int1 jackson - mind method and conditionals

file:///J|/1MyPhilEbooks/14Philosophy%20of%20Mind/Mind%20Method/htm.htm

MIND, METHOD AND CONDITIONALS Selected essays

Frank Jackson

London and New York

-iii-

First published 1998 by Routledge 11 New Fetter Lane, London EC4P 4EE

Simultaneously published in the USA and Canada by Routledge 29 West 35th Street, New York, NY 10001

© 1998 Frank Jackson

Typeset in Times by J&L Composition Ltd, Filey, North Yorkshire

Printed and bound in Great Britain by MPG Books Ltd., Bodmin, Cornwall

All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers.

British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library

Library of Congress Cataloguing in Publication Data

file:///J|/1MyPhilEbooks/14Philosophy%20of%20Mind/Mind%20Method/htm.htm (1 of 283)8/10/2006 10:44:49 μμ


Jackson, Frank, 1943- Mind, method, and conditionals: selected essays/Frank Jackson. p. cm. Includes bibliographical references and index. ISBN 0-415-16574-1 (hardcover) 1. Conditionals (Logic) 2. Philosophy of mind. 3. Methodology. I. Title. BC199.C56J34 1998 199 .94-dc21 97-50010 CIP

ISBN 0-415-16574-1

-iv-

CONTENTS Preface vii PART I Conditionals 1 1 On assertion and indicative conditionals 3 2 Classifying conditionals I 27 3 Classifying conditionals II 43 4 Postscript on truth conditions and assertability 51 PART II Mind 55 5 Epiphenomenal qualia 57 6 What Mary didn’t know 70 7 Postscript on qualia 76 8 Mental causation I 80 9 Mental causation without the language of thought 113

-v-



PART III Method in metaphysics 131 10 Metaphysics by possible cases 133 11 Armchair metaphysics 154 PART IV Ethics and action theory 177 12 Weakness of will 179 13 On the semantics and logic of obligation 197 14 Decision-theoretic consequentialism and the nearest and dearest objection 220 PART V Induction 247 15 Grue 249 Bibliography 270 Index 281

-vi-

PREFACE Bringing a number of one’s papers together is an interesting experience. Bad ways of saying things and obvious omissions leap to the eye. The papers that follow are in large part as originally published, but I have made a good number of small changes in phrasing and punctuation, and taken the opportunity to add a remark or two. The additions are marked ‘added note’ and are placed in the endnotes or, occasionally, when they seemed to warrant a more prominent position, in the text at the end of the relevant chapter. I have made the odd cut to avoid excessive duplication; the places where these cuts have been made are indicated in the relevant chapters. Also, I have reformatted the papers in the interests of uniformity.

Having noted two places where it is especially obvious that more than a ‘remark or two’ is called for, I included a postscript on truth conditions and on assertability as Chapter 4, and a postscript on qualia as Chapter 7.

The two papers expounding and defending the knowledge argument now seem to me to be mistaken. I have (reluctantly) come to the conclusion that there is a good reply to the knowledge argument, and give this reply in the postscript on qualia. I still think that the knowledge argument is the best argument for the view that physicalism leaves out the sensory side of psychology—it is the argument that physicalists have to meet. I also think that the mistake it makes is not some more or less elementary blunder, as has been suggested by some critics. The mistake is a subtle one that can



only be appreciated after the strength of the case for physicalism has been properly absorbed.

The details of first publication are, in chronological order: ‘Grue’, Journal of Philosophy, 71(1975):113-31; ‘On assertion and

-vii-

indicative conditionals’, Philosophical Review, 88 (1979):565-89; ‘Epiphenomenal qualia’, Philosophical Quarterly, 32 (1982): 127-36; ‘Weakness of will’, Mind, 93 (1984):1-18; ‘On the semantics and logic of obligation’, Mind, 94 (1985):177-96; ‘What Mary didn’t know’, Journal of Philosophy, 83 (1986): 291-5; ‘Classifying conditionals I’, Analysis, 50 (1990):134-47; ‘Classifying conditionals II’, Analysis, 51 (1991):137-43; ‘Decision-theoretic consequentialism and the nearest and dearest objection’, Ethics, 101 (1991):461-82; ‘Metaphysics by possible cases’, Monist, 77 (1994):93-110; ‘Armchair metaphysics’ in John O’Leary Hawthorne and Michaelis Michael, eds, Philosophy in Mind, Dordrecht, Kluwer, 1994, pp. 23-42; ‘Mental causation: The state of the art’, Mind, 105 (1996):377-13; ‘Mental causation without the language of thought’ in Maria Luisa Dalla Chiara, et al., eds, Structures and Norms in Science, vol. 2 of the 10th International Congress of Logic, Methodology and Philosophy of Science, Dordrecht, Kluwer, 1997, pp. 303-18. I am grateful to the publishers and editors for kind permission to reprint.

Frank Jackson

Canberra

September 1997

-viii-



Part I CONDITIONALS

-1-

[This page intentionally left blank.]

-2-

1 ON ASSERTION AND INDICATIVE CONDITIONALS The circumstances in which it is natural to assert the ordinary indicative conditional ‘If P then Q’ are those in which it is natural to assert ‘Either not P, or P and Q’, and conversely. For instance, the circumstances in which it is natural to assert ‘If it rains, the match will be cancelled’ are precisely those in which it is natural to assert ‘Either it won’t rain, or it will and the match will be cancelled’. Similarly, the circumstances in which it is natural to assert ‘Not both P and Q’ are precisely those in which it is natural to assert ‘Either not P or not Q’. We explain the latter coincidence of assertion conditions by a coincidence of truth conditions. Why not do the same in the case of the conditional? Why not, that is, hold that ‘If P then Q’ has the same truth conditions as ‘Either not P, or P and Q’?

This hypothesis—given the standard and widely accepted truth functional treatments of ‘not’, ‘or’, and ‘and’—amounts to the Equivalence thesis: the thesis that (P→Q) is equivalent to (P⊃Q). (I will use ‘→’ for the indicative conditional, reserving

for the subjunctive or counterfactual conditional.) In this chapter I defend a version of the Equivalence thesis.

As a rule, our intuitive judgements of assertability match up with our intuitive



judgements of probability, that is, S is assertable to the extent that it has high subjective probability for its assertor. Now it has been widely noted that when (P⊃Q) is highly probable but both ~P and Q are not highly probable, it is proper to assert (P→Q). 1 The problem for the Equivalence thesis is to explain away the putative

counter-examples to and , the only too familiar cases where despite the high probability of ~P or of Q, and so of (P⊃Q), (P→Q) is not highly assertable.

-3-

I will start in §1 by considering the usual way of trying to explain away these counter-examples and argue that it fails. An obvious reaction to this failure would be (is) to abandon the Equivalence thesis, but I argue in §2 that another is possible, namely, that the general thought behind the usual way of explaining away the paradoxes of material implication is mistaken. This leads, in §3, to the version of the Equivalence thesis I wish to defend. In §4 I point out some of the advantages of this account of indicative conditionals, and in §5 I reply to possible objections.

1.1 The usual way of explaining away the counter-examples

Suppose S1 is logically stronger than S2: S1 entails S2, but not conversely. And suppose S1 is nearly as highly probable as S2. (It cannot, of course, be quite as probable, except in very special cases.) Why then assert S2 instead of S1? There are many possible reasons: S2 might read or sound better, S1 might be unduly blunt or obscene, and so on. But if we concentrate on epistemic and semantic considerations widely construed, and put aside more particular, highly contextual ones like those just mentioned, it seems that there would be no reason to assert S2 instead of S1. There is no significant loss of probability in asserting S1 and, by the transitivity of entailment, S1 must yield everything and more that S2 does. Therefore, S1 is to be asserted rather than S2, ceteris paribus.

This line of thought, which I will tag ‘Assert the stronger instead of the weaker (when probabilities are close)’, has been prominent in defences of the Equivalence thesis that the ordinary indicative conditional (P→Q) is equivalent to the material conditional, (P⊃Q). 2 The Equivalence theorist explains away the impropriety of asserting (P⊃Q) when one of ~P or Q is highly probable, by saying that, in such a case, you should



come right out and assert the logically stronger statement, namely, either ~P, or Q, as the case may be.

The same idea can be put in terms of evidence instead of probability. 3 If your evidence favours (P⊃Q) by favouring one of ~P or Q, you should simply assert ~P, or assert Q, whichever it is, and not the needlessly weak conditional. 4 But I will concentrate in the main on the probabilistic formulation when presenting my objections.

My first objection is that a conditional like ‘If the sun goes out

-4-

of existence in ten minutes’ time, the Earth will be plunged into darkness in about eighteen minutes’ time’ is highly assertable. However the probability of the material conditional and the probability of the negation of its antecedent are both very close to one; and so at most the probability of the conditional is only marginally the greater. Hence this is a case where the weaker is assertable despite the absence of any appreciable gain in probability, contrary to the maxim ‘Assert the stronger instead of the weaker’.

The second objection is that conditionals whose high probability is almost entirely due to that of their consequents may be highly assertable. Suppose we are convinced that Carter will be re-elected whether or not Reagan runs. We say both ‘If Reagan runs, Carter will be re-elected’ and ‘If Reagan does not run, Carter will be re-elected’. The high subjective probability can only be due to that of the common consequent, yet the consequent is allegedly logically stronger and so, by the maxim, the conditionals ought not to be assertable.

Moreover, such cases cannot be handled by a conventional exemption from the maxim in the case of conditionals with very probable consequents. 5 Both the following conditionals are highly unassertable, but have very probable consequents: ‘If the history books are wrong, Caesar defeated Pompey in 48 BC’; ‘If the sun goes out of existence in ten minutes’ time, the Earth will not be plunged into darkness in eighteen minutes’ time’.

The third objection is that there is a third paradox of material implication. If the Equivalence thesis is true, then ((P→Q) v (Q →R)) is a logical truth. But evidently it is not in general highly assertable. Of course logical truths are as logically weak as you



can get, nevertheless ‘Assert the stronger instead of the weaker’ is of no assistance in explaining away the third paradox. Whatever you think about this maxim in general, it does not apply universally to logical truths. ‘If that’s the way it is, then that’s the way it is’; ‘George must either be here or not here’; ‘The part is not greater than the whole’ and so on, are all highly assertable.

The fourth objection is that ‘Assert the stronger instead of the weaker’ is, of necessity, silent about divergences in assertability among logical equivalents, simply because logical equivalents do not differ in strength. But equivalence theorists must acknowledge some marked divergences among equivalents. According to them, ((P & (~P→R)) and (P & (~P→S)) are logically equivalent, both being equivalent to P. But their assertability

-5-

can differ sharply. ‘The sun will come up tomorrow but if it doesn’t, it won’t matter’ is highly unassertable, while ‘The sun will come up tomorrow but if it doesn’t, that will be the end of the world’ is highly assertable.

My final objection is that if the standard way of trying to explain away the paradoxes is right, ‘or’ and ‘→’ are on a par. It would, for instance, be just as wrong, and just as right, to assert ‘P or Q’ merely on the basis of knowing P as to assert (P⊃Q) merely

on the basis of knowing not P. And, more generally, and

should strike us as just as much a problem for the thesis that ‘P or Q’ is equivalent to (P v Q) as do the paradoxes of material implication for the Equivalence thesis. It is a plain fact that they do not. The thesis that ‘P or Q’ is equivalent to (P v Q) is relatively noncontroversial; the thesis that (P→Q) is equivalent to (P⊃Q) is highly controversial.

This objection, of course, applies not just to attempts to explain away the paradoxes in terms of ‘Assert the stronger’, but to any attempt which appeals simply to considerations of conversational propriety. It leaves it a mystery why we—who are, after all, reasonably normal language users—find it so easy to swallow one thesis and so hard to swallow the other.

Should we respond to these objections by abandoning the Equivalence thesis, or by looking for a different way of explaining away the paradoxes? An argument for the



latter is that the thought behind ‘Assert the stronger rather than the weaker’ contains a serious lacuna, as I now argue.

1.2 A reason for sometimes asserting the weaker

Suppose, as before, that S1 is logically stronger than S2, and that S1’s probability is only marginally lower than S2’s. Consistent with this, it may be that the impact of new information, I, on S1 is very different from the impact of I on S2; in particular, it may happen that I reduces the probability of S1 substantially without reducing S2 ’s to any significant extent (indeed S2’s may rise). I will describe such a situation as one where S2 but not S1, is robust with respect to /. If we accept Conditionalisation, the plausible thesis that the impact of new information is given by the relevant conditional probability, then ‘P is robust with respect to /’ will be true just when both Pr(P) and Pr(P/I) are close and high. 6 (Obviously, a more general account would simply require that

-6-

can differ sharply. ‘The sun will come up tomorrow but if it doesn’t, it won’t matter’ is highly unassertable, while ‘The sun will come up tomorrow but if it doesn’t, that will be the end of the world’ is highly assertable.

My final objection is that if the standard way of trying to explain away the paradoxes is right, ‘or’ and ‘→’ are on a par. It would, for instance, be just as wrong, and just as right, to assert ‘P or Q’ merely on the basis of knowing P as to assert (P⊃Q) merely on the basis of knowing not P. And, more generally, and should strike us as just as much a problem for the thesis that ‘P or Q’ is equivalent to (P v Q) as do the paradoxes of material implication for the Equivalence thesis. It is a plain fact that they do not. The thesis that ‘P or Q’ is equivalent to (P v Q) is relatively noncontroversial; the thesis that (P→Q) is equivalent to (P⊃Q) is highly controversial.

This objection, of course, applies not just to attempts to explain away the paradoxes in terms of ‘Assert the stronger’, but to any attempt which appeals simply to considerations of conversational propriety. It leaves it a mystery why we—who are, after all, reasonably normal language users—find it so easy to swallow one thesis and so hard to swallow the other.

Should we respond to these objections by abandoning the Equivalence thesis, or by



looking for a different way of explaining away the paradoxes? An argument for the latter is that the thought behind ‘Assert the stronger rather than the weaker’ contains a serious lacuna, as I now argue.

1.2 A reason for sometimes asserting the weaker

Suppose, as before, that S1 is logically stronger than S2, and that S1’s probability is only marginally lower than S2’s. Consistent with this, it may be that the impact of new information, I, on S1 is very different from the impact of I on S2; in particular, it may happen that I reduces the probability of S1 substantially without reducing S2 ’s to any significant extent (indeed S2’s may rise). I will describe such a situation as one where S2 but not S1, is robust with respect to /. If we accept Conditionalisation, the plausible thesis that the impact of new information is given by the relevant conditional probability, then ‘P is robust with respect to /’ will be true just when both Pr(P) and Pr(P/I) are close and high. 6 (Obviously, a more general account would simply require that

-6-

Pr(P) and Pr(P/I) be close, but throughout we will be concerned only with cases where the probabilities are high enough to warrant assertion, other things being equal.)

We can now see the lacuna in the line of thought lying behind ‘Assert the stronger instead of the weaker’. Despite S1 and S2 both being highly probable, and S1 entailing everything S2 does, there may be a good reason for asserting S2 either instead of or as well as S1. It may be desirable that what you say should remain highly probable should I turn out to be the case, and further it may be that Pr(S2 /I) is high while Pr(SI /I) is low. In short, robustness with respect to / may be desirable, and (consistent with S1 entailing S2) S2 may have it while S1 lacks it.

Examples bear this out. Robustness is an important ingredient in assertability. Here are two examples taken from those which might be (are) thought to be nothing more than illustrations of ‘Assert the stronger instead of the weaker’.

Suppose I read in the paper that Hyperion won the 4.15. George asks me who won the 4.15. I say, ‘Either Hyperion or Hydrogen won’. Everyone agrees that I have done the wrong thing. Although the disjunction is highly probable, it is not highly assertable. Why? The standard explanation is in terms of ‘Assert the stronger instead of the



weaker’. 7 But is this the whole story? Consider the following modification to our case. What I read is that H…won. The name is too blurred for me to do more than pick out the initial letter. However, I happen to know that Hyperion and Hydrogen are the only two horses in the 4.15 whose names begin with ‘H’, and in addition I know that Hydrogen is a no-hoper from the bush. Clearly, it is still the case that ‘Hyperion won’ is highly probable and it would be quite proper for me to say that Hyperion won. But it would also be quite proper for me to say ‘Hyperion or Hydrogen won’, despite its being weaker and only marginally more probable. Indeed, the natural thing to do would be to say something like: ‘Either Hyperion or Hydrogen won. It can’t have been Hydrogen—he’s a no-hoper. So it must have been Hyperion’.

The obvious explanation for the marked change in the assertability of the disjunction is that in the original case it was not robust with respect to the negation of both its disjuncts taken separately, while in the modified case it is. In the original case, were I to learn that Hyperion was not the winner I would have to abandon the disjunction. In the modified case I would not, though I would have to abandon my low opinion of Hydrogen. Therefore,

-7-

in the modified case, there is point to asserting that Hyperion or Hydrogen won instead of simply that Hyperion won, even if the probabilities are very close. This disjunction possesses a relevant robustness that its left disjunct lacks.

Indeed, surely there are many cases where disjunctions are highly assertable even though they have probabilities for their assertors only marginally greater than that of one of their disjuncts. Consider ‘Either Oswald killed Kennedy or the Warren Commission was incompetent’. This is highly assertable even for someone convinced that the Warren Commission was not incompetent. Yet they are in a position to assert the stronger ‘Oswald killed Kennedy’. The disjunction is nevertheless highly assertable for them, because it would still be probable were information to come to hand that refuted one or the other disjunct. The disjunction is robust with respect to the negation of either of its disjuncts taken separately—and just this may give point to asserting it, because it makes it acceptable to a possible hearer who denies one or other of the disjuncts.

Moreover, we can have highly probable disjunctions which are, unlike the two just considered, significantly more probable than either of their disjuncts which are, nevertheless, not highly assertable. Suppose I propose to toss a fair coin five times in such a way that the tosses are probabilistically independent. Then ‘At least one of the



five tosses will be a head’ is probable enough (~97 per cent) to warrant assertion; consequently, so is the equivalent disjunction ‘Either at least one of the first three tosses or at least one of the last two tosses will be a head’, and, moreover, each disjunct is significantly less probable than the disjunction. But it would be highly misleading to assert the disjunction in preference to the equivalent sentence for it would create in hearers the mistaken expectation that should the first three tosses fail to yield a head, at least one of the last two will yield one.

The second example is one from David Lewis.

We are gathering mushrooms; I say to you ‘You won’t eat that one and live.’ A dirty trick: I thought that one was safe and especially delicious, I wanted it myself, so I hoped to dissuade you from taking it without actually lying. I thought it highly probable that my trick would work, that you would not eat the mushroom, and therefore that I would turn out to have told the truth. But though what I said had a high subjective probability of

-8-

truth, it had a low assertability and it was a misdeed to assert it. Its assertability goes not just by probability but by the resultant of that and a correction term to take account of the pointlessness and misleadingness of denying a conjunction when one believes it false predominantly because of disbelieving one conjunct. 8

But this explanation faces two difficulties. First, suppose I am not that confident that my trick will work. I am pretty sure but not certain enough to warrant outright assertion. Further suppose that I am also pretty certain that you will die for reasons unconnected with mushrooms. The two factors combined bring the probability of ‘You won’t eat that one and live’ up to a level sufficient to warrant assertion. In this case, the probability of falsity of each conjunct contributes significantly to the probability that the negated conjunction is true, but nevertheless it would still be a misdeed to assert it. Second, suppose the mushroom really is dangerous and I say, ‘You won’t eat that one and live’ while crushing it under my foot for safety’s sake. The difference in probability between the negated conjunction and ‘You won’t eat that one’ will then be minuscule. But the negated conjunction is nevertheless highly assertable in this case.

It seems to me, therefore, that a better explanation is one in terms of robustness. You take me to be providing information relevant to mushroom-eating pleasures and so construct for yourself the following piece of practical reasoning. I won’t eat that one



and live. (Premise supplied by me.) I eat that one. (Premise you can make true.) Therefore, I won’t live. The conclusion is undesirable, hence you are led to refrain from making the second premise true.

Why were you tricked? The argument is valid, the premise I supplied does have a high probability, and you are able to give the second premise a high probability. But in order to infer the conclusion of a valid argument, all premises need to be highly probable together, and if you were to make the second premise highly probable, the first premise (supplied by me) would no longer be highly probable. In the circumstances, you were entitled to take it that not only was ‘You won’t eat that one and live’ highly probable, it was also robust with respect to ‘You eat that one’. My misdeed lay in asserting something lacking appropriate robustness.

Therefore when considering propriety of assertion, we should

-9-

take account of robustness as well as high probability, relevance, informativeness, and so on.

1.3 The application of robustness to conditionals

Robustness is a relative affair. A highly probable sentence may be very robust relative to one possible piece of information and the opposite relative to another. 9 Often the possible information relative to which robustness is desirable is given by the context. In the mushroom-gathering story, it was obvious that the hearer expected sentences that were robust relative to his eating the mushroom. That is how he was tricked. But context will not always be enough. It makes sense that we should have syntactical constructions which signal the possible information relative to which we take what we are saying to be robust.

Their role would be akin to that of ‘but’ in signalling or indicating a contrast without the obtaining of this contrast being a necessary condition for speaking truly. 10 The truth-conditions for ‘P but Q’ are the same as those for ‘P & Q’ In familiar jargons, their literal content is the same, but the use of the first carries a conventional (not conversational) implicature that the second does not; or they differ in tone but are alike in sense. 11 I will, however, talk mainly of signalling and indicating, rather than implicature or tone. What follows does not depend crucially on the precise way such a distinction should be drawn.



It is, of course, vital that we allow the possibility of distinguishing signalling or indicating an attitude towards a sentence from making that attitude part of the truth conditions, sense or literal content of what we say. There is a great difference between producing a sentence S as something accepted, and thereby asserted, and producing it as an example or as something granted for the sake of argument. It is thus important that we can signal this—perhaps by using Frege’s assertion sign—and such a signal cannot be taken simply as part of the content of what is said. Because ‘S and I accept (or assert) S’ may as easily as S itself be produced as an example or granted for the sake of argument rather than being asserted. 12

I am suggesting, then, that when we assert a sentence it makes sense that we should have ways of indicating that, as well as obeying the base rule that requires that S be highly probable, we also take it that, for some /, S is robust with respect to /.

One way of doing this is to put your sentence in disjunctive

-10-

form when it would be shorter and simpler not to. Suppose I am asked what colour Harry’s car is. It is perfectly acceptable for me to reply simply that it is blue, even if my ground for being confident that it is blue is that it is light blue. Unless there is reason to think that the precise shade matters, near enough is good enough here. Suppose, however, that I replied that Harry’s car is either light blue or dark blue. This reply is not acceptable in the circumstances even if the precise shade does not matter, despite the fact that (ignoring borderline cases) it is equivalent to the acceptable one (and so, incidentally, the difference in assertability cannot be explained by reference to ‘Assert the stronger instead of the weaker’). The reason that the second reply is not acceptable is that in putting it in explicitly disjunctive form, you signal robustness with respect to the negation of each disjunct taken separately. The reply would be proper only if both the (subjective) probability of its being light blue or dark blue given it is not light blue, and the probability of its being light blue or dark blue given it is not dark blue were high. And in our case the former is low.

In general, we are happiest asserting disjunctions which are two-sidedly robust. We most happily assert ‘P or Q’ when Pr(P v Q), Pr(P v Q/~P), and Pr(P v Q/~Q) are all high. (Thus, the oft noted ‘exclusive feel’ about the inclusive ‘or’.) Accordingly, when we are not in a position to so assert, we should expect to have a way of signalling merely one-sided robustness in order to avoid misleading our hearers into assuming two-sided robustness. And it seems that we do.



Consider the following, common enough kind of case. You are pretty sure that George lives in Boston but not quite sure enough to warrant outright assertion. You are, though, sure enough that he lives somewhere in New England. You say, ‘He lives in Boston or anyway somewhere in New England’. Likewise, we say things like ‘He is a fascist (communist) or anyhow on the far right (left)’ and ‘Caesar defeated Pompey in 48 BC, or at least that’s what George told me’. We use the ‘P or anyway Q’ construction to indicate that ‘P v Q’ is robust with respect to ~P, but not with respect to ~Q. Should you learn against your expectation that George does not live in Boston, the disjunction will still be highly probable for you due to its right disjunct ‘George lives in New England’ still being so; but obviously you will have to abandon the disjunction altogether should you learn that George does not live in New England at all.

-11-

A consequence of this asymmetry is that commutation can give strange sounding results. ‘He lives in Boston or anyway somewhere in New England’ is a happy saying, whereas ‘He lives somewhere in New England or anyway in Boston’ is not. Nevertheless, commutation is valid; for the truth conditions of ‘P or anyway Q’ are just those of ‘P v Q’. ‘George lives in Boston or anyway somewhere in New England’ is true if and only if either ‘George lives in Boston’ is true or ‘George lives somewhere in New England’ is true. ‘Caesar defeated Pompey in 48 BC, or at least that is what George told me’ is true if either disjunct is true, and false if neither is, and so on and so forth. Signalling robustness does not invade truth conditions.

Before I apply these ideas to indicative conditionals, let me review the course of the argument. High probability is an important ingredient in assertability. Everyone accepts that. But so is robustness. Commonly, cases cited to illustrate ‘Assert the stronger instead of the weaker’ really illustrate the importance of robustness. The relevant robustness, however, is relative to statements other than the one being asserted. (Every highly probable statement is trivially robust with respect to itself.) Thus we need devices and conventions to signal relative to which statements our assertions are robust. We have just been looking at some of these devices and have noted that their presence does not alter truth conditions. Accordingly, I suggest that the indicative conditional construction is such a device. It signals robustness with respect to its antecedent: it is proper to assert (P→Q) when (P⊃Q) is highly probable and robust with respect to P, that is, when Pr(P⊃ Q/P) is also high. But, by analogy with explicit disjunction and ‘…or anyway…’, the truth conditions of (P→Q) are those of (P⊃Q). It is like ‘Nevertheless P’ in this regard. The use of ‘nevertheless’ signals



the robustness of P with respect to what has gone before, but the whole sentence is true if and only if P is.

At first glance it may appear that this version of the Equivalence thesis is totally opposed to those theories which assign conditionals assertion and acceptance but not truth conditions. 13 But in fact it is a half-way house. Consider again ‘Nevertheless P’. Although the whole is true if and only if P is true, a part—‘nevertheless’—contributes to assertion conditions without affecting truth conditions. We can give the conditions under which it is proper to use ‘nevertheless’ but not those under which using it is saying something true. The same may be said of the signalling role of the indicative conditional construction. Our

-12-

theory is thus a supplemented Equivalence theory. In the widest sense of ‘meaning’, (P→Q) and (P⊃Q) do not mean the same. But their truth conditions are the same—they agree in sense or literal content. The extra element is that in using (P→Q), you explicitly signal the robustness of (P⊃Q) with respect to P, and this element affects assertion conditions without affecting truth conditions.

We could have gone further and made the robustness of (P⊃Q) with respect to P a necessary condition for the truth of (P→Q). But this seems, as a simple fact of linguistic usage, too strong. First, we allow that a person may speak truly in the conditional mode without deserving to do so. Suppose it is highly probable that it will rain tomorrow and, in consequence, that the match will be cancelled. But, with the intention of misleading Fred, I say that if it rains, the match will go ahead. In this case ‘It rains ⊃ the match will go ahead’ is neither probable nor robust with respect to ‘It rains’. Further, suppose that it does indeed rain but, against the odds, the match goes ahead. We allow that I have spoken truly without of course deserving to do so. Second, we allow that one member of the set of conditionals of the form ‘If I write down the number…, I will write down the number of molecules in this room’ is true. Yet none is robust with respect to its antecedent.

What is the point of signalling the robustness of (P⊃Q) with respect to P? The answer lies in the importance of being able to use modus ponens. Although ‘(P⊃Q), P, ∴ Q’ is certainly valid, there is a difficulty about using it in practice. Suppose my evidence makes (P⊃Q) highly probable, but that I have no evidence concerning P. Q is of interest to me, so I set about finding evidence for P if I can. The difficulty is that finding evidence that makes P highly probable is not enough in itself for me to



conclude Q by modus ponens. For the evidence that makes P probable may make (P⊃Q) improbable. Indeed, except in special cases of extreme probability, Pr(P⊃Q/P) < Pr (P⊃Q); that is, normally on learning P, I must lower the probability I give (P ⊃Q), so endangering the inference to Q. It is thus of particular interest whether or not (P⊃Q)’s high probability would be unduly diminished by learning P; that is, it is important whether or not (P⊃Q) is robust with respect to P. In sum, we must distinguish the validity of modus ponens from its utility in a situation where I know (P⊃Q) but do not know P. 14 The robustness of (P⊃Q) relative to P is what is needed to ensure the utility of modus ponens in such situations.

-13-

It does not, though, ensure the utility of modus tollens. Pr(P⊃ QIP) can be high when Pr(P⊃Q/~Q) is low. And this is how things should be. You may properly assert (P→Q) when you would not infer ~P on learning ~Q. Suppose you say ‘If he doesn’t live in Boston, then he lives somewhere in New England’; or ‘If he works, he will still fail’. You will—despite the validity of modus tollens—neither infer that he lives in Boston on learning (to your surprise) that he doesn’t live in New England, nor infer that he didn’t work on learning (to your surprise) that he passed. Rather, on learning either you would abandon the original conditional as mistaken. 15

It is not only the robustness of material conditionals with respect to their antecedents that is important. Accordingly, if our approach is along the right lines, we should expect a linguistic device to signal the robustness of Q with respect to P, not merely of (P⊃Q) with respect to P. But if the Supplemented Equivalence thesis is right, the latter is sufficient for the former. Consider (Q & (P→Q)). According to the Equivalence thesis, it is equivalent to Q, and according to our supplementation, the right-hand conjunct signals that Pr(P⊃Q/P) is high. But Pr(P ⊃QIP) simplifies to Pr(Q/P). Hence, asserting (Q & (P→Q)) is equivalent to asserting Q and also signals the robustness of Q with respect to P—just what we are looking for. When we assert both Q and (P→Q), we commonly use a ‘still’ construction: ‘The match will be played, and it will still be played if it rains’; ‘Carter will be re-elected, and if the Camp David talks fail, he will still be re-elected’. Often we don’t bother to repeat the common element, Q. Context makes it clear that we think that the match will be played or that Carter will be re-elected, and we simply say ‘The match will still be played if it rains’ or ‘(Even) if the Camp David talks fail, Carter will still be re-elected’. A stronger position is that ‘If P, then still Q’ entails Q. 16 But consider one who makes, all in one breath, the following perfectly acceptable remark: ‘If it rains lightly, the match will still be played. But if it rains heavily, as it well may, the match will be cancelled’. Surely he or she is



not asserting inter alia that the match will be played.

1.4 Defence

I take for granted one negative argument for our Supplemented Equivalence thesis, namely that all its competitors face well-

-14-

known objections. One obvious positive argument for it would consist in assembling a large number of examples of indicative conditionals and testing our intuitions concerning assertion against the results our theory predicts. Fortunately this is not necessary. Ernest Adams has provided a simple formula governing our intuitions, and the Supplemented Equivalence theory explains this formula.

Adams has shown that the (intuitively justified) assertability of (P→Q) is given by Pr(Q/P)=df Pr(PQ)/Pr(P). 17 Thus I assent to ‘If it rains, the match will be cancelled’ to the extent that my subjective probability of the match being cancelled, given it rains, is high.

We explain Adams’s thesis as follows. On our theory, the assertability of (P→Q) will be the product of two factors: the extent to which Pr(P⊃Q) is high, and the extent to which (P⊃Q) is robust with respect to P. But we have from the probability calculus that Pr(P⊃Q/P)=Pr(Q/P) and that Pr(P⊃Q) Pr(Q/P). Consequently, both conditions are satisfied to the extent that Pr(Q/P) is high.

(An important recent result of Lewis’s highlights the significance of this derivation. He proves that the obvious alternative explanation of Adams’s thesis fails. He proves (by a reductio argument) that (P→Q) does not differ in truth conditions from (P⊃Q) in such a way as to make Pr(P→Q)=Pr(Q/P)). 18

Consequently, we can explain why (P→Q) and (P→~Q) are not assertable together when P is consistent. Pr(Q/P) and Pr(~Q/P) cannot (from the calculus) both be high. 19 Or, more precisely, they cannot both be high relative to the same body of evidence. Robustness, like probability in general, is relative to evidence and, of course, Pr(Q/P&R) and Pr(~Q/P&S) can both be high. Accordingly, our theory predicts that we should be happy to assert both (P→Q) and (P→~Q) when it is explicit that the relevant bodies of evidence are appropriately different. Exactly this happens.



Harry and George are discussing whether Fred went to the rock concert. Harry says, ‘If Fred went, he must have gone by car, because there was a transport strike at the time’. George says, ‘But Fred regards the private car as exploitative and never goes anywhere by car on principle; so if he went, it cannot have been by car’. They conclude, ‘In that case, obviously Fred did not go to the rock concert’. Instead of regarding their statements as mutually inconsistent, Harry and George draw from them the conclusion that Fred did not go. 20

-15-

Our theory, then, makes highly assertable just those conditionals intuition judges to be highly assertable. But what of our intuitive judgements of validity? I am committed to taking ‘~P, ∴ P→Q’ and ‘Q, ∴ P→Q’ to be valid, and they notoriously lack intuitive appeal. But this lack of appeal seems to derive from our reluctance to assert (P→Q) merely because we are confident that ~P, and our (less marked) reluctance to assert (P→Q) merely because we are confident that Q, and our theory can explain these easily enough. Neither the fact that Pr(~P) is high, nor the fact that Pr(Q) is high, is sufficient for Pr(P⊃Q/P) to be high. The reason our reluctance is less marked in the case of asserting (P→ Q) on the basis of our certainty that Q, is that Pr(Q) being high, together with P and Q being probabilistically independent, is sufficient for Pr(P⊃Q/P) to be high.

Similarly, what I referred to earlier as the third paradox—that ((P→Q) v (Q→R)) is a logical truth and yet is far from invariably highly assertable—is not a decisive objection to our supplemented version of the Equivalence thesis, because the presence of signals can make logical truths unassertable. We have already noted the plausibility of giving ‘Nevertheless P’ the same truth conditions as P. Consequently, ‘Nevertheless P or nevertheless not P’ is a logical truth, but it is not highly assertable.

What of strengthening the antecedent, hypothetical syllogism and contraposition, all of which are of course valid on our theory? Take contraposition (similar points apply to all three). The problem is not that it seems invalid stated in symbols; exactly the reverse is the case, as is evinced by its appearance in natural deduction systems. The problem is rather a certain class of apparent counter-examples like: ‘If George works hard, he will (still) fail; therefore, if he passes, he won’t have worked hard’, and ‘If Carter is re-elected, it won’t be by a large margin; therefore, if Carter is re-elected by a large margin, he won’t be re-elected’.

But these apparent counter-examples are paralleled by ones against the



commutativity of ‘…or anyway…’: for instance, ‘It won’t rain or anyway not heavily; therefore, it won’t rain heavily or anyway it won’t rain’. Despite this, we noted that it seemed clearly right to give the same truth conditions to ‘P or anyway Q’ as are standardly given to ‘P or Q’. The explanation for the counter-intuitive feel must, therefore, lie not in the failure of commutation, but in the failure of what is signalled by ‘anyway’ to ‘commute’. Similarly, addition is hardly appealing when applied to ‘…or at

-16-

least…’. Consider, ‘Harry said that Caesar defeated Pompey in 48 BC; therefore, Harry said that Caesar defeated Pompey in 48 BC, or at least Caesar defeated Pompey in 48 BC’.

It seems, therefore, not unreasonable to attribute the counter-intuitive feel of certain instances of contraposition to the failure of what is signalled by the indicative conditional construction to ‘contrapose’ (and likewise for hypothetical syllogism and so on). And it may be confirmed by inspection that the putative counter-examples to contraposition are all ones where Pr(P⊃Q/P)= Pr(Q/P) is high, and Pr(~Q⊃~P/~Q)=Pr(~P/~Q) is low. For example, the probability of Carter not being re-elected by a large margin given he is re-elected may be high, when the probability of Carter not being re-elected given he is re-elected by a large margin is zero. Accordingly, we can explain our reluctance to assert ‘If Carter is re-elected by a large margin, then Carter will not be re-elected’ even when we are happy to assert ‘If Carter is re-elected, then it will not be by a large margin’ in terms not of the first being false and the second true, but in terms of what is signalled by saying the first being false, and what is signalled by saying the second being true. 21

1.5 On three objections

(i) It may be objected that the account offered above is circular. The Equivalence thesis itself is not circular, obviously, but the supplemented thesis involves a story about the role of the indicative conditional construction as signalling robustness, and it might be objected that robustness can only be elucidated via a conditional construction. P is robust for person S relative to I just if P’s high probability for S would not be substantially reduced if S were to acquire the information that /. One reply would be to urge that we simply define robustness in terms of conditional probability: Pr(P/I)=df Pr(PI)/Pr(I). No conditionals there.

This reply is open to challenge. For the defence of so defining robustness must



involve a defence of Conditionalisation, the thesis that the impact of new information is given by the relevant conditional probability, and it might be urged that talk of the impact of new information can best be understood as talk of what one’s probabilities would or should be if one were to acquire the new information. But another reply is possible. The conditionals involved here are essentially subjunctive and counterfactual in character, and as such are importantly distinct from indicative

-17-

conditionals. It is not, therefore, uselessly circular to appeal to the former in one’s story about the latter.

I take the case for separating out the problem of indicative conditionals from the problem of subjunctive conditionals to be familiar. 22 It derives from pairs like ‘If Carter is bald, no one knows it’ and ‘If Carter were bald, no one would know it’, and ‘If Oswald did not shoot Kennedy, someone else did’ and ‘If Oswald had not shot Kennedy, someone else would have’. For each pair, we assent to the first and dissent from the second, and our dissent from the second member of each pair is accompanied by assent to, respectively, ‘If Carter were bald, everyone would know it’ and ‘If Oswald had not shot Kennedy, no one would have’.

It sometimes seems to be thought that the contrast between indicative and subjunctive or counterfactual conditionals only shows up when (i) the consequents are known one way or the other, and (ii) they are not about the future. 23 But here is a pair involving a future, doubtful event. I have been told that Fred’s birthday and George’s birthday fall next week. But I cannot remember the exact days, only that Fred’s is the day before George’s. I say ‘If Fred’s is next Tuesday, George’s will be next Wednesday’, and I don’t say ‘If Fred’s were next Tuesday, George’s would be next Wednesday’ (unless of course I know that the hospital specially arranged George’s birth one day after Fred’s, or something of that kind). Likewise, suppose that station B is half-way between A and C. A person at B observing a train passing through on time may well affirm ‘If the train had been late leaving A, it would be late pulling into C’, while denying ‘If the train was late leaving A, it will be late pulling into C’.

It is, nevertheless, undeniable that the contrast between indicative and counterfactual conditionals is less marked in the case of conditionals pertaining to the future. For example, before the assassination of Kennedy, we would say both ‘If Oswald were not to shoot Kennedy, no one would’ and ‘If Oswald does not shoot Kennedy, no one will’. It is only now, after the event, that we say ‘If Oswald did not shoot Kennedy,



someone else did’. But this is a point in favour of our theory for it can explain why the contrast is less marked in the case of the future. It is a fact that we know more about the past than about the future. I know more about who won last year’s election than I do about who will win next year’s. In particular, our beliefs about the future, by and large, depend on relatively tenuous beliefs

-18-

about what present and past conditions will give rise to, while our beliefs about the past are frequently independent of our beliefs about how the past came about. My beliefs about next year’s election winner rest on my beliefs about present conditions and their effect on electoral popularity, while my beliefs about last year’s winner are by and large independent of my views as to what led to her success. Predicting election winners calls for a theory of what makes for electoral popularity, retrodicting them only calls for an ability to read the newspapers. Consequently, the probabilities we assign future events depend on our views about what would lead to what. But, by our theory, it is these very probabilities that settle the indicative conditionals we assert, and it is these very views about what would lead to what that are expressed in subjunctive and counterfactual conditionals. 24 Hence the general match between the two in the case of the future. On the other hand, the probabilities we assign past events may be largely independent of our views about what would lead to what. When Q is past, we may give Pr(Q/P) a high value independently of what we believe gave rise to it, and so may assert (P→Q) largely independently of our stance on

counterfactuals of the form , including, in particular,

It might also be thought that the contrast between indicative and subjunctive conditionals is simply due to the different role of what is being taken for granted, presupposed or regarded as common knowledge in the context. 25 When we consider indicative conditionals, we ‘hold onto’ common knowledge; when we consider subjunctive conditionals, we need not. 26 It is taken for granted that someone shot Kennedy; hence even under the indicatively expressed supposition that Oswald did not shoot Kennedy, we hold that someone (else) did. But under the subjunctively expressed supposition that Oswald had not, we may abandon this presupposition. Likewise with the other examples given. It is common knowledge that Carter is not bald and also (we were supposing) that Fred’s birthday is the day before George’s and that the train was on time at B; and these facts were what was being retained when the indicatives were in question and being abandoned when the subjunctives were in question. However, the reverse happens with other examples.



You and I have been taking the date of Caesar’s defeat of Pompey as common knowledge; and it is just this we hold onto in asserting the subjunctive ‘If the historians had reported the date of Caesar’s victory as 50 BC, they would have been wrong’ (not

-19-

even historians can change the past), and it is just this we abandon in asserting the indicative ‘If the historians do report the date as 50 BC, then I am wrong in giving it as 48 BC’. Perhaps it will be objected that what was taken as common knowledge was that the historians have the date right, rather than the date itself. But then my point can be made with a different pair. We assert ‘If Caesar’s victory was in 50 BC, the historians have the date wrong’, while denying ‘If Caesar’s victory had been in 50 BC, the historians would have got the date wrong’. Likewise, it is common ground that the declared winner of a presidential election is the person with the most votes. Yet it is just this we are prepared to abandon when we consider indicative conditionals starting ‘If Ford got more votes than Carter,…’, and just this we hold onto when considering subjunctive conditionals starting ‘If Ford had got more votes than Carter,…’.

(ii) Thus far, I have focused on explaining assent patterns to conditionals in terms of the Supplemented Equivalence theory. What of dissent? Standardly, you dissent from an assertion just when its subjective probability of falsity is high (neglecting, as before, highly contextual factors like obscenity). The probable falsity of what may be signalled by the assertion is by and large irrelevant. 27 You dissent from ‘He is poor but happy’ just when it is probable that he is either not poor or not happy, not when you dissent from the signalled contrast. Dissent is typically dissent from what is literally said. But it is clear and generally acknowledged that we dissent from conditionals in circumstances other than those where it is probable that the antecedent is true and the consequent false. Even anti-Warrenites dissent from ‘If Oswald killed Kennedy, then the Warren Commission got the killer’s identity wrong’; but they do not regard ‘Oswald killed Kennedy and the Warren Commission got the killer’s identity right’ as highly probable. They think, rather, that Oswald did not kill Kennedy and that the Warren Commission got the killer’s identity wrong.

There are, however, exceptions to the rule that we dissent just when it is probable that what is literally said is false. Suppose I say in a serious tone of voice ‘I believe that it will rain tomorrow’. There are two circumstances in which you naturally dissent. One is when you think I am lying, that is, when the probability of falsity of what is literally said is high (by your lights). The other is when you think it will not rain. In this



case, your dissent is not

-20-

from what I literally say but from what I signal by saying it in a serious tone of voice, namely, that its raining is highly probable.

Another example is when I say The winner of the election for club president will come from Tom, Dick and Harry.’ What I say counts as true if anyone of these three wins. But you won’t dissent only if you think this improbable. You may grant it probable because of the excellent chance Tom has of winning but nevertheless dissent because I left out George, and in your view George has the best chance after Tom. In other words, you will dissent when what you would assert is that the winner will come from Tom and George, and not only when you are prepared to say that none of Tom, Dick and Harry has a chance.

The explanation for these two cases being exceptions to the rule that dissent is prompted by low probability of literal truth appears to lie in the peculiarly intimate relationship that obtains in them between what is said and what is signalled. In the second example what is signalled is sufficient for the high probability of what is said. In saying that the election is out of Tom, Dick and Harry, I signal that the high probability for me of the triple disjunction is robust with respect to the conjunction of the negations of any two of the disjuncts (for example, that it is highly probable that Dick will be elected given that Harry won’t and Tom won’t). This is sufficient (by the calculus) for the high probability of the disjunction. In the first example what is signalled is arguably sufficient for the truth of what is said. If I say ‘I believe it will rain tomorrow’ in an appropriately serious tone, it is arguable that I signal that I do indeed believe it will rain tomorrow, and I am not, say, producing the sentence merely as a handy example of a belief-sentence. At any rate, it can hardly be denied that there is a close connection between what is said and what is signalled in this case.

According to our account, conditionals are yet another example of the same general kind. What is signalled by the assertion of (P→ Q) amounts to Pr(Q/P) being high. This is sufficient for Pr(P⊃Q) being high. So what is signalled is sufficient for the high probability of what is literally said. Hence, drawing on the moral of the two examples just discussed, dissent from (P→Q) may be prompted by dissenters giving a low value to Pr(Q/P) as much as by their giving a low value to Pr(P⊃Q). Moreover, the latter is sufficient for the former by the calculus, so all cases of dissent from (P→Q) are ones where Pr(Q/P) is low. This result squares with our intuitions. I dissent from ‘If it rains, the match will be



-21-

cancelled’ when the conditional probability of the match being cancelled, given rain, is low. Further, if Pr(Q/P) is low, both Pr(P⊃~Q) and Pr(~Q/P) are high. So our theory predicts assent to (P→~Q) when you dissent from (P→Q). And this is just how it turns out in practice. If you dissent from ‘If Fred went, he went by car’, you assent to ‘If Fred went, he did not go by car’.

(iii) Conditionals like ‘If he is speaking the truth, I’m a dutchman’ are often cited as being more hospitable to the Equivalence thesis than most. But they present a prima facie objection to our version. ‘He is speaking the truth ⊃ I’m a dutchman’ is not robust with respect to ‘He is speaking the truth’. Should he turn out to be speaking the truth, I won’t conclude that I’m a dutchman. The probability that I’m dutch given he is speaking the truth is low. But there is good reason to hold that dutchman conditionals are a very special case. For suppose what he is saying is that I am a dutchman. Then ‘If he is speaking the truth, I’m a dutchman’, standardly interpreted, is certainly true, but I would not use it in this case to express my utter disbelief in his truthfulness. Instead, I would say something like ‘If he’s speaking the truth, pigs have wings’. Therefore, the use of a dutchman conditional to express disbelief in its antecedent is not the standard one. The very circumstances in which ‘If he’s speaking the truth, I’m a dutchman’, standardly interpreted, is beyond doubt true are the very ones in which we would not use it in the way in question. Hence it is not an objection to our theory that it does not cover them. Our theory is a theory of the standard indicative conditional.

1.6 Postscript

Why not simply say the following about (P→Q)? We can distinguish truth conditions from assertion conditions. The truth conditions of (P→Q) are those of (P⊃Q). There are good and well-known arguments for this. And the assertion condition for (P→Q) is that Pr(Q/P) be high. There are good and well-known arguments for this. End of story.

My reason is that conjoining is not explaining. The problem is to explain one in terms of the other. Given the widely accepted view that the best approach to meaning and analysis is via truth conditions, we should hope for a theory which explains the assertion conditions in terms of the truth conditions. This is essentially what I have attempted. I have tried to show how a



-22-

plausible thesis about (P→Q)’s truth conditions, namely the Equivalence thesis, can, in the light of the importance of robustness for assertability, explain the plausible thesis about (P→Q)’s assertion condition, namely Adams’s thesis. In my view this puts a very different complexion on certain putative counter-examples to the Equivalence thesis. We saw, for instance, how granting the validity of contraposition can force the equivalence theorist into holding that ‘If Carter is re-elected by a large margin, then he will not be re-elected’ is true. But what is it that is immediately evident about this putative counter-example? Surely that it has very low assertability. But the Supplemented Equivalence theory explains this, and what a theory explains well cannot be an objection to that theory. 28 , 29

Notes 1 Though the point is commonly put in terms of evidence, see, e.g., Charles L.

Stevenson, ‘If-iculties’, Philosophy of Science, 37 (1970): 27-9 and G.H. Von Wright, ‘On Conditionals’ in Logical Studies, London, Routledge & Kegan Paul, 1957, p. 139.

2 Particularly in discussion, but see Richard C. Jeffrey, Formal Logic, New York, McGraw-Hill, 1967, ch. 3 and David Lewis, ‘Probabilities of Conditionals and Conditional Probabilities’, Philosophical Review, 75 (1976):297-315. For support for the general idea and other arguments for the Equivalence thesis, see Michael Clark, ‘Ifs and Hooks’, Analysis, 32 (1971):33-9.

3 I understand that this was the emphasis in H.P. Grice’s influential, unpublished William James Lectures; see L. Jonathan Cohen, ‘Some Remarks on Grice’s Views about the Logical Particles of Natural Language’ in Y. Bar-Hillel, ed., Pragmatics of Natural Languages, Dordrecht, Reidel, 1971, Michael Clark, ‘Ifs and Hooks’, and particularly ‘Ifs and Hooks: A Rejoinder’, Analysis, 34 (1974):77-83, A.J. Ayer, Probability and Evidence, London, Macmillan, 1972, Stevenson, ‘If-iculties’, and J.L. Mackie, Truth, Probability and Paradox, Oxford, Clarendon Press, 1973. Added note: Grice’s lectures have now been published as Studies in the Ways of Words, Cambridge, Mass., Harvard University Press, 1989.



4 In their presentation of Grice’s (tentative) views, Cohen et al. sometimes use formulations that are ambiguous about whether it is all or part of your evidence that is meant. If it is all, things are as above; but if it is part, the view being reported is that (P→Q) is assertable if part of your evidence favours (P⊃Q) without favouring one of ~P or Q, even if your total evidence favours one of them. There is immediate trouble for such a view. Suppose I know that Fred and Bill both live in Oak Street. Even though my evidence strongly favours the material conditional, it would normally be wrong to assert ‘If Fred lives in Elm Street, Bill lives in Elm Street’ in such a case;

-23-

nevertheless, part of what I know, namely, that they live in the same street, favours the material conditional without favouring its consequent and without favouring the negation of its antecedent.

5 Pace what appears to be Lewis’s suggestion, ‘Probabilities of Conditionals and Conditional Probabilities’, p. 308.

6 See, e.g., Richard C. Jeffrey, The Logic of Decision, New York, McGraw-Hill, 1965, and P.P. Ramsey, Foundations of Mathematics, London, Routledge & Kegan Paul, 1931, ch. 7. Robustness is a notion I first heard about some years ago from Manfred Von Thun in the context of weight in J.M. Keynes’s sense in the latter’s Treatise on Probability, London, Macmillan, 1921. Brian Skyrms uses ‘resilience’ for a similar notion: see ‘Physical Laws and Philosophical Reduction’ in G. Maxwell and R.M. Anderson Jr, eds, Induction, Probability, and Confirmation, Minnesota Studies in Philosophy of Science, vol. 6, Minneapolis, University of Minnesota Press, 1975. Neither should be held responsible for my use of the notion in what follows.

7 ‘Standard’ in that it is offered by opponents of the Equivalence thesis as well as by supporters; see, e.g., J.L. Mackie, Truth, Probability and Paradox, p. 76.

8 ‘Probabilities of Conditionals and Conditional Probabilities’, pp. 306-7. 9 The possible information may be actual. Obviously we are often interested in

robustness relative to what we might, but don’t to date, know. But this is not part of the definition of robustness. If it was, P would automatically become non-robust with respect to I on learning I! When I is known at t, our definition makes P robust with respect to I if and only if P is highly probable at t.

10 See, e.g., Michael Dummett, Frege, London, Duckworth, 1973, pp. 85-6.



11 See, e.g., Dummett, Frege and various of H.P. Grice’s papers, including ‘Logic and Conversation’ in Peter Cole and Jerry L. Morgan, eds, Syntax and Semantics, New York, Academic Press, 1975, vol. 3, and ‘Further Notes on Logic and Conversation,’ in Peter Cole, ed., Syntax and Semantics, New York, Academic Press, vol. 9, 1978. Added note: I discuss conversational and conventional implicature in a little detail in Frank Jackson, Conditionals, Oxford, Basil Blackwell, 1987, ch. 5.

12 Cf., Dummett, Frege, p. 316. 13 For recent examples see Ernest W. Adams, The Logic of Conditionals,

Dordrecht, Reidel, 1975, and J.L. Mackie, Truth, Probability and Paradox. 14 In my view, the objection to disjunctive syllogism in A.R. Anderson and N.D.

Belnap, Entailment, Princeton, Princeton University Press, 1975, conflates these two questions. Note particularly the top paragraph of their p. 177.

15 I here dissent from W.E. Johnson’s illuminating remarks in ch. 3 of Logic, Pt. I, New York, Dover, 1964. (Incidentally, saying ‘P only if Q’ does seem to signal robustness of (P⊃Q) with respect to ~Q.)

16 Some have held the similar position that ‘Q even if P’ entails Q. See,

-24-

e.g., Mackie, Truth, Probability and Paradox, p. 72, and John Pollock, Subjunctive Reasoning, Dordrecht, Reidel, 1976, p. 29. I would advance a similar objection to this position.

17 Adams, The Logic of Conditionals, and his earlier papers ‘The Logic of Conditionals’, Inquiry, 8, 2 (1965):166-97 and ‘Probability and the Logic of Conditionals’ in J. Hintikka and P. Suppes, eds, Aspects of Inductive Logic, Amsterdam, North-Holland, 1966, pp. 265-316. Strong evidence that he is essentially right is the number of authors of very different philosophical persuasions who have found this general kind of thesis congenial: e.g., Brian Ellis, ‘An Epistemological Concept of Truth’ in R. Brown and C.D. Rollins, eds, Contemporary Philosophy in Australia, London, Allen and Unwin, 1969, pp. 52-72; Richard C. Jeffrey, ‘If’, Journal of Philosophy, 61 (1964): 702-3; Robert Stalnaker, ‘Probability and Conditionals’, Philosophy of Science, 37 (1970):64-80, and David Lewis, ‘Probability of Conditionals and Conditional Probabilities’. Adams’s formula does not, of course, take into account the kind of ‘local’ sources of unassertability set to one side in section 1 such as obscenity, rudeness, and longwindedness.



18 ‘Probability of Conditionals and Conditional Probabilities’, pp. 300-2. Added note: this proof has generated a considerable literature—suggested ways around the result, alternative arguments to the same or similar conclusions, and the like. The upshot, as I read the literature, is to confirm Lewis’s original claim.

19 When P is inconsistent, Pr(P)=0, and Pr(Q/P) is undefined; accordingly, we need a ruling about the assertability of (P→Q) in such cases. The ruling I will follow is that all such conditionals are assertable. Others are possible.

20 For other examples of this kind, see Clark, ‘Ifs and Hooks: A Rejoinder’. 21 Added note: the failure of what is signalled to be preserved amounts to the

inference patterns in question failing to be necessarily assertibility-preserving, so another way of making the point of the preceding paragraphs is to say that intuitions about validity for inference patterns involving certain indicative conditionals track assertibility preservation rather than truth preservation. (And why shouldn’t they?)

22 From, e.g., David Lewis, Counterfactuals, Oxford, Basil Blackwell, 1973, and Ernest Adams, ‘Subjunctive and Indicative Conditionals’, Foundations of Language, 6 (1970):89-94. Added note: see also Chapter 2, ‘Classifying Conditionals’ and Chapter 3, ‘Classifying Conditionals II’.

23 See, e.g., Brian Ellis, ‘A Unified Theory of Conditionals’, Journal of Philosophical Logic, 7 (1978):107-24, and (with reservations) Adams, The Logic of Conditionals, ch. 4.

24 I argue this in detail in ‘A Causal Theory of Counterfactuals’, Australasian Journal of Philosophy, 55 (1977):3-21, but the general idea is widely accepted.

25 I am indebted here to the referee for drawing my attention to Robert

-25-

Stalnaker, ‘Indicative Conditionals’ in A. Kasher, ed., Language in Focus, Dordrecht, Reidel, 1976.



26 Stalnaker, ‘Indicative Conditionals’, pp. 182-7 shows how to express this in terms of the familiar possible worlds approach to counterfactuals (due to him, ‘A Theory of Conditionals’ in N. Rescher, ed., Studies in Logical Theory, Oxford, Basil Blackwell, 1968, and Lewis, Counterfactuals). According to Stalnaker, this approach works for both indicative and subjunctive conditionals, the difference between the two being due to the fact that in the case of the former but not the latter the similarity relation is constrained by the need to preserve common knowledge. When we consider (P→Q) at world i, we are to look for the closest P-world which shares with i what is being taken to be common knowledge in the context of assertion and ask whether it is a Q-world. But see the counter-examples below. Added note: I argue against the general programme of analysing indicative conditional in the possible worlds way in ‘Conditionals and Possibilia’, Proceedings of the Aristotelian Society, 81 (1980/81):126-37.

27 Cf., Cohen’s report of Grice’s views in ‘Some Remarks on Grice’s Views about the Logical Particles of Natural Language’.

28 I am conscious of a more than usually large debt to many discussions with Brian Ellis, Lloyd Humberstone, and Robert Pargetter; and also to comments from the referee.

29 Added note: this paper is a defence of the equivalence thesis for indicative conditionals with simple antecedents and consequents. Although, in an appendix to Conditionals, I make some sketchy remarks about certain special examples of conditionals embedded in conditionals, I have been unable to find anything systematic and satisfying to say on the subject. Some seem to think that anyone who defends the Equivalence thesis for simple conditionals must also maintain it for more complex varieties (which, I grant, would be very implausible). This seems to me like insisting that anyone who thinks that ‘or’ in ‘There will be red wine or white wine’ is equivalent to inclusive disjunction understood truth-functionally must say the same for the ‘or’ in ‘You may have red wine or white wine’. Of course, there is a presumption in favour of uniformity and simplicity, but it is only a presumption.

-26-



2 CLASSIFYING CONDITIONALS I Consider:

(1) If Booth had not killed Lincoln, someone else would have

(2) If Booth does not kill Lincoln, someone else will and

(3) If Booth did not kill Lincoln, someone else did.

Many writers agree that (1) is importantly different from (3). It is not just that we deny (1) and affirm (3)—a matter everyone agrees about—rather the explanation of why we deny (1) and affirm (3) is to be found in an important semantic difference between (1) and (3). 1 The details vary (needless to say), but there is considerable agreement that the key to the meaning of (1) lies in the fact that (1)’s truth conditions are to be given in terms of possible worlds in the style of Robert Stalnaker or of David Lewis, or of some reasonable variant thereon. 2 By contrast, the key to the meaning of (3) lies in the fact that the justified assertability of (3) is given by the conditional probability of (3)’s consequent given its antecedent, or in terms of some reasonable variant thereon. 3 This approach to (3)’s meaning is sometimes associated with the doctrine that (3) does not have truth conditions, and sometimes with the doctrine that (3) has the truth conditions of the material conditional. 4

This chapter takes for granted this general attitude to the relationship between (1) and (3). Our question is what then to say about (2). Should we group (2) with (1), or with (3)? If—as is common—we describe (1) as a subjunctive conditional and (3) as an indicative conditional, and go on to express the generalisation

-27-



of the idea that (1) and (3) are semantically distinct in an important way by saying that subjunctive and indicative conditionals are importantly semantically distinct, we presumably place (2) with (3). For (2) would commonly be classified as indicative. But this would be weak qua argument for classifying (2) with (3). 5 The fact that (1) is subjunctive whereas (3) is indicative is far from the only difference between them which might reasonably be supposed to be the relevant one, particularly in view of the notorious elusiveness of the subjunctive/indicative distinction in English. Moreover, one of the main reasons for holding that (1) and (3) are semantically distinct is the noted point that we deny (1) and affirm (3); but if we go by denial and affirmation, then (2) most naturally goes with (1). As Brian Ellis has particularly emphasised, our reasons for denying (1)—the absence of a backup killer, for instance—are those which would justify denying (2) from a temporal perspective just prior to Booth shooting. 6 What is more, suppose we acquired reason to affirm (1) by learning that Booth’s wife was ready to fire from the stalls if he missed from the balcony; we would in that case hold that (2) was the right thing to say beforehand, and so going by affirmation and denial would in this case as well put (1) and (2) together.

However, the evidence from affirmation and denial for classing (1) and (2) together is not as strong as it might seem at first. There is an important difference between the data about affirmation and denial as it applies to separating (1) from (3), and as it applies to where to locate (2). From a given epistemological standpoint—roughly, that which we actually occupy—we deny (1) and affirm (3). That is the datum that most often motivates giving different accounts of (1) and (3). But it is not true that from a given epistemological standpoint we deny (2) while denying (1) and affirming (3). We deny (1) and affirm (3) from the standpoint we actually occupy after the event, but deny (2) from a standpoint we imagine ourselves to be occupying beforehand. There is no one epistemological standpoint from which we give different answers about (2) and (3). Hence, in the case of (2) and (3) it is an open possibility that the difference in regard to affirmation and denial can be explained by the difference in epistemological standpoints, thus allowing us to class (2) and (3) together, semantically speaking, and apart from (1).

The principal burden of this chapter is that we should, semantically speaking, class (2) and (3) together, and apart from (1). The semantic line goes where the mood line goes, between (1)

-28-



and (2), not between (2) and (3). After I have argued this, I will return to the question of providing an epistemological explanation of why we affirm (3) but, from the imagined perspective before the event, deny (2).

I think that there are a number of reasons for classifying (2) with (3), 7 but the case I will develop here turns on that provided by Allan Gibbard’s Sly Pete example. 8 I think that the significance of this much-discussed example has been left rather obscure in the literature. Indeed, Bennett takes the example to provide reason for classifying (2) with (1). 9 I will argue, though, that once the lesson of the example is properly understood it provides a strong reason for classifying (2) with (3).

It will be useful to have labels for conditionals relevantly like (1), like (2), and like (3). I will call conditionals relevantly like (1) ‘past subjunctives’, like (2) ‘future indicatives’, and like (3) ‘past indicatives’. I use these labels because they are convenient—you will know the conditionals I have in mind—and set to one side whether there are better labels to be had. Thus the main issue we will be concerned with can be expressed as how to locate semantically future indicative conditionals: with past subjunctives as Bennett holds, or with past indicative conditionals as I hold. 10

2.1 The natural approach to conditionals

My argument will be that although the natural approach to conditionals applies to past subjunctive conditionals, the Sly Pete example shows that it does not apply to past indicative conditionals and does not apply to future indicative conditionals. And, as will be clear when I describe the natural approach, this difference between past subjunctive on the one hand, and past and future indicative conditionals on the other, is most certainly a semantic one.

The natural approach is, as its name suggests, the natural one to take, the one we would like to take if we could. It has two clauses: (Support) and (Conditional non-contradiction).

(Support). On the face of it, acquiring information which justifies the assertion of a conditional is not different in kind from acquiring evidence which justifies the assertion of a non-conditional sentence. Reading the weather report justifies my saying that it will rain, reading the rules of cricket justifies my saying that if it rained heavily, the match was cancelled. Learning about the stock exchange may justify my saying that a certain stock will

-29-



go up, or it may justify my saying that if a certain stock goes up, certain other stocks will go down, and so on and so forth. By saying that there is, on the face of it, not a difference in kind, I simply mean that both cases seem alike in essentially involving the acquisition of evidence that makes what is asserted more likely to be true. The information is support in the standard sense of making very probably true. (Support) is thus incompatible with the view that conditionals do not have truth conditions, only assertion conditions. But I think that it is important to appreciate that this deflationary position is one we need to be driven into; it is not one to embrace early on. For it makes a mystery of the sense in which information justifies, in the relevant sense, asserting a conditional. For instance, a lot of evidence that, in the past, economic problems have been followed by election reverses for the government justifies saying that if the economy continues to go badly, the government will be in trouble at the next election. But this is not justification in the sense of making the saying morally justified, or pragmatically advisable, or in accord with good etiquette. What then is the relevant sense of justification? The obvious answer is that it is justification in the sense of making probably true.

(Conditional non-contradiction). There is something very wrong with a given person saying both ‘If A then B’ and ‘If A then not B’, or at least there is when A is consistent. In tandem with this is the fact that if you say ‘If A then B’ and I say ‘If A then not B’, we count as disagreeing, or at least we do when A is consistent. (Reductio proofs in logic and mathematics show that when A is inconsistent the situation is different.) The obvious explanation of these facts is that ‘If A then B’ and ‘If A then not B’ are (logical) contraries when A is consistent. We will follow the practice of referring to this as (Conditional non-contradiction). (The name perhaps suggests the doctrine that, when A is consistent, not only are ‘If A then B’ and ‘If A then not B’ contraries, they are also contradictories in the sense that they must have opposite truth values. The argument that follows goes through on either interpretation, but we will work in terms of the weaker one.)

If a possible worlds semantics of one of the more familiar kinds applies to a given class of conditionals, then there is no problem about holding that the natural approach applies to that class. If ‘If A then B’ is true if and only if (something along the lines of) the closest A-worlds are all B-worlds, then ‘If A then B’ is (a) truth-

-30-



valued, (b) supported by evidence in the standard sense because information garnered about how things actually are can make it more or less likely that nearby antecedent worlds are consequent worlds, and (c) inconsistent with ‘If A then not B’ when A is consistent, for the closest A-worlds cannot be both B-worlds and not-B-worlds. I think, in common with many, that the possible worlds semantics applies to past subjunctive conditionals. Hence, I take it that there is no problem about adopting the natural approach for past subjunctive conditionals, including of course (1). The problem is with past and future indicative conditionals. The Sly Pete example shows that (unfortunately) the natural approach fails for them; although, as we will see, it takes a little work to see exactly how to argue the point.

2.2 The Sly Pete example and some false starts

There are a number of ways of setting out this example. We will work with the following version. It is in terms of past indicatives. We will discuss future indicatives shortly.

Sly Pete is playing poker. He has to decide between calling and folding. Informant X knows that Sly Pete is cheating, and that he is doing so by knowing what is in his opponent’s hand. (Perhaps Sly Pete can see his opponent’s hand in a mirror in the ceiling.) Informant Y has surreptitiously observed both hands and knows that Sly Pete has the weaker hand. We can suppose that it is the last hand for the night in order to avoid the complication that Pete might want to lose the hand in order to encourage his opponent to bet higher later in the game. Informants X and Y leave the room just before Pete makes his decision and report to you as follows. X says that if Pete called, he won. Y says that if Pete called, he lost. You respond (correctly) by inferring that Pete did not call, he folded.

How does this example make trouble for the natural approach as applied to past indicative conditionals? Gibbard sees things this way. He points out that both informant X and informant Y are fully justified in their assertions, are working from information properly so called, that is, which contains no mistakes, and are being sincere. From this he argues that ‘we can see that neither is asserting anything false’, and so that if ‘both… utterances express propositions,…both express true propositions’. 11 To put matters in our terms, Gibbard is arguing that the (Support) part of the natural approach when combined with the

-31-



fact that informant X and informant Y are sincere and are proceeding in a fully justified manner from data that contains no mistakes would mean that what they each say is true, and so that (Conditional non-contradiction) fails. Hence, the natural approach is false, for its two clauses cannot be true together.

The trouble with this argument is that it is perfectly possible to proceed in a fully justified manner from information properly so-called to a false conclusion. That is the point behind the familiar distinction between deductive and inductive arguments. In particular, it is perfectly possible for two different bodies of information relevant to a common subject matter to sustain incompatible conclusions about that subject matter. Smith knows that Petersen is a churchgoing Swede and that nearly all churchgoing Swedes are Protestants. She is entitled to infer with considerable confidence that Petersen is a Protestant. Jones knows that Petersen has Catholic parents and that nearly all churchgoers with Catholic parents are Catholic. She is entitled to infer with considerable confidence that Petersen is a Catholic. All this is compatible with Smith’s information being information properly speaking, correct in every detail, and likewise for Jones’s information. Of course, their two bodies of information are incomplete, but the same goes for the bodies of information available to informant X and to informant Y in the Sly Pete example.

There is, of course, a special feature of the Sly Pete example. In addition to the fact that both informants proceed justifiably from correct if incomplete information, there is the point that you, the hearer, are able to use their two utterances constructively to infer that Sly Pete did not call. This suggests that (a) ‘If Sly Pete called, he won’ conjoined with ‘If Sly Pete called, he lost’ entails that Sly Pete did not call, and that (b) it is rational to accept, in the circumstances as specified in the Sly Pete example, the conjunction of ‘If Sly Pete called, he won’ with ‘If Sly Pete called, he lost’.

This way of construing the argument based on the Sly Pete example avoids the mistake of supposing that what comes rationally from what is the case must be true. It rests on the idea that if it is rational to accept ‘If Sly Pete called, he won’ together with ‘If Sly Pete called, he lost’, they can hardly be contraries. 12 The problem for this way of construing the argument is that it is not sufficiently obvious that your inference to ‘Sly Pete did not call’ depends inter alia on accepting the conjunction of the two con-

-32-



ditionals. There is another hypothesis available to explain what is going on.

An important feature of the Sly Pete story is that you know the kind of information available to your two informants. You know that one informant knows that Sly Pete is cheating by knowing his opponent’s hand, and that the other informant knows that Sly Pete’s opponent has the stronger hand; and this fact plays a crucial role in warranting your inference from their two utterances to the conclusion that Sly Pete did not call. For suppose that you had had no idea of the evidence available to them, and that all you knew was that they were sincere and sensible. In that case you could have done nothing constructive with their two utterances, and in particular you could not have inferred that Sly Pete did not call. For suppose you cannot now remember whether you had your child inoculated against whooping cough and one doctor says to you, ‘If you had your child inoculated against whooping cough, then you did the right thing’, while another says to you, ‘If you had your child inoculated, you did the wrong thing’; you do not infer without further ado that you did not have your child inoculated against whooping cough. If in a sporting broadcast you hear Fred Stolle say that if Chris Evert won last night, she went one ahead in the head to head tally against Martina Navratilova, and then hear John Newcombe say that if Evert won, the tally is dead level, you do not take your high regard for the sincerity and good sense of Stolle and Newcombe as justifying the conclusion that Chris Evert lost. You simply infer that one of them has made a mistake.

Now if it is crucial in the Sly Pete example that you know a good deal about your informants’ evidence, there is an alternative hypothesis about how you reached your conclusion that Sly Pete did not call. We do not have to suppose that you inferred from the two conditional utterances taken together—a supposition which would indeed count against (Conditional non-contradiction)—we can instead suppose that you inferred from what you know of the evidence available to your informants. You know that informant X is using evidence that Sly Pete is cheating, informant Y is using evidence that Sly Pete’s hand is the weaker, and those two bits of evidence taken together are in themselves enough in the circumstances to warrant inferring that he did not call. We do not need to go via the informants’ conditional utterances to explain how you were entitled to your conclusion.

In a variant on the Sly Pete example, your informants’ conditional

-33-



utterances tell you something important about their respective bodies of evidence. In the variant you do not know initially that informant X knows that Sly Pete is cheating. You know merely that informant X knows whether or not Sly Pete is cheating, and knows nothing else relevant. Similarly, you do not know initially that informant Y knows that Sly Pete has the weaker hand. You know merely that informant Y knows who has the weaker hand, and knows nothing else relevant. In this variant case the conditional utterances of X and Y do give you highly relevant information. When X says ‘If Sly Pete called, he won’, you learn that your informant knows that Sly Pete is cheating, for otherwise, given what you already know about X, X would not be justified in saying that if Sly Pete called, he won. When Y says ‘If Sly Pete called, he lost’, you learn that your informant knows that Sly Pete’s hand is the weaker one, for otherwise, given what you already know about Y, Y would not be justified in saying that if Sly Pete called, he lost. (We assume known sincerity and good sense in X and in Y.) But even in this version of the Sly Pete example where your two informants’ utterances play a major role in enabling you to conclude that Sly Pete did not call, we do not have to suppose that it is what they say as such that you are using. A believer in (Conditional non-contradiction) can insist that you are using what their two utterances tell you about their evidence, not the contents of the utterances themselves, to conclude that Sly Pete did not call. And this insistence would not be ad hoc. Believers in (Conditional non-contradiction) are entitled to make much of the point already emphasised that the two assertions of your informants, even if known to be sincere and sensible, in the absence of any information about the evidence that lies behind them, would not enable you to conclude that Sly Pete did not call. How so, if the evidence that backs their assertions is not the key?

2.3 The Sly Pete argument and a principle about evidence

Despite these failures, I think that it is possible to mount an argument from the Sly Pete example against the natural approach. I will refer to such an argument as a Sly Pete argument. It rests on a principle about evidence in favour of hypotheses which are logical contraries.

You might be tempted by the following principle about evidence

-34-



in favour of contraries: if e1 strongly supports H1 , e2 strongly supports H2 , and H1 and H2 are contraries, then e1 is unlikely given e2 , and e2 is unlikely given e1 . But here is a counter-example to it. Set e1 to be ‘Petersen is a Swedish Catholic’, e2 to be ‘Petersen is a Swede’, H1 to be ‘Petersen is Catholic’, and H2 to be ‘Petersen is not Catholic’. e1 strongly supports H1 (by entailing it), e2 strongly supports H2 (relative to a background which includes the fact that nearly all Swedes are Protestants), H1 and H2 are contraries, and yet it is false that e2 is improbable given e1 . In fact e1 entails e2 . It is, however, true that e1 is improbable given e2 (relative to the background evidence), and indeed the following principle is valid.

(P) If e1 is strong evidence for H1 , e2 is strong evidence for H2 , and H1 and H2 are contraries, then e1 is improbable given e2 , or e2 is improbable given e1 (or both).

More precisely, we can specify a value such that if Pr(H1 /e1 ) and Pr(H2 /e2 ) both exceed it, one or both of Pr(e1 /e2 ) and Pr(e2 /e1 ) are <0.5.

It is by reference to (P) that we show that the Sly Pete example refutes the natural approach to conditionals as applied to past indicatives. Informant X’s information is that Sly Pete knows what is in his opponent’s hand. This information, along with such facts as that Sly Pete wants not to lose money on this hand and understands the rules of poker, very strongly warrants X’s assertion that if Sly Pete called, he won. Similarly, informant Y’s information that Sly Pete has the weaker hand, in the appropriate circumstances very strongly warrants Y’s assertion that if Sly Pete called, he lost. However, Sly Pete’s knowing his opponent’s hand need not count against Sly Pete’s having the weaker hand. Similarly, Sly Pete’s having the weaker hand need not count against his knowing his opponent’s hand. Typically, ‘Sly Pete knows his opponent’s hand’ and ‘Sly Pete has the weaker hand’ are evidentially neutral with respect to each other. In the example as originally described by Gibbard, Sly Pete has an accomplice who signals the contents of the opponent’s hand to Pete. We imagined that Pete can see his opponent’s hand reflected in a mirror in the ceiling. Typically, the successful operation of systems of these kinds is probabilistically quite independent of which hand is the stronger of the two.

The same point can be made in terms of other examples.

-35-



Information that a certain motion put at a certain meeting was a bad one would strongly warrant asserting ‘If Fred voted for the motion, he did the wrong thing’. Information that Fred has excellent judgement in such matters would strongly warrant asserting ‘If Fred voted for the motion, he did the right thing’. However, the fact that a certain motion was bad does in itself not count against (or for) the quality of Fred’s judgement; and, conversely, Fred’s excellent judgement does not in itself count against (or for) the motion being bad. We must conclude, therefore, that the Sly Pete example shows via (P) that the natural approach fails for past indicative conditionals. It cannot simultaneously be the case that (a) the two bodies of information that very strongly warrant the assertion, respectively, of ‘If Sly Pete called, he won’ and ‘If Sly Pete called, he lost’ are evidence in the standard sense of making the conditional whose assertion is warranted highly probably true, and (b) the conditionals are logical contraries, (a) and (b) cannot be true together because the two bodies of information may each be not improbable given the other. It follows that (Support) and (Conditional non-contradiction) clash in the case of past indicatives. But as we saw (or rather noted had been widely granted) the natural theory is true for past subjunctives, so we have the result that past indicative conditionals are semantically distinct from past subjunctive conditionals.

By itself this result is no advance. We started by noting the semantic distinctness of (1) from (3). We are now, though, in a position to argue that (2), the future indicative, should be classed with (3), the past indicative, rather than with (1), the past subjunctive. We can, that is, extend the result to show the distinctness of (1) from (2). We can now say with some precision how a Sly Pete argument can show that the natural approach to conditionals fails for conditionals of some specified class. It does so by describing a case where there are two bodies of evidence, e1 and e2 , and a pair of conditionals, ‘If A then B’ and ‘If A then not B’ of the class in question with A consistent, such that against some given background: (a) e1 by itself strongly warrants (as highly as you care to make it by varying the background) asserting ‘If A then B’, (b) e2 by itself strongly warrants asserting ‘If A then not B’, and yet (c) e1 is not improbable given e2 , and e2 is not improbable given e1 .

A Sly Pete argument can be mounted for future indicative conditionals in much the same way as for past indicative conditionals. 13 Although the Sly Pete example is commonly described

-36-



in the past tense, this is an accidental feature of the example. We imagined your two informants reporting to you after Sly Pete’s decision, but we could equally have imagined them to be reporting just beforehand. Informant X knows that Sly Pete knows his opponent’s hand and says to you, with full justification, ‘If Sly Pete calls, he will win’. Informant Y knows that Sly Pete’s hand is the weaker and says, with full justification, ‘If Sly Pete calls, he will lose’. And, as before, X’s information does not count against Y’s, and Y’s information does not count against X’s. Thus the case against the natural theory applying to future indicative conditionals is as strong as that against its applying to past indicative conditionals.

2.4 Sly Pete and past subjunctives

An obvious question to ask at this point is whether a Sly Pete argument is possible for past subjunctive conditionals. It had better not be, given our earlier presumption that the natural approach applies to past subjunctives.

I think however that we can see that a Sly Pete argument is not possible for past subjunctives. If we take the Sly Pete example and replace the pair of past indicative conditionals by the corresponding pair of past subjunctive conditionals, we get ‘If Sly Pete had called, he would have won’ and ‘If Sly Pete had called, he would have lost’. Now the second of these conditionals is supported by the information that warrants the assertion of the corresponding past indicative (or the corresponding future indicative, if it comes to that). The information that Sly Pete had the weaker hand justifies saying both that if he called, he lost, and that if he had called, he would have lost. However, the first of these past subjunctives—‘If Sly Pete had called, he would have won’—is not warranted by the information about his cheating that warrants the assertion of the corresponding past indicative. 14

To see this, consider the matter from the perspective of informant X who is in possession of this information, that is, the information that Sly Pete is cheating by knowing his opponent’s hand. We noted before that this information is neutral concerning who has the stronger hand, so let us suppose that X gives equal credence to the hypothesis that Sly Pete’s hand is stronger, and to the alternative hypothesis that it is weaker than his opponent’s. What X can be sure about though is that Sly Pete will do the right, in the sense of the most rewarding, thing. That is his return

-37-



for cheating. But what is the right thing for Sly Pete to do? The answer depends on whether or not he has the stronger hand. If he has the stronger hand, the right thing to do is to call; if he has the weaker hand, the right thing to do is to fold. Hence, informant X gives a 50 per cent credence to calling being the right action for Sly Pete, and 50 per cent to folding being the right action for Sly Pete. But what makes calling the right action for Sly Pete? Well, he had two options—to call or to fold, and calling is best if it had or would have had the better consequences of the two. But that is the case precisely if had Sly Pete called, he would have won. That is to say, the right credence for informant X to give ‘If Sly Pete had called, he would have won’ is the same as the credence he gives to calling being the right thing for Sly Pete to have done, which is 50 per cent. The upshot is that although informant X’s information that Pete is cheating makes ‘If Sly Pete called, he won’ very highly assertable, it leaves the credence of the corresponding past subjunctive unchanged at 50 per cent. 15 In brief, informant X’s information that Pete is cheating justifies asserting ‘If Sly Pete did not call, then had he called, he would have lost’, but does not justify asserting outright either that had he called he would have lost, or that had he called he would have won.

2.5 Wherein lies the semantic difference?

Our argument has been that there is a semantic difference between past subjunctive conditionals on the one hand, and past and future indicative conditionals on the other: the natural approach applies to the former but not to the latter. But our argument has been silent on wherein lies the difference, for we have not addressed the question of precisely where the natural approach fails for the past and future indicative conditionals. There would appear to be three live possibilities. I will not try to adjudicate between them here. I mention them to give a sense of the possibilities opened up by the Sly Pete argument.

You might hold that (Support) fails. Information that (strongly) justifies asserting past and future indicative conditionals does not make them (highly) probably true. The most familiar version of this response holds that past and future indicative conditionals are not truth-valued to start with, and so that there is no question of anything making them probably true. 16

Second, you might hold that it is (Conditional non-contradiction) which fails. Even when A is consistent, ‘If A then B’ and ‘If A

-38-



then not B’ can be true together. The most familiar version of this response gives past and future indicative conditionals the truth conditions of the material conditional.

Finally, you might hold that what is perhaps best described as a presupposition of (Conditional non-contradiction) fails. For you might hold that the truth conditions of past and future indicative conditionals are a function of the context of assertion, that is to say that the proposition expressed by ‘If A then B’ varies with that context. More particularly, you might hold that (Conditional non-contradiction) is valid in the following sense: relative to a given context, the proposition expressed by ‘If A then B’ and the proposition expressed by ‘If A then not B’ are contraries provided A is consistent. This is why it is wrong for a given person at a given time to assert both together. However, what may happen is that the proposition expressed by ‘If A then B’ relative to one context is consistent with the proposition expressed by ‘If A then not B’ relative to some other context. 17 And you might hold that exactly this happens in the Sly Pete example: that is to say, you might hold that the proposition that X expresses by saying ‘If Sly Pete called, he won’ is consistent with the proposition that Y expresses by saying ‘If Sly Pete called, he lost’, though it is not consistent with the proposition that X would have expressed had X used Y’s words.

(Of course it just might be that the way in which the standard approach fails for future indicative conditionals is different from the way it fails for past indicative conditionals. This is a consequence of the fact that the standard approach has two clauses.)

2.6 Epistemology and the data about affirmation and denial

We noted at the very beginning that if we went by affirmation and denial, we would class (1) and (2) together, and so separate them from (3), and I acknowledged an obligation qua proponent of classing (2) with (3) semantically, to give an epistemological explanation of the fact that we deny (1), would beforehand deny (2), and affirm (3), and I noted the abstract possibility of doing this. I now want to make this abstract possibility concrete by describing a situation in which we would deny (1) and affirm both (2) and (3), a case, that is, where going by affirmation and denial, places (2) and (3) together. It will be apparent that the

-39-



difference between this situation and our actual one is an epistemological one.

Our justified beliefs about what will happen in the future often depend heavily on our opinions about what the past and present will give rise to. My beliefs about who will win the next election depend heavily on my opinions about the likely consequences of present and past conditions. By contrast, my opinion—indeed, knowledge—about who won last year’s election is relatively independent of the more or less speculative views I have about what caused that electoral victory. Similarly, my view beforehand about Lincoln’s being killed would normally depend on my view about how it might happen that he comes to be killed. But we can have exceptions to this norm. I may expect Lincoln to be killed by being killed by Booth working alone, and yet be more confident that he will be killed by someone or other working alone than I am that it will be by Booth. Perhaps someone I know to be very knowledgable about the plot has assured me that Lincoln will be assassinated by someone acting alone, but she remained silent on who it would be. Booth is just my educated guess.

What I do I say beforehand? I affirm (2), that if Booth does not kill Lincoln, someone else will. I expect it to be Booth, but, be I right or wrong about that part of the story, what I am confident of is that it will be someone or other, and so if not Booth, then someone else. What do I say afterward when I learn that Lincoln was indeed killed as predicted, that is, by someone working alone, and that it was almost certainly, as I had anticipated, Booth? Clearly I do not affirm that if Booth had not killed Lincoln, someone else would have. Booth acted alone. If he had not killed Lincoln, the plot would have failed and Lincoln would not have been killed. So I deny (1). Finally, provided that although I am confident that it was Booth who killed Lincoln, I am even more confident that Lincoln was killed, I will affirm (3), namely, that if Booth did not kill Lincoln, someone else did.

We have, therefore, concrete reason to hold that the fact that in the normal case we would, going by affirmation and denial, put (2) with (1), and not with (3), has an epistemological explanation not a semantic one, a reason to do with the difference in the epistemic status that normally obtains between opinions about the future and the past. For if we change that status in the appropriate way, going by affirmation and denial puts (2) with (3), not (1).

My case for semantically separating past subjunctive conditionals on the one hand, from past indicative conditionals and

-40-



future indicative conditionals on the other, is now before you. Is it theoretically desirable to describe this by saying that the, or anyway an, important line between conditionals is between indicative conditionals and subjunctive conditionals? This depends on what should be said about future subjunctive conditionals, about which we have said nothing, and on the more general question of the viability in English grammar of dividing indicative from subjunctive moods. Here I am content to insist that in respect to one major feature, the line has (1) on one side and (2) and (3) on the other. 18

Notes 1 See, e.g., Jonathan Bennett, ‘Farewell to the Phlogiston Theory of Conditionals’,

Mind, 97 (1988):509-27; Allan Gibbard, ‘Two Recent Theories of Conditionals’ in W.L. Harper et al, eds, Ifs, Dordrecht, Reidel, 1981, pp. 211-47; David Lewis, Counterfactuals, Oxford, Basil Blackwell, 1973, and Frank Jackson, Conditionals, Oxford, Basil Blackwell, 1987. For a dissenting voice, see Brian Ellis, ‘A Unified Theory of Conditionals’, Journal Of Philosophical Logic, 7 (1978): 107-24. I follow Bennett in here using Booth and Lincoln instead of the often-used Oswald and Kennedy example.

2 David Lewis, Counterfactuals, and Robert Stalnaker, ‘A Theory of Conditionals’ in N. Rescher, ed., Studies In Logical Theory, American Philosophical Quarterly monograph series no. 2, Oxford, Basil Blackwell, 1968, pp. 98-112.

3 See, e.g., Ernest Adams, ‘The Logic of Conditionals’, Inquiry, 8, 2 (1965):166-97, and The Logic of Conditionals, Dordrecht, Reidel, 1975.

4 For examples of the former response, see Adams, The Logic of Conditionals, and Dorothy Edgington, ‘Do Conditionals have Truth Conditions?’, Critica, 98 (1986):3-30, reprinted in Frank Jackson, ed., Conditionals, Oxford, Oxford University Press, 1991; for an example of the latter response, see Chapter 1, ‘On Assertion and Indicative Conditionals’, and Lewis, ‘Probabilities of Conditionals and Conditional Probabilities’, Philosophical Review, 75 (1976): 297-315.

5 As Bennett, ‘Farewell to the Phlogiston Theory of Conditionals’ and Michael Pendelbury, ‘The Projection Strategy and the Truth Conditions of Conditional Statements’, Mind, 98 (1989):179-205, note.

6 Ellis, ‘A Unified Theory of Conditionals’. 7 For some reasons different from the one I give here, see Jackson, ‘Two Theories

of Indicative Conditionals: Reply to Brian Ellis’, Australasian Journal of Philosophy, 62 (1984):67-76.

8 In ‘Two Recent Theories of Conditionals’. 9 See Bennett, ‘Farewell to The Phlogiston Theory of Conditionals’; Pendelbury,

‘The Projection Strategy and the Truth Conditions of

-41-



Conditional Statements’, also holds (but for different reasons) that (2) should be classified with (1).

10 I am, of course, also disagreeing with Pendelbury, ‘The Projection Strategy and the Truth Conditions of Conditional Statements’. But he labels (2) as well as (1) as ‘subjunctive’; see p. 192. Hence, we come out in verbal agreement but remain in disagreement over the substantive question of (2)’s placement. Added note: Bennett has recanted; see ‘Classifying Conditionals: The Traditional Way is Right’, Mind, 104 (1995):331-54.

11 Gibbard, ‘Two Recent Theories of Conditionals’, p. 231. Helpful correspondence with Gibbard has convinced me that though my interpretation of Gibbard’s argument is the best interpretation of the printed word, it is probably not the line of argument he intended (although I confess that the precise direction of the intended line of argument is still obscure to me).

12 If S and S* are sentences which are hard to comprehend fully, it can be rational to accept S along with S* even when S and S* are (unobviously) contraries. This is a point familiar to logicians and mathematicians. But it is not plausible to think that our two sentences fall into this category.

13 As Gibbard, ‘Two Recent Theories of Conditionals’, p. 228, notes, though he also notes some complications that space precludes our pursuing.

14 Added note: but see Chapter 3 ‘Classifying Conditionals II’, where I grant to E.J. Lowe that there is a back-tracking (and in my view forced) reading of this subjunctive, on which the information that he is cheating would warrant asserting the subjunctive; but, as I note there, the existence of two readings of this kind of subjunctive leaves the essential point undisturbed.

15 This point has, of course, played a central role in recent discussions of decision theory, see, e.g., Allan Gibbard and William L. Harper, ‘Counterfactuals and Two Kinds of Expected Utility’, reprinted in W.L. Harper et al, eds, Ifs.

16 See, e.g., Adams, The Logic of Conditionals; Gibbard, ‘Two Recent Theories of Conditionals’, and Edgington, ‘Do Conditionals have Truth Conditions?’.

17 See, e.g., Stalnaker, ‘Indicative Conditionals’, reprinted in Harper et al., eds, Ifs. 18 I am indebted to too many helpful comments on earlier versions to list, but must

mention Lloyd Humberstone, David Lewis, and especially Peter Menzies who gave me invaluable help with (P).

-42-



3 CLASSIFYING CONDITIONALS II Consider a past indicative conditional like ‘If it rained heavily, then the match was cancelled’. Depending on the evidence to hand, this conditional may be one which I am highly justified in asserting, one which I am highly justified in denying, or it may fall somewhere in between. And when I am not highly justified in asserting it, typically there will be a proposition which is such that if I knew it, then I would be highly justified in asserting the conditional. For instance, if I do not know whether the game in question is rugby or cricket, I may well not be in a position to assert that if it rained heavily, then the match was cancelled, but there will be a proposition, namely, that the game in question was cricket, which is such that if I knew it, then I would be fully justified in asserting that if it rained heavily, the match was cancelled. The obvious way to account for these facts is to hold that the sentence ‘If it rained heavily, then the match was cancelled’ expresses a proposition (and so is truth-valued), and moreover a proposition which relative to the evidence I actually have is not particularly probable, but which given that the game is cricket, is very highly probable.

What makes the philosophy of past indicative conditionals so hard, or part of what makes it so hard, is that the plausible view just described (labelled (Support) in Chapter 2) conflicts with an equally plausible view (labelled (Conditional non-contradiction) in Chapter 2), namely, that the proposition expressed by ‘If A then B’ cannot be true together with that expressed by ‘If A then not B’ when A is consistent: the proposition expressed by ‘If it rained heavily, then the match was cancelled’ cannot be true together with the proposition expressed by ‘If it rained heavily, then the match was not cancelled’.

I made the case for the incompatibility of (Support) and

-43-



(Conditional non-contradiction) via Allan Gibbard’s Sly Pete example using the pair

(1) If Sly Pete called, he won

(2) If Sly Pete called, he lost.

But the point applies generally to past indicative conditionals. I will presume familiarity with the Sly Pete example and the details of the argument (the ‘Sly Pete argument’), but we need a bare outline to refer back to. In Chapter 2 I developed the argument using

(P) If H1 and H2 are contraries and P(H1 /e1 ) and P(H2 /e2 ) are sufficiently high, then either P(e1 /e2 ) or P(e2/e1 ) is less than 0.5.

I could equally have used the more informative

(P*) If H1 and H2 are contraries and P(H1 /e1 ) and P(H2 le2 ) are both greater than 1 x, then either P(e1 /e2 ) or P(e2 / e1 ) is less than 2x.

The key point in the argument was that there is an e1 , namely,

(A) Sly Pete is cheating and knows what is in his opponent’s hand

and an e2, namely,

(B) Sly Pete has the weaker hand

such that someone who knows just (A)—informant X—can assert with full justification ‘If Sly Pete called, he won’, whereas someone who knows just (B)—informant Y—can assert with full justification ‘If Sly Pete called, he lost’. And yet it need not be the case that one of P((A)/(B)) and P((B)/(A)) is less than 0.5. (Or using the quantitative (P*): given (A) ‘If Sly Pete called, he won’ may be 0.9 assertable, given (B) ‘If Sly Pete called, he lost’ may be 0.9 assertable, and yet it need not be the case that one of P((A)/(B)) and P((B)/(A)) is less than 0.2.) It follows that one (or



-44-

both) of (Support) and (Conditional non-contradiction) fail for past indicative conditionals.

I take it that E.J. Lowe’s principal objection is to the use I made of this argument to drive a wedge between past indicative conditionals and past subjunctive conditionals. 1 I sought to drive that wedge by arguing that Sly Pete examples could not be used to show that (Support) and (Conditional non-contradiction) were incompatible in the case of past subjunctive conditionals. I think that Lowe is right that the reason I gave for saying that the argument failed in the case of past subjunctive conditionals was mistaken. But I think, nevertheless, that the argument does fail for past subjunctive conditionals, and in a way which supports my contention in Chapter 2 that Sly Pete examples show that there is an important difference between past indicative and past subjunctive conditionals.

Consider the past subjunctives corresponding to (1) and (2), namely,

(3) If Sly Pete had called, he would have won

(4) If Sly Pete had called, he would have lost.

I argued that although it is the case that (B)—Sly Pete has the weaker hand—strongly justifies asserting (4) as well as (2), it is not the case that (A)—Sly Pete knows what is in his opponent’s hand—strongly justifies asserting (3) as well as (1). For it does not justify asserting (3). Information that Sly Pete knows what is in his opponent’s hand justifies being confident that Pete did not call unless his hand was the stronger, but does not justify a claim about what would have happened had he called. Cheating does not alter what would have happened had Sly Pete called, but rather enables Pete to know what would have happened, and so to select with safety the best strategy.

I think that Lowe is right when he urges that the situation is more complex than this. Both (3) and (4) are ambiguous—their interpretation is context dependent—in a way most simply expressible in terms of the familiar possible worlds semantics. When we consider the closest possible worlds where the antecedent common to (3) and (4) is true, namely, worlds where Sly Pete calls, we have a choice between keeping constant Pete’s state of knowledge or of keeping constant the relative strengths of



the hands. If we keep constant the relative strengths of the

-45-

hands, what I said in Chapter 2 applies. Information that Pete knows what is in his opponent’s hand does not warrant asserting that had he called, he would have won. However, it is perfectly acceptable to keep constant Pete’s state of knowledge, and if you do that then information that Pete knows what is in his opponent’s hand does warrant asserting that had Pete called, he would have won. This is a kind of ‘back-tracking’ reading. You ask how it might have come about that Pete called. And if you know that he knows what is in his opponent’s hand, you know that he would have called only if he had the best hand, and so only if he would have won. I was wrong to insist on only one reading (though I still think that the ‘no back-tracking’ reading I insisted on is the most natural one).

None of this, however, negates my contention in Chapter 2 that Sly Pete arguments do not show that (Support) and (Conditional non-contradiction) cannot be true together for past subjunctive conditionals. The crucial point about past indicative conditionals is that it is true simultaneously that information that Pete is cheating would strongly warrant asserting (1), and that information that Pete has the weaker hand would strongly warrant asserting (2). However, the situation with past subjunctive conditionals is different. There is no uniform way of resolving the potential ambiguity in (3) and (4) on which it is true both that information that Pete is cheating supports (3) and information that Pete has the weaker hand supports (4). Rather, there is one way of resolving the ambiguity—the way which keeps Pete’s state of knowledge constant when you go to the closest antecedent worlds—on which information that Pete is cheating would warrant asserting (3) but information that Pete has the weaker hand would not warrant asserting (4), and there is another way of resolving the ambiguity—the way which keeps the relative strengths of the hands constant when you go to the closest possible worlds—on which information that Pete has the weaker hand would warrant asserting (4) but information that Pete is cheating would not warrant asserting (3).

Lowe rightly points out that in Chapter 2 I mentioned three ways you might respond to the Sly Pete argument as applied to past indicative conditionals, and that the third of these ways involves recognising a kind of ambiguity. And he argues that adopting this third way would undermine the case provided by Sly Pete for separating past indicative conditionals from past subjunctive conditionals. I will argue that this is a mistake.



-46-

The three responses to Sly Pete that I mentioned were: abandon (Conditional non-contradiction), abandon (Support) by holding that past indicative conditionals lack truth values, or abandon what I called a presupposition of (Support) by holding that the proposition expressed by a past indicative conditional is a function of context. In particular, the proposition expressed by a given past indicative conditional might be a function of the evidence that backs its assertion. Relative to any given body evidence, a past indicative conditional ‘If A then B’ expresses a proposition (is truth valued), and moreover a proposition which is incompatible with the proposition expressed by ‘If A then not B’ relative to the same evidence. But the proposition expressed by a given utterance of ‘If A then B’ varies with the evidence.

According to this last position, what is happening in the Sly Pete example is that informant Y who knows that Pete has the weaker hand and says in consequence that if Pete called, he lost expresses a (certainly true) proposition which is inconsistent with the proposition which she would have expressed had she said that if Pete called, he won. And informant X who knows that Sly Pete is cheating and says in consequence that if Pete called, he won, expresses a (certainly true) proposition which is inconsistent with the proposition which she would have expressed had she said that if Pete called, he lost. But it is not the case that the proposition expressed by one informant’s declaration that if Pete called, he lost is incompatible with that expressed by the other informant’s declaration that if Pete called, he won; and this is so precisely because the evidence that they have at their disposal is different.

Lowe’s view is that if you adopted this last response to the Sly Pete argument as applied to past indicative conditionals—and I take it that he thinks that one should—then the point that I conceded to him earlier concerning the potential ambiguity of the past subjunctive conditionals means that the Sly Pete argument does not establish a significant difference between past indicative conditionals and past subjunctive conditionals. We would be learning the same lesson about both from the Sly Pete example: be alert to ambiguity and context dependence when dealing with conditionals, be they past subjunctive or past indicative ones. However, there would still be a very significant difference between past subjunctive conditionals and past indicative conditionals. Consider informant X who knows that Sly Pete is cheating and who has no opinion one way or the other about

-47-



whose hand is the stronger. What should be X’s attitude be to ‘If Pete called, he won’? The single correct answer is, Assert strongly. What should that informant’s attitude be to ‘If Pete had called, he would have won’? There is no single answer for, as I conceded to Lowe, there is an ambiguity to be resolved (explicitly, by context, by charity principles, or whatever), and which attitude should be taken depends on how the ambiguity is resolved. If the resolution which allows back-tracking is in play, then the informant will be in a position to assert strongly that if Pete had called, he would have won. For given her knowledge that Pete is cheating, she knows that the only way it could come about that Pete calls would be in the case where Pete has the stronger hand, and in that case he would win. If, on the other hand, back-tracking is not allowed (the default position, in my view), then how it might have come to pass that Pete called is beside the point. What matters is simply how things are at the putative time of calling, and as the informant does not know which hand is the stronger, she is not, on this resolution of the ambiguity, in a position to assert that if Pete had called, he would have won.

It is one thing to acknowledge the possibility that a given sentence may express different propositions depending on context, and quite another to specify the circumstances in which the possibility is realised. If the proposition expressed by a past indicative conditional is a function of the evidence for that conditional, then past indicative conditionals are very special—the usual situation after all is that gaining or losing evidence is, or at least can be, gaining or losing evidence for the very same thing—and in particular past indicative conditionals are importantly different from past subjunctive conditionals. If the right response to the Sly Pete argument applied to past indicative conditionals is that the very act of getting evidence for (or against) them changes the proposition you are getting evidence for (or against), the Sly Pete argument by virtue of revealing this fact supports the view that past indicative and past subjunctive conditionals are importantly different.

What about future indicative conditionals? I have just highlighted as central the fact that for informant X who knows that Pete is cheating but has no opinion about whose hand is the stronger, the right thing to say in the indicative is ‘If Pete called, he won’, but that X has a choice about asserting ‘If Pete had called, he would have won’. For in the latter case, what he ought

-48-



to assert depends on how an ambiguity is resolved. Likewise, for informant Y who knows that Pete’s hand is weaker but has no opinion about whether or not Pete knows what is in his opponent’s hand, the right thing to say indicatively is that if Pete called, he lost. It was this point which showed that even if Lowe is right and it is all a matter of context dependence, the way in which it is a matter of context dependence is sharply different for past indicative conditionals and past subjunctive conditionals. I think that the same is true for future indicative conditionals, and this was my reason in Chapter 2 for holding that the Sly Pete argument gave us reason to classify future and past indicative conditionals together and apart from past subjunctive conditionals. V.H. Dudman disagrees. 2 He thinks that the situation with future indicative conditionals is like the situation I belatedly acknowledged for past subjunctive conditionals. For instance, according to Dudman, informant X who knows that Pete is cheating but has no opinion about whose hand is the stronger could well assert that if Pete calls, he will win, but she could also properly refrain from asserting it. It all depends on the context in which she looks at the matter.

We have here a clash of linguistic intuitions. For me, it is clear that someone who knows that Sly Pete is cheating while having no opinion on whether Pete’s hand is the stronger has only one way to answer the question, What will happen if Pete calls? namely, by saying that Pete will win. I take it that Dudman simply denies this intuition.

It is, of course, important to separate the question of the assertability for X of ‘If Pete calls, he will win’ from the assertability for X of ‘If Pete calls and…, Pete will win’. ‘If A then B’ can be very highly assertable when ‘If A and C, then B’ is not—a point prominent in discussions of strengthening the antecedent. And it does seem to me that when Dudman seeks to make appealing his view that X could properly assert ‘If Pete calls, he will win’ but also properly refrain from doing so (he all but grants that (a normal) Y has to assert that if Pete calls, he will lose), he is in danger of conflating these two separate questions. Consider, for example, his footnote discussion of the underlying reasoning of X. He observes correctly that ‘it makes perfect sense [for X] to say ‘Sly Pete will not call without a winning hand. If he does he will lose his shirt’. But ‘If he does he will lose his shirt’ here should be expanded to ‘If he calls without a winning hand he will lose his shirt’, not to ‘If he calls he will lose his shirt’. Later,

-49-



in the body of the text, he suggests that an X who withholds assent from ‘If Pete calls, he will win’ is moved ‘by some such reasoning’ as this: Sly Pete may for all I know hold the weaker hand. If Sly Pete calls while holding the weaker hand, he will lose. So for all I know Sly Pete may not win if he calls. But, of course, the fact that X may assert ‘If Sly Pete calls while holding the weaker hand, he will lose’ does not mean that he may properly withhold assent from ‘If Sly Pete calls, he will win’. For if Pete calls, he will hold the stronger hand. That is the crucial piece of information X has to hand.

Notes 1 E.J. Lowe, ‘Jackson on Classifying Conditionals’, Analysis, 51 (1991): 126-30. 2 V.H. Dudman, ‘Jackson Classifying Conditionals’, Analysis, 51 (1991): 131-6.

-50-

4 POSTSCRIPT ON TRUTH CONDITIONS AND ASSERTABILITY 4.1 On having truth conditions

Many wonder why I hold that indicative conditionals have truth conditions. They note that I view

(Adams) Ass(A→B)=Pr(B/A)=bydef. Pr(AB)/Pr(A), when Pr(A)≠0

very favourably, and ask, Why suppose that indicative conditionals have truth conditions in addition to the assertion conditions given by (Adams)? My response is that having truth conditions follows from having assertion conditions of the appropriate kind. You can’t have one without the other. In particular, if a hypothesis about indicative conditionals that assigns them the truth conditions of the corresponding material conditional is the best way to explain (Adams), as I argue in Chapter 1, it follows that indicative conditionals have the truth conditions of material conditionals. 1 I stand by this response, but have come to think that there is an interesting variant on it that should be catalogued. In order to explain it, I need to say something about what it is for a sentence to have truth conditions. 2



A principal use of language is to represent how things are. This is why we find a sentence like ‘There is tiger nearby’ so alarming, and why we are reassured when we hear soon after a sentence like ‘The tiger is in a very strong cage’. Thus, the conditions under which a sentence is true are the conditions under which things are as the sentence represents them, and for a sentence to have truth conditions is for it to be in the business of representing how things are.

Now, how a sentence represents things as being is an a posteriori,

-51-

contingent matter. Language represents how things are as a result of conventions implicitly entered into for the communication of information, just as Locke said in An Essay Concerning Human Understanding, book III. More particularly, what settles how sentences represent things as being are the beliefs about how things are that the words and sentences, under the contingently adopted conventions of the language, are used to express. ‘Roses are red’ might have been the sentence to use to convey the information that ‘There is a tiger nearby’ in fact conveys and, if it had been, it would have been because we had settled on conventions of word and sentence usage that imply that ‘Roses are red’ is the sentence to use when we wish to communicate our belief that there is a tiger nearby.

This means that the question as to whether ‘If P then Q’ has truth conditions is the question as to whether the way we use the sentence implies that there is a belief that the sentence serves to express. Accordingly, if there was a belief with roughly the property that the more we believe it (that is, the greater the relevant degree of belief), the more right it is to assert ‘If P then Q, then ‘If P then Q’ would express that belief, and be true iff that belief is true—that is, iff things are as that belief represents them. But (Adams) tells us there is no such belief. It tells us that the assertability of an indicative conditional is given by a quotient of degrees of belief, not by the degree to which we hold any given belief. 3 This means that the way patterns of usage determine truth conditions in the case of indicative conditionals must deviate from the paradigm way, must deviate from the way encapsulated in: S’s truth conditions are the way things are such that, roughly, the more confident users of S are that these way things are are actual, the more assertable they regard S. 4

However, how much deviance from the paradigm is consistent with having truth conditions is a judgement call. I think the ‘robustness’ explanation I give in Chapter 1 of (Adams), in terms that give indicative conditionals the truth conditions of material conditionals, is close enough to the paradigm to mean that indicative conditionals



have the truth conditions of material conditionals. I allow, though, that an alternative position well worth cataloguing holds that it is indeterminate whether or not indicative conditionals have truth conditions on the ground that it is indeterminate whether or not the deviance from the paradigm is too great to be consistent with having truth conditions.

-52-

4.2 What is assertability in (Adams)?

There is a puzzle about what it is that (Adams) identifies with the conditional probability of Q given P. What, precisely, is the right reading to give to assertability as it figures in (Adams)? It isn’t how advisable it is to assert ‘If P then Q’; that is settled by one’s best estimate of the consequences of so asserting, a matter that will often have little to do with the probability of Q given P. It isn’t how justified it is to assert ‘If P then Q’; that is settled by how advisable it is to so assert going by the consequences one ought to attend to, a matter that will often having little to do with the probability of Q given P. It isn’t how likely ‘If P then Q’ is to be true; that is what we learn from the Lewis proof and its literature.

In Chapter 1, I talk, when considering questions of propriety of assertion, of concentrating ‘on epistemic and semantic considerations widely construed’ and of putting ‘aside more particular, highly contextual ones’. This suggests that we might distinguish local factors that affect assertability from more general semantic and epistemic ones and read (Adams) as identifying the non-local assertability of ‘If P then Q’ with the conditional probability of Q given P. 5 However, if assertability means how right it is to utter or write the sentence, or anything much like that, assertability would seem to be always a matter of highly contextual factors in the sense of involving the likely effects of uttering or writing the sentence; it is, therefore, obscure how we should follow the suggested recipe for isolating the required sense of assertability.

Some argue that we should read (Adams) as identifying the acceptability of an indicative conditional with the probability of its consequent given its antecedent, where acceptability is explained as independent of the advisability of producing the sentence in public. But sometimes inner sayings to oneself can have bad effects—it is not a good idea for nervous flyers to say to themselves ‘If the landing gear malfunctions, the plane will crash’—nevertheless, the probability of consequent given antecedent is high. But if we insisted that acceptability had nothing to do with



questions of advisability of acceptance, it would then not be sufficiently clear how acceptability differed from probability of truth—as it must, if (Adams) is to be the true, interesting claim so many of us take it to be.

I now think that we should simply observe that indicative conditionals seem to have a probability of truth given by the

-53-

probability of their consequents given their antecedents—call this, their intuitive probability—and that this intuitive probability plays for indicative conditionals the role that (subjective) probability of truth typically plays elsewhere in governing assertion. In other words, intuitive probability is defined functionally: it is that property of indicative conditionals that plays the role that subjective probability of truth plays for sentences like ‘Grass is green’. (Adams) is then the thesis that the intuitive probability of ‘If P then Q’ is the probability of Q given P, or, in other words, (Adams) is the interesting (and I hold true) claim that, in the case of indicative conditionals, that which plays the role in governing assertion typically played by (subjective) probability of truth is played instead by the probability of the conditional’s consequent given its antecedent. 6

Notes 1 The details are given in Frank Jackson, Conditionals, Oxford, Basil Blackwell,

1987, pp. 56-9. 2 For how what I am about to say goes against minimalism about truth aptness—

the view that with the exception of liar sentences and the like, meaningful sentences that have the syntactic marks of truth aptness are automatically truth apt—see Frank Jackson, ‘II—Naturalism and the Fate of the M-Worlds’, Proceedings of the Aristotelian Society, supp. vol. 71 (1997):269-82.

3 And the Lewis proof referred to in Chapter 1, ‘On Assertion and Indicative Conditionals’, along with the literature sparked by it, gives us good reason to resist the suggestion that the quotient of degrees of belief might turn out to be the degree of some third belief.

4 For much less rough formulations, see David Lewis, Convention, Cambridge, Mass., Harvard University Press, 1969, and Jonathan Bennett, Linguistic Behaviour, Cambridge, Cambridge University Press, 1976. The assertability in my rough formulation should be read in the functional way discussed in §4.2.



5 Added note: in Frank Jackson, Conditionals, Oxford, Basil Blackwell, 1987, pp. 10-11, I used ‘assertibility’ (the ‘i’ spelling) to mark this sense; I now think that assertibility is best thought of as the intuitive probability I discuss below.

6 With hindsight I realise that this view was one I suggested in ‘Conditionals and Possibilia’, Proceedings of the Aristotelian Society, 81 (1980/81):126-37, but in different words.

-54-

Part II MIND

-55-


-56-

5 EPIPHENOMENAL QUALIA It is undeniable that the physical, chemical and biological sciences have provided a great deal of information about the world we live in and about ourselves. I will use the label ‘physical information’ for this kind of information, and also for information that automatically comes along with it. For example, if a medical scientist tells me enough about the processes that go on in my nervous system, and about how they relate to happenings in the world around me, to what has happened in the past and is likely to happen in the future, to what happens to other similar and dissimilar organisms and the like, he or she tells me—if I am clever enough to fit it together appropriately—about what is often called the functional role of those states in me (and in organisms in general in similar cases). This information, and its kin, I also label ‘physical’.



I do not mean these sketchy remarks to constitute a definition of ‘physical information’, and of the correlative notions of physical property, process and so on, but to indicate what I have in mind here. It is well known that there are problems with giving a precise definition of these notions, and so of the thesis of physicalism that all (correct) information is physical information. 1 But—unlike some—I take the question of definition to cut across the central problems I want to discuss in this chapter.

I am what is sometimes known as a ‘qualia freak’. I think that there are certain features of the bodily sensations especially, but also of certain perceptual experiences, which no amount of purely physical information includes. Tell me everything physical there is to tell about what is going on in a living brain, the kind of states, their functional roles, their relation to what goes on at other times and in other brains, and so on and so forth, and be I as clever as can be in fitting it all together, you won’t have told me

-57-

about the hurtfulness of pains, the itchiness of itches, pangs of jealousy, or about the characteristic experience of tasting a lemon, smelling a rose, hearing a loud noise or seeing the sky.

There are many qualia freaks, and some of them say that their rejection of physicalism is an unargued intuition. 2 I think that they are being unfair to themselves. They have the following argument. Nothing you could tell of a physical sort captures the smell of a rose, for instance. Therefore, physicalism is false. By our lights this is a perfectly good argument. It is obviously not to the point to question its validity, and the premise is intuitively obviously true both to them and to me.

I must, however, admit that it is weak from a polemical point of view. There are, unfortunately for us, many who do not find the premise intuitively obvious. The task then is to present an argument whose premises are obvious to all, or at least to as many as possible. This I try to do in §5.1 with what I will call the ‘knowledge argument’. In §5.2 I contrast the knowledge argument with the modal argument and in §5.3 with the ‘What is it like to be’ argument. In §5.4 I tackle the question of the causal role of qualia. The major factor in stopping people from admitting qualia is the belief that they would have to be given a causal role with respect to the physical world and especially the brain; 3 and it is hard to do this without sounding like someone who believes in fairies. I seek in §5.4 to turn this objection by arguing that the view that qualia are epiphenomenal is a perfectly possible one.



5.1 The knowledge argument for qualia

People vary considerably in their ability to discriminate colours. Suppose that, in an experiment to catalogue this variation, Fred is discovered. Fred has better colour vision than anyone else on record; he makes every discrimination that anyone has ever made, and moreover he makes one that we cannot even begin to make. Show him a batch of ripe tomatoes and he sorts them into two roughly equal groups and does so with complete consistency. That is, if you blindfold him, shuffle the tomatoes up, and then remove the blindfold and ask him to sort them out again, he sorts them into exactly the same two groups.

We ask Fred how he does it. He explains that all ripe tomatoes do not look the same colour to him, and in fact that this is true of a great many objects that we classify together as red. He sees two colours where we see one, and he has in consequence developed

-58-

for his own use two words, ‘red1’ and ‘red2 ’, to mark the difference. Perhaps he tells us that he has often tried to teach the difference between red1 and red2 to his friends but has got nowhere and has concluded that the rest of the world is red1 and red2 colour-blind—or perhaps he has had partial success with his children; it doesn’t matter. In any case he explains to us that it would be quite wrong to think that because ‘red’ appears in both ‘red1 ’ and ‘red2 ’ that the two colours are shades of the one colour. He only uses the common term ‘red’ to fit more easily into our restricted usage. To him, red1 and red2 are as different from each other and all the other colours as yellow is from blue. And his discriminatory behaviour bears this out: he sorts red1 from red2 tomatoes with the greatest of ease in a wide variety of viewing circumstances. Moreover, an investigation of the physiological basis of Fred’s exceptional ability reveals that Fred’s optical system is able to separate out two groups of wavelengths in the red spectrum as sharply as we are able to sort out yellow from blue. 4

I think that we should admit that Fred can see, really see, at least one more colour than we can; red1 is a different colour from red2 . We are to Fred as a totally red-green colour-blind person is to us. H.G. Wells’s story ‘The Country of the Blind’ is about a sighted person in a totally blind community. 5 This person never manages to convince them that he can see, that he has an extra sense. They ridicule this sense



as quite inconceivable, and treat his capacity to avoid falling into ditches, to win fights and so on as precisely that capacity and nothing more. We would be making their mistake if we refused to allow that Fred can see one more colour than we can.

What kind of experience does Fred have when he sees red1 and red2 ? What is the new colour or colours like? We would dearly like to know but do not; and it seems that no amount of physical information about Fred’s brain and optical system tells us. We find out, perhaps, that Fred’s cones respond differentially to certain light waves in the red section of the spectrum that make no difference to ours (or perhaps he has an extra cone) and that this leads in Fred to a wider range of those brain states responsible for visual discriminatory behaviour. But none of this tells us what we really want to know about his colour experience. There is something about it we don’t know. But we know, we may suppose, everything about Fred’s body, his behaviour and dispositions to behaviour and about his internal physiology, and

-59-

everything about his history and relation to others that can be given in physical accounts of persons. We have all the physical information. Therefore, knowing all this is not knowing everything about Fred. It follows that physicalism leaves something out.

To reinforce this conclusion, imagine that as a result of our investigations into the internal workings of Fred, we find out how to make everyone’s physiology like Fred’s in the relevant respects; or perhaps Fred donates his body to science and on his death we are able to transplant his optical system into someone else—again the fine detail doesn’t matter. The important point is that such a happening would create enormous interest. People would say, ‘At last we will know what it is like to see the extra colour, at last we will know how Fred has differed from us in the way he has struggled to tell us about’. But then it cannot be that we knew all along all about Fred. But ex hypothesi we did know all along everything about Fred that features in the physicalist scheme; hence the physicalist scheme leaves something out.

Put it this way. After the operation, we will know more about Fred and especially about his colour experiences. But beforehand we had all the physical information we could desire about his body and brain, and indeed everything that has ever featured in physicalist accounts of mind and consciousness. Hence there is more to know than all that. Hence physicalism is incomplete. Fred and the new colour(s) are of course essentially rhetorical devices. The same point can be made with normal people and familiar colours. Mary is a brilliant scientist who is, for whatever reason, forced to



investigate the world from a black and white room via a black and white television monitor. She specialises in the neurophysiology of vision and acquires, let us suppose, all the physical information there is to obtain about what goes on when we see ripe tomatoes, or the sky, and use terms like ‘red’, ‘blue’, and so on. She discovers, for example, just which wavelength combinations from the sky stimulate the retina, and exactly how this produces via the central nervous system the contraction of the vocal chords and expulsion of air from the lungs that results in the uttering of the sentence ‘The sky is blue’. (It can hardly be denied that it is in principle possible to obtain all this physical information from black and white television, otherwise the Open University would of necessity need to use colour television.)

What will happen when Mary is released from her black and white room or is given a colour television monitor? Will she

-60-

learn anything or not? It seems just obvious that she will learn something about the world and our visual experience of it. But then it is inescapable that her previous knowledge was incomplete. But she had all the physical information. Ergo, there is more to have than that, and physicalism is false.

Clearly the same style of knowledge argument could be deployed for taste, hearing, the bodily sensations and, generally speaking, for the various mental states which are said to have (as it is variously put) raw feels, phenomenal features or qualia. The conclusion in each case is that the qualia are left out of the physicalist story. And the polemical strength of the knowledge argument is that it is so hard to deny the central claim that one can have all the physical information without having all the information there is to have.

5.2 The modal argument

By the modal argument I mean an argument of the following style. 6 Sceptics about other minds are not making a mistake in deductive logic, whatever else may be wrong with their position. No amount of physical information about another logically entails that he or she is conscious or feels anything at all. Consequently there is a possible world with organisms exactly like us in every physical respect (and remember that includes functional states, physical history, etc.) but which differ from us profoundly in that they have no conscious mental life at all. But then what is it that we have and they lack? Not anything physical ex hypothesi. In all physical regards



we and they are exactly alike. Consequently there is more to us than the purely physical. Thus physicalism is false. 7 It is sometimes objected that the modal argument misconceives physicalism on the ground that that doctrine is advanced as a contingent truth. 8 But to say this is only to say that physicalists restrict their claim to some possible worlds, including especially ours; and the modal argument is only directed against this lesser claim. If we in our world, let alone beings in any others, have features additional to those of our physical replicas in other possible worlds, then we have non-physical features or qualia. The trouble rather with the modal argument is that it rests on a disputable modal intuition: disputable because it is disputed. Some sincerely deny that there can be physical replicas of us in other possible worlds which, nevertheless, lack consciousness.

-61-

Moreover, at least one person who once had the intuition now has doubts. 9 Head-counting may seem a poor approach to a discussion of the modal argument. But frequently we can do no better when modal intuitions are in question, and remember our initial goal was to find the argument with the greatest polemical utility.

Of course, qua protagonists of the knowledge argument we may well accept the modal intuition in question; but this will be a consequence of our already having an argument to the conclusion that qualia are left out of the physicalist story, not our ground for that conclusion. Moreover, the matter is complicated by the possibility that the connection between matters physical and qualia is like that sometimes held to obtain between aesthetic qualities and natural ones. Two possible worlds which agree in all ‘natural’ respects (including the experiences of sentient creatures) must agree in all aesthetic qualities also, but it is plausibly held that the aesthetic qualities cannot be reduced to the natural. 10

5.3 The ‘what is it like to be’ argument

Thomas Nagel argues that no amount of physical information can tell us what it is like to be a bat, and indeed that we, human beings, cannot imagine what it is like to be a bat. 11 His reason is that what this is like can only be understood from a bat’s point of view, which is not our point of view and is not something capturable in physical terms, which are essentially terms understandable equally from many points of view.

It is important to distinguish this argument from the knowledge argument. When I complained that all the physical knowledge about Fred was not enough to tell us what



his special colour experience was like, I was not complaining that we weren’t finding out what it is like to be Fred. I was complaining that there is something about his experience, a property of it, of which we were left ignorant. And if and when we come to know what this property is, we still will not know what it is like to be Fred, but we will know more about him. No amount of knowledge about Fred, be it physical or not, amounts to knowledge ‘from the inside’ concerning Fred. We are not Fred. There is thus a whole set of items of knowledge expressed by forms of words like ‘that it is / myself who is…’ which Fred has and we simply cannot have because we are not him. 12

When Fred sees the colour he alone can see, one thing he knows is the way his experience of it differs from his experience

-62-

of seeing red and so on; another is that he himself is seeing it. Physicalist and qualia freaks alike should acknowledge that no amount of information of whatever kind that others have about Fred amounts to knowledge of the second. My complaint, though, concerned the first and was that the special quality of his experience is certainly a fact about it, and one which physicalism leaves out because no amount of physical information told us what it is.

Nagel speaks as if the problem he is raising is one of extra-polating from knowledge of one experience to another, of imagining what an unfamiliar experience would be like on the basis of familiar ones. In terms of Hume’s example, from knowledge of some shades of blue, we can work out what it would be like to see other shades of blue. Nagel argues that the trouble with bats etc., is that they are too unlike us. It is hard to see an objection to physicalism here. Physicalism makes no special claims about the imaginative or extrapolative powers of human beings, and it is hard to see why it need do so. 13

Anyway, our knowledge argument makes no assumptions on this point. If physicalism were true, enough physical information about Fred would obviate any need to extrapolate or to perform special feats of imagination or understanding in order to know all about his special colour experience. The information would already be in our possession. But it clearly isn’t. That was the nub of the argument.

5.4 The bogey of epiphenomenalism 14



Is there any really good reason for refusing to countenance the idea that qualia are causally impotent with respect to the physical world? I will argue for the answer no, but in doing this I will say nothing about two views associated with the classical epiphenomenalist position. The first is that mental states are inefficacious with respect to the physical world. All I will be concerned to defend is that it is possible to hold that certain properties of certain mental states, namely those I’ve called qualia, are such that their possession or absence makes no difference to the physical world. The second is that the mental is totally causally inefficacious. For all I will say, it may be that you have to hold that the instantiation of qualia makes a difference to other mental states though not to anything physical. Indeed, general considerations to do with how you could come to be aware of the instantiation of qualia suggest such a position. 15

-63-

Three reasons are standardly given for holding that a quale, like the hurtfulness of a pain, must be causally efficacious in the physical world, and so, for instance, that its instantiation must sometimes make a difference to what happens in the brain. None, I will argue, has any real force. (I am much indebted to Alec Hyslop and John Lucas for convincing me of this.)

(i) It is supposed to be just obvious that the hurtfulness of pain is partly responsible for the subject seeking to avoid pain, saying ‘It hurts’, and so on. But, to reverse Hume, anything can fail to cause anything. No matter how often B follows A, and no matter how initially obvious the causality of the connection seems, the hypothesis that A causes B can be overturned by an overarching theory which shows the two as distinct effects of a common underlying causal process.

To the untutored, the image on the screen of Lee Marvin’s fist moving from left to right immediately followed by the image of John Wayne’ s head moving in the same general direction looks as causal as anything. 16 And of course throughout countless Westerns images similar to the first are followed by images similar to the second. All this counts for precisely nothing when we know the overarching theory concerning how the relevant images are both effects of an underlying causal process involving the projector and the film. The epiphenomenalist can say exactly the same about the connection between, for example, hurtfulness and behaviour. It is simply a consequence of the fact that certain happenings in the brain cause both.

(ii) The second objection relates to Darwin’s theory of evolution. According to natural selection, the traits that evolve over time are typically those conducive to physical



survival. We may assume that qualia evolved over time—we have them, the earliest forms of life do not—and so we should expect qualia to be conducive to survival. The objection is that they could hardly help us to survive if they do nothing to the physical world.

The appeal of this argument is undeniable, but there is a good reply to it. Polar bears have particularly thick, warm coats. The theory of evolution explains this (we suppose) by pointing out that having a thick, warm coat is conducive to survival in the Arctic. But having a thick coat goes along with having a heavy coat, and having a heavy coat is not conducive to survival. It slows the animal down.

-64-

Does this mean that we have refuted Darwin because we have found an evolved trait—having a heavy coat—which is not conducive to survival? Clearly not. Having a heavy coat is an unavoidable concomitant of having a warm coat (in the context, modern insulation was not available), and the advantages for survival of having a warm coat outweighed the disadvantages of having a heavy one. The point is that all we can extract from Darwin’s theory is that we should expect any evolved characteristic to be either conducive to survival or a by-product of one that is so conducive. The epiphenomenalist holds that qualia fall into the latter category. They are a by-product of certain brain processes that are highly conducive to survival. 17

(iii) The third objection is based on a point about how we come to know about other minds. We know about other minds by knowing about other behaviour, at least in part. The nature of the inference is a matter of some controversy, but it is not a matter of controversy that it proceeds from behaviour. 18 That is why we think that stones do not feel and dogs do feel. But, runs the objection, how can a person’s behaviour provide any reason for believing he or she has qualia like mine, or indeed any qualia at all, unless this behaviour can be regarded as the outcome of the qualia. Man Friday’s footprint was evidence of Man Friday because footprints are causal outcomes of feet attached to people. And an epiphenomenalist cannot regard behaviour, or indeed anything physical, as an outcome of qualia.

But consider my reading in The Times that Spurs won. This provides excellent evidence that the Telegraph has also reported that Spurs won, despite the fact that (I trust) the Telegraph does not get the results from The Times. They each send their own reporters to the game. The Telegraph’s report is in no sense an outcome of The Times’s, but the latter provides good evidence for the former nevertheless.



The reasoning involved can be reconstructed thus. I read in The Times that Spurs won. This gives me reason to think that Spurs won because I know that Spurs’ winning is the most likely candidate to be what caused the report in The Times. But I also know that Spurs’ winning would have had many effects, including almost certainly a report in the Telegraph.

I am arguing from one effect back to its cause and out again to another effect. The fact that neither effect causes the other is irrelevant. Now the epiphenomenalist allows that qualia are

-65-

effects of what goes on in the brain. Qualia cause nothing physical but are caused by something physical. Hence the epiphenomenalist can argue from the behaviour of others to the qualia of others by arguing from the behaviour of others back to its causes in the brains of others and out again to their qualia.

You may well feel for one reason or another that this is a more dubious chain of reasoning than its model in the case of newspaper reports. You are right. The problem of other minds is a major philosophical problem, the problem of other newspaper reports is not. But there is no special problem of epiphenomenalism as opposed to, say, Interactionism here.

There is a very understandable response to the three replies I have just made. ‘All right, there is no knockdown refutation of the existence of epiphenomenal qualia. But the fact remains that they are an excrescence. They do nothing, they explain nothing, they serve merely to soothe the intuitions of dualists, and it is left a total mystery how they fit into the world view of science. In short we do not and cannot understand the how and why of them.’

This is perfectly true; but is no objection to qualia, for it rests on an overly optimistic view of the human animal, and its powers. We are the products of evolution. We understand and sense what we need to understand and sense in order to survive. Epiphenomenal qualia are totally irrelevant to survival. At no stage of our evolution did natural selection favour those who could make sense of how they are caused and the laws governing them, or in fact why they exist at all. And that is why we can’t.

It is not sufficiently appreciated that physicalism is an extremely optimistic view of our powers. If it is true, we have, in very broad outline admittedly, a grasp of our place in the scheme of things. Certain matters of sheer complexity defeat us—there are an



awful lot of neurons—but in principle we have it all. But consider the antecedent probability that everything in the Universe be of a kind that is relevant in some way or other to the survival of homo sapiens. It is very low surely. But then one must admit that it is very likely that there is a part of the whole scheme of things, maybe a big part, which no amount of evolution will ever bring us near to knowledge about or understanding. For the simple reason that such knowledge and understanding is irrelevant to survival.

Physicalists typically emphasise that we are a part of nature on their view, which is fair enough. But if we are a part of nature, we are as nature has left us after however many years of evolution it

-66-

is, and each step in that evolutionary progression has been a matter of chance constrained just by the need to preserve or increase survival value. The wonder is that we understand as much as we do, and there is no wonder that there should be matters which fall quite outside our comprehension. Perhaps exactly how epiphenomenal qualia fit into the scheme of things is one such.

This may seem an unduly pessimistic view of our capacity to articulate a truly comprehensive picture of our world and our place in it. But suppose we discovered living on the bottom of the deepest oceans a sort of sea slug which manifested intelligence. Perhaps survival in the conditions required rational powers. Despite their intelligence, these sea slugs have only a very restricted conception of the world in comparison with ours, the explanation for this being the nature of their immediate environment. Nevertheless they have developed sciences which work surprisingly well in these restricted terms. They also have philosophers, called slugists. Some call themselves tough-minded slugists, others confess to being soft-minded slugists.

The tough-minded slugists hold that the restricted terms (or ones pretty like them which may be introduced as their sciences progress) suffice in principle to describe everything without remainder. These tough-minded slugists admit in moments of weakness to a feeling that their theory leaves something out. They resist this feeling and their opponents, the soft-minded slugists, by pointing out—absolutely correctly—that no slugist has ever succeeded in spelling out how this mysterious residue fits into the highly successful view that their sciences have and are developing of how their world works. Our sea slugs don’t exist, but they might. And there might also exist super beings which stand to us as we stand to the sea slugs. We cannot adopt the perspective of these super beings, because we are not them, but the possibility of



such a perspective is, I think, an antidote to excessive optimism. 19

Notes 1 See, e.g., D.H. Mellor, ‘Materialism and Phenomenal Qualities’, Proceedings of

the Aristotelian Society, supp. vol. 47 (1973):107-19, and J.W. Cornman, Materialism and Sensations, New Haven, Yale University Press, 1971. Added note: I discuss some of the ways we might specify what it is to be physical in Chapter 1 of From

-67-

Metaphysics to Ethics: A Defence of Conceptual Analysis, Oxford, Oxford University Press, 1998.

2 Particularly in discussion, but see, e.g., Keith Campbell, Metaphysics, Encino, Calif., Dickenson, 1976, p. 67.

3 See, e.g., Daniel Dennett, ‘Current Issues in the Philosophy of Mind’, American Philosophical Quarterly, 15 (1978):249-61.

4 Put this, and similar simplifications below, in terms of Land’s theory if you prefer. See, e.g., Edwin H. Land, ‘Experiments in Color Vision’, Scientific American, 200 (5 May 1959):84-99.

5 H.G. Wells, The Country of the Blind and Other Stories, London, n.d. 6 See, e.g., Keith Campbell, Body and Mind, London, Macmillan, 1971, and Robert

Kirk, ‘Sentience and Behaviour’, Mind, 83 (1974):43-60. 7 I have presented the argument in an inter-world rather than the more usual intra-

world fashion to avoid inessential complications to do with supervenience, causal anomalies and the like.

8 See, e.g., W.G. Lycan, ‘A New Lilliputian Argument against Machine Functionalism’, Philosophical Studies, 35 (1979):279-87, p. 280, and Don Locke, ‘Zombies, Schizophrenics and Purely Physical Objects’, Mind, 85 (1976):97-9. Added note: the point that follows in the text is addressed in greater detail in §11.2 of Chapter 11, ‘Armchair metaphysics’.

9 See Robert Kirk, ‘From Physical Explicability to Full-Blooded Materialism’, Philosophical Quarterly, 29 (1979):229-37. See also the arguments against the modal intuition in, e.g., Sydney Shoemaker, ‘Functionalism and Qualia’, Philosophical Studies, 27 (1975):291-315.



10 Added note: I now think that aesthetic qualities can be reduced to natural ones. In a critical notice of Susan Hurley, Natural Reasons, Australasian Journal of Philosophy, 70 (1992):475-87,1 sketch how to reduce ethical qualities to natural or descriptive ones and would argue that the same basic strategy applies to aesthetic qualities; see also the last two chapters of From Metaphysics to Ethics.

11 Thomas Nagel, ‘What is it Like to be a Bat?’, Philosophical Review, 83 (1974):436-50. Two things need to be said about this article. One is that, despite my dissociations to come, I am much indebted to it. The other is that the emphasis changes through the article, and by the end Nagel is objecting not so much to physicalism as to all extant theories of mind for ignoring points of view, including those that admit (irreducible) qualia.

12 Knowledge de se in the terms of David Lewis, ‘Attitudes De Dicto and De Se’ Philosophical Review, 88 (1979):513-43.

13 See Laurence Nemirow’s comments on ‘What is it Like to be a Bat?’ in his review of Thomas Nagel, Mortal Questions, Philosophical Review, 89 (1980):473-7. I am indebted here in particular to a discussion with David Lewis.

14 Added note: I now think this section is an ingenious but flawed attempt to try and live with the unlivable; however, I stand by the

-68-

remarks about the optimistic nature of physicalism—see Chapter 7, ‘Postscript on Qualia’.

15 See my review of Keith Campbell, Body and Mind, in the Australasian Journal of Philosophy, 50 (1972):77-80.

16 Cf. Jean Piaget, ‘The Child’s Conception of Physical Causality’, reprinted in Howard E. Gruber and J. Jacques Vonedue, eds, The Essential Piaget, London, Routledge & Kegan Paul, 1977.

17 Added note: the reading of the theory of evolution in play in the preceding three paragraphs now seems to me unduly adaptationist, but not in a way that affects the key point.

18 Added note: but see the comments in §8.2 of Chapter 8 ‘Mental causation’. 19 I am indebted to Robert Pargetter for a number of comments and, despite his

dissent, to §IV of Paul E. Meehl, ‘The Compleat Autocerebroscopist’ in Paul Feyerabend and Grover Maxwell, eds, Mind, Matter, and Method, Minneapolis, University of Minnesota Press, 1966.

-69-



6 WHAT MARY DIDN’T KNOW Mary is confined to a black-and-white room, is educated through black-and-white books, and through lectures relayed on black-and-white television. In this way she learns everything there is to know about the physical nature of the world. She knows all the physical facts about us and our environment, in a wide sense of ‘physical’ which includes everything in completed physics, chemistry, and neurophysiology, and all there is to know about the causal and relational facts consequent upon all this, including of course functional roles. If physicalism is true, she knows all there is to know. For to suppose otherwise is to suppose that there is more to know than every physical fact, and that is just what physicalism denies.

Physicalism is not the noncontroversial thesis that the actual world is largely physical, but the challenging thesis that it is entirely physical. This is why physicalists must hold that complete physical knowledge is complete knowledge simpliciter. For suppose it is not complete: then our world must differ from a world, W(P), for which it is complete, and the difference must be in nonphysical facts; for our world and W(P) agree in all matters physical. Hence, physicalism would be false at our world (though contingently so, for it would be true at W(P)). 1

It seems, however, that Mary does not know all there is to know. For when she is let out of the black-and-white room or given a colour television, she will learn what it is like to see something red, say. This is rightly described as learning—she will not say ‘Ho, hum’. Hence, physicalism is false. This is the knowledge argument against physicalism in one of its manifestations. 2 This note is a reply to three objections to it mounted by Paul M. Churchland. 3

-70-



6.1 Three clarifications

The knowledge argument does not rest on the dubious claim that logically you cannot imagine what sensing red is like unless you have sensed red. Powers of imagination are not to the point. The contention about Mary is not that, despite her fantastic grasp of neurophysiology and everything else physical, she could not imagine what it is like to sense red; it is that, as a matter of fact, she would not know. But if physicalism is true, she would know; and no great powers of imagination would be called for. Imagination is a faculty that those who lack knowledge need to fall back on.

Second, the intensionality of knowledge is not to the point. The argument does not rest on assuming (falsely) that, if S knows that a is F, and if a=b, then S knows that b is F. It is concerned with the nature of Mary’s total body of knowledge before she is released: is it complete, or do some facts escape it? What is to the point is that S may know that a is F and know that a=b, yet arguably not know that b is F, by virtue of not being sufficiently logically alert to follow the consequences through. If Mary’s lack of knowledge were at all like this, there would be no threat to physicalism in it. But it is very hard to believe that her lack of knowledge could be remedied merely by her explicitly following through enough logical consequences of her vast physical knowledge. Endowing her with great logical acumen and persistence is not in itself enough to fill in the gaps in her knowledge. On being let out, she will not say, ‘I could have worked all this out before by making some more purely logical inferences.’

Third, the knowledge Mary lacked which is of particular point for the knowledge argument against physicalism is knowledge about the experiences of others, not about her own. When she is let out, she has new experiences, colour experiences she has never had before. It is not, therefore, an objection to physicalism that she learns something on being let out. Before she was let out, she could not have known facts about her experience of red, for there were no such facts to know—that physicalist and non-physicalist alike can agree on. After she is let out, things change; and physicalists can happily admit that she learns this; after all, some physical things will change, her brain states and their functional roles, for instance. The trouble for physicalism is that, after Mary sees her first ripe tomato, she will realise how impoverished her conception of the mental life of others has been all along. She will realise that there was, all the time she was

-71-



carrying out her laborious investigations into the neurophysiologies of others and into the functional roles of their internal states, something about these people she was quite unaware of. All along their experiences (or many of them, those got from tomatoes, the sky,…) had a feature conspicuous to them but until now hidden from her (in fact, not in logic). But she knew all the physical facts about them all along; hence, what she did not know until her release is not a physical fact about their experiences. But it is a fact about them. That is the trouble for physicalism. 6.2 Churchland’s three objections (i) Churchland’s first objection is that the knowledge argument contains a defect that ‘is simplicity itself (p. 23). The argument equivocates on the sense of ‘knows about’. How so? Churchland suggests that the following is ‘a conveniently tightened version’ of the knowledge argument:

(1) Mary knows everything there is to know about brain states and their properties. (2) It is not the case that Mary knows everything there is to know about sensations and their properties.

Therefore, by Leibniz’s law, (3) Sensations and their properties ≠ brain states and their properties (p. 23) .

Churchland observes, plausibly enough, that the type or kind of knowledge involved in premise 1 is distinct from the kind of knowledge involved in premise 2. We might follow his lead and tag the first ‘knowledge by description’ and the second ‘knowledge by acquaintance’; but, whatever the tags, he is right that the displayed argument involves a highly dubious use of Leibniz’s law.

My reply is that the displayed argument may be convenient, but it is not accurate. It is not the knowledge argument. Take, for instance, premise 1. The whole thrust of the knowledge argument is that Mary (before her release) does not know everything there is to know about brain states and their properties, because she does not know about certain qualia associated with them. What is

-72-



complete, according to the argument, is her knowledge of matters physical. A convenient and accurate way of displaying the argument is:

(1)* Mary (before her release) knows everything physical there is to know about other people. (2)* Mary (before her release) does not know everything there is to know about other people (because she learns something about them on her release).

Therefore, (3)* There are truths about other people (and herself) which escape the physicalist story.

What is immediately to the point is not the kind, manner, or type of knowledge Mary has, but what she knows. What she knows beforehand is ex hypothesi everything physical there is to know, but is it everything there is to know? That is the crucial question. There is, though, a relevant challenge involving questions about kinds of knowledge. It concerns the support for premise 2*. The case for premise 2* is that Mary learns something on her release, she acquires knowledge, and that entails that her knowledge beforehand (what she knew, never mind whether by description, acquaintance, or whatever) was incomplete. The challenge, mounted by David Lewis and Laurence Nemirow, is that on her release Mary does not learn something or acquire knowledge in the relevant sense. What Mary acquires when she is released is a certain representational or imaginative ability; it is knowledge how rather than knowledge that. Hence, a physicalist can admit that Mary acquires something very significant of a knowledge kind—which can hardly be denied—without admitting that this shows that her earlier factual knowledge is defective. She knew all that there was to know about the experiences of others beforehand but lacked an ability until after her release. 4 Now it is certainly true that Mary will acquire abilities of various kinds after her release. She will, for instance, be able to imagine what seeing red is like, be able to remember what it is like, and be able to understand why her friends regarded her as so deprived (something which, until her release, had always mystified her). But is it plausible that that is all she will acquire?

-73-



Suppose she received a lecture on scepticism about other minds while she was incarcerated. On her release she sees a ripe tomato in normal conditions and so has a sensation of red. Her first reaction is to say that she now knows more about the kind of experiences others have when looking at ripe tomatoes. She then remembers the lecture and starts to worry. Does she really know more about what their experiences are like, or is she indulging in a wild generalisation from one case? In the end she decides she does know, and that scepticism is mistaken (even if, like so many of us, she is not sure how to demonstrate its errors). What was she to-ing and fro-ing about—her abilities? Surely not; her representational abilities were a known constant throughout. What else then was she agonising about than whether or not she had gained factual knowledge of others? There would be nothing to agonise about if ability was all she acquired on her release.

I grant that I have no proof that Mary acquires on her release, as well as abilities, factual knowledge about the experiences of others—and not just because I have no disproof of scepticism. My claim is that the knowledge argument is a valid argument from highly plausible, though admittedly not demonstrable, premises to the conclusion that physicalism is false. And that, after all, is about as good an objection as one could expect in this area of philosophy.

(ii) Churchland’s second objection (pp. 24-5) is that there must be something wrong with the argument, for it proves too much. Suppose Mary received a special series of lectures over her black-and-white television from a full-blown dualist, explaining the ‘laws’ governing the behaviour of ‘ectoplasm’ and telling her about qualia. This would not affect the plausibility of the claim that on her release she learns something. So if the argument works against physicalism, it works against dualism too.

My reply is that lectures about qualia over black-and-white television do not tell Mary all there is to know about qualia. They may tell her some things about qualia, for instance, that they do not appear in the physicalist’s story, and that the quale we use ‘yellow’ for is nearly as different from the one we use ‘blue’ for as is white from black. But why should it be supposed that they tell her everything about qualia? On the other hand, it is plausible that lectures over black-and-white television might in principle tell Mary everything in the physicalist’s story. You do not need colour television to learn physics or functionalist

-74-



psychology. To obtain a good argument against dualism (attribute dualism; ectoplasm is a bit of fun), the premise in the knowledge argument that Mary has the full story according to physicalism before her release, has to be replaced by a premise that she has the full story according to dualism. The former is plausible; the latter is not. Hence, there is no ‘parity of reasons’ trouble for dualists who use the knowledge argument.

(iii) Churchland’s third objection is that the knowledge argument claims ‘that Mary could not even imagine what the relevant experience would be like, despite her exhaustive neuroscientific knowledge, and hence must still be missing certain crucial information’ (p. 25), a claim he goes on to argue against.

But, as we emphasised earlier, the knowledge argument claims that Mary would not know what the relevant experience is like. What she could imagine is another matter. If her knowledge is defective, despite being all there is to know according to physicalism, then physicalism is false, whatever her powers of imagination. 5

Notes 1 The claim here is not that, if physicalism is true, only what is expressed in

explicitly physical language is an item of knowledge. It is that, if physicalism is true, then if you know everything expressed or expressible in explicitly physical language, you know everything. Pace Terence Morgan, ‘Jackson on Physical Information and Qualia’, Philosophical Quarterly, 34 (1984):147-52. Added note: for a fuller discussion of what physicalism implies concerning other possible worlds, see Chapter 11, ‘Armchair metaphysics’.

2 Namely, that in Chapter 5, ‘Epiphenomenal Qualia’. See also Thomas Nagel, ‘What is it Like to be a Bat?’, Philosophical Review, 83 (1974):43-50, and Howard Robinson, Matter and Sense, New York, Cambridge University Press, 1982.

3 ‘Reduction, Qualia, and the Direct Introspection of Brain States’, Journal of Philosophy, 83 (1985):8-28. Unless otherwise stated, future page references are to this paper.

4 See Laurence Nemirow, review of Thomas Nagel, Mortal Questions, in the Philosophical Review, 89 (1980):473-7 and David Lewis, ‘Postscript to “Mad Pain and Martian Pain”’, Philosophical Papers, vol. 1, New York, Oxford University Press, 1983. Churchland mentions both Nemirow and Lewis, and it may be that he intended his objection to be essentially the one I have just given. However, he says quite explicitly (bottom of p. 23) that his objection does not need an ‘ability’ analysis of the relevant knowledge.

5 I am much indebted to discussions with David Lewis and with Robert Pargetter.

-75-



7 POSTSCRIPT ON QUALIA Suppose I undertook to tell you everything there is to know about where some object is at each and every time. Could you in principle deduce all there is to know about that object’s motion? If the at-at theory of motion is true, the answer is yes. But the at-at theory of motion is an empirical claim about the nature of motion in our world. It is not properly a claim about the concept of motion. I argue for this (and for a similar position for a number of other examples) in detail in Chapter 10, ‘Metaphysics by possible cases’, but we can put the key point very quickly. The precise nature of what we are talking about when we talk about motion is unclear. We are in a kind of ‘best candidate situation’, to borrow a term from discussions of personal identity. If there is an intrinsic property of objects that genuinely explains, and so is distinct from, their being at different places at different times, and which plays all the roles we centrally associate with motion, then that property is motion; but if there is no such property, then an object’s being at different places at different times is all that that object’s motion comes to.

This means that whether or not we can deduce all there is to know about motion—our motion, motion as it is in our world—from all there is to know about positions at times depends on an empirical fact about what our world is like. The deduction is possible just if the at-at theory’s candidate for motion is the best candidate for motion in our world. I now think that the same is true for qualia and, more generally, the sensory side of psychology. In some worlds, its nature cannot be deduced in principle from the full account of the physical nature of that world, but in other worlds, including ours, it can. The redness of our reds can be deduced in principle from enough about the physical nature of our world despite the manifest appearance to

-76-



the contrary that the knowledge argument trades on. This is why I now think that the knowledge argument fails.

Why do I think that the sensory side of psychology, as it is constituted in our world, is deducible in principle from enough about the world’s physical nature? Our knowledge of the sensory side of psychology has a causal source. Seeing red and feeling pain impact on us, leaving a memory trace which sustains our knowledge of what it is like to see red and feel pain on the many occasions where we are neither seeing red nor feeling pain. This is why it was always a mistake to say that someone could not know what seeing red and feeling pain is like unless they had actually experienced them: false ‘memory’ traces are enough. This places a constraint on our best opinion about the nature of our sensory states: we had better not have opinions about their nature which cannot be justified by what we know about the causal origin of those opinions. Now the precise connection between causal origin and rational opinion is complex, but for present purposes the following rough maxim will serve: do not have opinions that outrun what is required by the best theory of these opinions’ causal origins. 1 Often it will be uncertain what the best theory is, or the question of what it is will be too close to the question under discussion for the maxim to be of much use. But in the case of sensory states, the maxim has obvious bite. We know that our knowledge of what it is like to see red and feel pain has purely physical causes. We know, for example, that Mary’s transition from not knowing what it is like to see red to knowing what it is like to see red will have a causal explanation in purely physical terms. (Dualist interactionism is false.) It follows, by the maxim, that what she learns had better not outrun how things are physically.

Towards the end of Chapter 5, ‘Epiphenomenal qualia’, I point out that a report in one newspaper may be good evidence for a similar report in another newspaper without its being the case that one report causes the other. This is true but, I now think, does not blunt the force of the argument just rehearsed. As noted in Chapter 5, the reason we are entitled to hold that the reports are similar depends on our knowing inter alia that they have a common cause, namely, the event being reported on. But we know this only because of the way reports in newspapers in general impact on us. The fundamental point remains that our entitlement comes back to causal impacts of the right kinds.

I now think that the puzzle posed by the knowledge argument is

-77-



to explain why we have such a strong intuition that Mary learns something about how things are that outruns what can deduced from the physical account of how things are. I suggest that the answer is the strikingly atypical nature of the way she acquires certain relational and functional information. Suppose that you want to know on landing in Chicago if the weather is typical for this time of year. A good deal of collecting and bringing together of information is required. The same goes for information about functional roles. To know that a certain way of driving is dangerous, or that a certain drug slows the progression of AIDS, requires bringing together information from disparate sources. However, the most plausible approach for physicalists to sensory experience sees it as a striking exception to the rule that acquiring this kind of information requires collation. The most plausible view for physicalists is that sensory experience is putative information about certain highly relational and functional properties of goings on inside us. As it is often put nowadays, its very nature is representational: it represents inter alia certain highly relational and functional facts about what is happening to us. If this is right—and I have nothing to add to the detailed arguments by those physicalists who came to the position decades ahead of me—sensory experience is a quite unusually ‘quick and easy’ way of acquiring highly relational and functional information. (And evolutionary considerations tell us why we might have acquired this ability to access quickly and easily certain sorts of highly relational and functional information.) Sensory experience is in this regard like the way we acquire information about intrinsic properties—typically, we get the information that something is round more quickly and easily than the information that it is the second largest object in the room. In consequence, sensory experience presents itself to us as if it were the acquisition of information about intrinsic nature. But, very obviously, it is not information about intrinsic physical nature, so the information Mary acquires presents itself to us as if it were information about something more than the physical. This is, I now think, the source of the strong but mistaken intuition that Mary learns something new about how things are on her release.

I still think though that we should take seriously the possibility that we know little about the intrinsic nature of our world, that we mostly know its causal cum relational nature as revealed by the physical sciences. I hope and believe (on Occamist grounds) that this kind of ‘Kantian’ scepticism is mistaken, but I think that the

-78-



reflections at the end of Chapter 5 have to be taken seriously. But even if a large part of the intrinsic nature of our world is beyond our epistemic reach, the nature we know about supervenes on the mostly functional cum relational nature that the physical sciences tell us about. The considerations at the end of Chapter 5 can be no reason to hold that Mary learns something new about how things are on her release, but rather that there may (may) be a lot about fundamental nature that we and she can never know.

Note 1 For something less rough, see Frank Jackson and Robert Pargetter, ‘Causal

Origin and Evidence’, Theoria, 51(1985):65-76.

-79-

8 MENTAL CAUSATION I 8.1 Introduction

Hypotheses about what people believe and desire greatly facilitate the prediction of their behaviour. That is common ground. And that is where the common ground ends, I fear. Some hold that the common ground fact about prediction exhausts what little truth there is in the view that what we believe and desire explains behaviour; some hold that what we believe and desire explains behaviour, but without doing so causally; and some, the majority, hold that what we believe and desire causally explains behaviour, while differing markedly among themselves over the role of mental properties, and especially the role of the contents of belief and desire, in the causal explanation of behaviour. I will restrict myself to the range of opinions among the majority. I will presume a basically realist position that insists that what is believed and desired causally explains behaviour and I will focus on the issues that arise under that presumption. 1

The differences among the majority have many sources, but an important one is a fundamental disagreement over the significance for the debate over mental causation of the causal explanations of behaviour offered by the physical sciences. I start with this issue, and then, armed with a certain view on it, return to the debate over the place of what is believed and desired in the causation of behaviour. The debate over mental causation is not, of course, merely over the causal role of belief and desire



with respect to behaviour; other mental states make their entries and exits, as do connections between mental states. But in the interests of keeping a huge topic manageable, I will be focusing in the main on the connection between behaviour and what is believed and desired.

-80-

8.2 Transparency versus strong autonomy 8.2.1 Preamble

The physical sciences—by which I do not mean physics alone but rather the physical sciences in general, including especially neuroscience—have a story to tell about the causation of behaviour. It is, as of now, radically incomplete and yields only relatively crude predictions of behaviour and rough generalisations. Unlike belief—desire psychology, the physical sciences cannot predict such facts as that a certain body, after interacting with a telephone in the distinctive fashion that results in the production of some such sentence as ‘Yes, I would love to come’, will end up in a relatively distant place and consume food and drink for some hours before returning to somewhere near the telephone. Nevertheless, it is widely believed that the physical sciences, or rather some natural extension of them, can in principle give a complete explanation for each and every bodily movement, or at least can do so up to whatever completeness is compatible with indeterminism in physics. In this regard we do not differ from motor cars or plants. It is not plausible that in order to explain the behaviour of a motor car we need to go outside the resources of the physical sciences, and the same goes for the growth of plants since the demise of vitalism. I will take it for granted that the same goes for us; but I do need to say something quickly about what I mean by the resources of the physical sciences.

I mean not only the kinds of properties, relations, laws and particulars that appear explicitly in one or another of the physical sciences, but also whatever supervenes logically on these ingredients. Thus, the property of being two grams more massive than anything of twenty grams counts as a physical property for our purposes, despite the fact that it does not appear explicitly in any physical science. Likewise, appropriate functional properties count as physical for our purposes: they logically supervene on enough about the explicitly physical, including enough about physical law. 2 Also, it makes no difference if a property is given a name that does not appear in any physical science. We might use ‘Fred’ for our property of being two grams heavier than anything of twenty grams, but this exercise in renaming would not turn the property into a non-physical one.



Transparency theorists, as I will call them, insist that whatever we say about mental causation must make perfect sense when

-81-

viewed in terms of what the physical sciences say about the causation of behaviour. We should regard the story told by the physical sciences as the ‘real McCoy’ and seek a story about mental causation which makes it transparent how the mental story coheres with the physical story. What coherence comes to is itself a matter of debate, but the general idea is that, in some way or other, the mental story is a telling of the physical story, or a fragment of the physical story, except that it is framed in the language of psychology and not in the language or languages of the physical sciences. One guiding model is the relation between the causal explanations offered by the molecular theory of gases and those offered by the thermodynamic theory. Once you properly understand the explanations offered by the molecular theory, it is transparent how they cohere with those offered by the thermodynamic theory—the terms of the second theory pick out certain molecular properties that figure in the first. There is no ‘explanatory gap’ that needs to be filled, merely a grasp of how words like ‘temperature’ and ‘pressure’, as applied to gases, pick out the energy and momentum properties of gas molecules which do the causing of the phenomena we explain in terms of temperature and pressure. In consequence, the story transparency theorists will tell when we know enough of the neuroscience about mental causation will have an essentially dual character: there will be a part that tells how behaviour, internal states and surroundings, all described in physical language—as neurostates playing certain functional roles, as sentences of mentalese, as stimulations of sense organs by the environment, as movements of limbs, and the like—causally interconnect, and there will be a part that tells about which of these goings on should be described in psychological terms—which neurostates or functional states are to be called beliefs, which limb movements are to be called actions, and so on. Why does this mean that the physical story counts as the ‘real McCoy’? Because (of course) calling something a belief or an action does not affect its causal powers or ancestry.

By contrast, strong autonomy theorists insist that we have two separate, and entirely proper in their own spheres, stories about the origins of behaviour, or maybe about behaviour in two different senses of ‘behaviour’. The two stories are not, of course, entirely separate. For example, if it is physically impossible for a body to move at all, we can be confident that there is no true explanation in terms of belief and desire of the body’s moving.

-82-



Nevertheless, each style of explanation should be judged by its own lights and standards, and each has its own virtues and aims. Neither is the ‘real McCoy’. I call these latter views strong autonomy views in order to distinguish them from non-controversial positions which restrict themselves to emphasising that we often do not know enough of the physical detail and laws to predict and explain behaviour at all precisely, and to pointing out that, even if we did have the needed physical information, its complexity would often defeat our attempts to extract interesting predictions from it. The situation is not the relatively simple one presented by the reduction of thermodynamics to the molecular theory via statistical mechanics (its simplicity being, of course, one reason philosophers are so fond of the example). In consequence, run more modest autonomy views, we need psychology in practice, and, indeed, in principle, in anything but the most general sense of ‘in principle’. 3

Strong autonomy views are hard to maintain, or so it seems to me. I will distinguish three varieties. One turns on a view about how the language of psychology divides things up; the second is really a style of interactionist dualism of an attribute variety; the last is a view about kinds of causation and causal explanation. I will start with the first view (which is, I take it, in one form or another, the most widely entertained version of strong autonomy).

8.2.2 Problems for two language versions of strong autonomy

According to the first view, psychological language divides up objects, events, processes, states—most generally, things—in ways that are importantly different from the ways that physical language does. It is not that the things that are being divided up have, or are being said to have, extra properties that fall outside the ken of physical science—we are not being offered Cartesian mysteries!—but rather that the divisions being effected by psychological language cut across those being effected by physical language in a way that cannot be captured in physical terms. As Donald Davidson remarks, we are not being offered a view about things ‘in themselves’, but rather a view about things ‘as described in the vocabulary of thought and action’. 4 When we describe things in intentional terms we dance to a different tune, or, as Davidson says, ‘the conditions [for the application of psychological terms] have no echo in physical theory’. 5 There

-83-



is, therefore, no systematic way of moving between the physical and psychological stories about how things are, and in particular we cannot, as matter of principle, read off the psychological story about ourselves from the physical story about ourselves and our world. It is wrong to see the physical story as the full story and view the psychological one as what results from understanding aright which items in the physical story have names drawn from the vocabulary of psychology. This is why the two stories are in some important, strong sense autonomous. It will be useful to have a tag for views of this kind. I will call them ‘two language’ views.

The most influential exponent of this kind of view, Davidson, allows that beliefs and desires cause behaviour. Indeed, he famously played a prominent role in criticising the view that because there is some kind of conceptual connection between someone’s beliefs and desires, and what they do, there cannot be a causal connection. 6 He denies, though, that there are strict psychophysical laws. A common criticism of this view is that it is inconsistent with granting that the physical and psychological are causally interconnected. For the causal interconnections are not merely between physical and mental, but between physical and mental by virtue of their very physical and mental natures. But then, runs the objection, there must be laws relating the physical way things are with the psychological way they are that count as strict psychophysical laws. They will not, of course, be everyday laws like ‘Someone who desires water generally behaves in a way that leads to water’, or ‘Bodily damage often causes pain’ which are not strict, but rather descendants of them with all the exceptions handled in (presumably) physical terms. 7 However, the issue of strict psychophysical laws in this sense is tangential to the two language version of strong autonomy. (I am indebted here to comments from James Hopkins, though he should not be held responsible for anything I say.) For, first, the relation between a rich enough physical story about us and our world, and the psychological story is not causal. We know from the (widely accepted, despite the controversies over how to state it) supervenience of the psychological on the physical that, in some suitably circumscribed sense, the physical determines without remainder the psychological: settle enough of the physical story about us and our world, and you settle the psychological story about us. This means that although certain parts of the physical story about our world are causally responsible for its psycholo-

-84-



gical nature, the relation between the (full) physical story and the psychological story is one of determination without remainder; and however this should be spelt out in detail, it is implausible that it is a species of causal relation. It will be more like the way the distribution of H2 O determines the distribution of water, or the way detailed information about who is married to whom along with information about deaths determines who are widows, or the way sufficiently detailed information about what a computer program does determines that it is a word-processing program—none of which is causal determination. Second, Davidson is very explicit that by ‘strict laws’ he means the kind of strictness we (arguably, and anyway according to him) seek in the fundamental physical sciences (see ‘Thinking Causes’, especially), and it is plausible that any law that contains significant occurrences of psychological terms will automatically, by virtue of the widely acknowledged vagueness and indeterminacy of psychological language, not be strict in this sense. (Thus, the supervenience of the psychological on the physical must be stated in a way that counts, for example, its being vague whether or not something is a case of belief or mere information processing, as a way of being psychologically alike.)

The central claim made by the two language view that leads to strong autonomy is, rather, that no amount of physical information about us and our workings would allow one to read off—that is, infer without further empirical information—the psychological story about us. 8 This is the sense in which the physical story is not the whole story. By contrast, when we know enough about certain molecular properties of certain gases and how they affect the behaviour of these gases, we do not need extra empirical information to know about the temperature and pressure of these gases. The semantics of the terms ‘temperature’ and ‘pressure’ tell us that the situation as described in molecular terms is also correctly described in certain temperature and pressure terms.

I will press an epistemic and, in the next section, a semantic objection to the two language view.

There is debate over whether we infer the thoughts of others from their behaviour in circumstances. For some, the only alternative is telepathy. But others, without endorsing telepathy, insist that we don’t always, or even typically, infer the thoughts of others from behaviour: perhaps on the ground that we don’t always consciously do so; or perhaps on the ground that, at least sometimes, we cannot characterise the behaviour independently

-85-



of that which it is supposed to support (‘I know it was the look on her face that told me she was annoyed, but all I can say about the look is that it is the sort people have when they are annoyed’); or perhaps on the ground that we are often more certain of another’s mental state than we are of their behaviour; or perhaps on the ground that our fundamental evidence for attributions of thoughts to others is their behaviour characterised qua intentional actions. 9 We need not enter this debate. What is important for the argument I am about to offer is that our opinions about the thoughts of others are typically caused by what they do. It is the look on someone’s face that tells us that they are annoyed; the wine they drink that reveals their preference for red; the words they write that make public their views on politics; and so on. This is why, when we want to know what others think, we typically direct our attention towards their bodies and what those bodies do. 10 This point about causal origin places a constraint on when our opinions about the thoughts of others are trustworthy. It must be reasonable to hold that what they think is part of the causal explanation of how their bodies move. But we know that a complete causal explanation of behaviour can, in principle, be given in purely physical terms. It follows that behaviour is not a reliable guide to anything over and above what is to be found in the full physical story about the causation of behaviour. Analogy: once we have a complete story about the causal origins of the fossil record, we have a story which rules out the fossil record being a reliable guide to how things are that significantly outruns that story. If behaviour is to be a good guide to what people think, it must be possible to move without further empirical information from the right kind of physical story about our world to a psychological story about it or, better, in view of the indeterminacy of the psychological, to one or another of a limited range of psychological stories about our world. Many hold, plausibly it seems to me, that the supervenience of the intentional on the physical shows that the physical metaphysically necessitates the intentional. If this necessitation is interpreted to mean that we can move without further empirical ado from a rich enough physical story about us and our world to an intentional one, then there is no special epistemic worry generated by the combining possibility of completely explaining behaviour in physical terms with the fact that our opinions about what people believe and desire are causally grounded in behaviour. But, equally, we would not then have a version of strong autonomy. If, however, it is insisted that

-86-



the necessitation is a posteriori (necessary a posteriori), 11 we may well have a version of strong autonomy, but the epistemic problem remains. We need to justify what will be in this case be the extra step (if it is not an extra step, I wonder what ‘a posteriori’ means here) of supposing that behaviour is a reliable sign of psychological nature as well as of the physical nature that fully explains it. 12

8.2.3 A semantic objection to the two language view

We use psychological language to say how things are. When I say that Tom and Mary both believe that snow is white, I am seeking to capture something they share, some respect in which they are alike. I may fail, of course. They may not be alike in the respect that matters—in which case I speak falsely—but unless we seek, and sometimes succeed, in capturing similarities when we use psychological language, there would be little point in using it. Indeed, there would be no point as far as the task of representing how things are goes—all that would be left would be the achievement of goals like making a noise and passing the time. It is a nice question in analytic ontology how we should cash out what it is to be similar; what exactly is involved in acknowledging that language, qua vehicle of representation, divides things up according to how they are. Is it a matter of a common relation to a Platonic universal? Is it to be understood in terms of a sparse theory of properties of the kind championed by David Armstrong? Is it to be analysed in the deflationary style of resemblance nominalism? 13 I will talk variously of similarities, properties, and ways things are, but as far as I can see nothing I say depends on which account, if any, in analytic ontology is correct.

The semantic problem for two language theorists is to say which similarities psychological language is in the business of picking out. As they eschew Cartesian mysteries, they cannot say that the similarities lie in the sharing of special ‘dualist’ natures or properties. What then are left but similarities that count as physical? But in that case there must be some systematic relationship between the way physical and psychological language carves things up. The issue here is not about whether the similarities that psychological attributions seek to capture lie solely in the subjects of those attributions. Two language theorists often argue that S’s psychological states are in part a matter of which states expert

-87-



interpreters would ascribe to S if they had the requisite information (about S’s behaviour, as the view is usually articulated); they espouse some version or other of interpretationism. 14 The word ‘properties’ in this discussion is not just intended to be neutral with regard to the debates in analytic ontology but also to be neutral as between more or less intrinsic properties and relations. Even if the key similarities lie in part in relations between subjects, on the one hand, and concept and language-wielding interpreters, on the other, it remains the case that an austere ontology requires that they must be found somewhere in the physical account of what is going on, including perhaps physical relations between purely physical subjects and purely physical interpreters. Also, the similarities may be, to one extent or another, disjunctive and unobvious (it may have been important for our survival that we were good at picking up commonalities that are disjunctive and unobvious).

It might be objected that two language theorists can insist that the similarity being asserted between John and Mary by one who says that they both believe that snow is white is simply that they both believe that snow is white. But it is implausible that the similarity being asserted is that both satisfy the predicate ‘believes that snow is white’. It is true of two square things that they both satisfy ‘is square’, but their so doing is matter of how they are. If it wasn’t, language would fail in its purpose of representing how things are. But if language represents how things are, we can ask how things are with John and Mary (and maybe with their surroundings and interpreters) if they both satisfy ‘believes that snow is white’, and two language theorists must grant that it is some sharing of physical ways things are, including perhaps physical ways interpreters are; to be silent on the subject is to countenance Cartesian mysteries. In sum, unless there are more than physical properties (similarities, ways things might be, etc.) in play, there is no via media between the view that psychological language divides pretty much at random and the view that it divides at joints that, at some level, make good sense in physical terms. As David Lewis notes, we are finite purely physical entities and so can hardly be picking up patterns that in principle defy physical capture. 15 But then the psychological story about our world is not strongly autonomous with respect to the physical story about our world. For the physical story will be the whole story about our world, with the psychological story being recoverable from it by reference to the part

-88-



that tells us how to make good sense of the divisions psychological language makes in physical terms—just the picture that transparency theorists have of the relation between physical and psychological explanations of behaviour.

8.2.4 The extra properties version of strong autonomy

The objection just presented turns crucially on the metaphysical austerity of the two language version of strong autonomy. One response for a strong autonomist is therefore to add some properties to the physical picture of what things are like (which, for the resemblance nominalist, will be nothing more than adding to the ways things can resemble each other, so adding to the sets that contain things that resemble each other). This is what Tim Crane has recently done. He posits psychophysical laws connecting the psychological way we are with physical matters, insisting that the psychological way we are does some causing on its own behalf, distinct from that done by the physical. 16 This is, however, really an attribute style of interactionist dualism and faces its notorious problems, most pressingly the question of how to reconcile a causal role for the special non-physical properties that figure in the laws with the causal closure of the physical. Crane suggests, as does D.H. Mellor, that this problem has been overplayed in the literature. They argue that it is simply a physicalist dogma that a complete causal explanation in one set of terms, physical as it might be, excludes a real, independent causal role for properties that fail to supervene on them. 17 Mellor, in particular, emphasises the logical possibility of complete overdetermination of certain physical matters both by the physical and by the non-supervening mental. However, the issue is not whether epiphenomenalism of the non-supervening mental follows logically from the claim that anything physical can be completely causally explained physically, but whether it is the only rational position that remains open. The story the biological sciences tell about the growth of plants does not logically exclude the story vitalism tells, but it does eliminate it as a serious contender.

8.2.5 Strong autonomy via the denial of homogeneity

The final variety of strong autonomy denies the homogeneity of causation and causal explanation. Of course, runs the view, the mental causes and is caused by the physical, and of course

-89-



the physical causes and is caused by the physical, but there are two quite distinct causal relations in the two cases. It is therefore wrong to suppose that there is any special claim to completeness in the physical causal story, and a fundamental mistake to seek to make sense of mental to physical, and physical to mental causation, in the light of what we know about physical to physical causation. This view has few adherents. I will mention just one problem for it. It requires an interactionist, dual attribute metaphysics. This is because it is incredible that mental to physical causation is different in kind from physical to physical causation while holding between the same kinds of things, for then the difference could only be made by the words we use to pick out what is in nature identical, by whether we use words from the vocabulary of psychology or a physical science to pick out what are being said to explain causally. And that’s word magic.

The rest of this chapter will be concerned with mental causation under the assumption of transparency. We will view the problem of mental causation as that of identifying where in the physical story about us and our world lie the parts that in effect tell us about mental causation—or will tell us when we know enough of the physical story. (And, of course, many discussions of mental causation start from this position.)

8.3 The location of mental states 8.3.1 Preamble

If we are going to make sense of mental causation under the assumption of transparency, we need to identify mental states with physical states, at least to the extent of saying roughly what kinds of physical states they might be. We need to say something about where in the physical story the mental states are. The usual view is that mental states are certain internal states in subjects’ brains. But there is dissension, so it is worth pausing to note briefly why the brain is the plausible place to locate mental states, and more particularly belief, the example I will use to outline the case.

8.3.2 Mental states as brain states

Consider what happens when I enter a room and acquire the belief that there is a pot of coffee on the table which then leads me to

-90-



move towards the table. My belief is something that happens to me as a result of my body’s change of situation in the world which then affects the movements of my body in certain distinctive ways. So, in order to locate the belief, we need to ask for the location of the crucial changes consequent on my change of situation that affect how my body moves. One place there are changes is in the air around me, but we know that, unless they lead to changes inside me, I do not acquire the belief—which is why I need to be looking in the right direction with my eyes open. Also, there are changes in my eyes, but we know that unless they lead to changes ‘further in’, I will not acquire the belief, and my behaviour will not be affected in the distinctive ways. In short, the crucial changes that matter are somewhere in the brain—it is a discovery of science that unless the brain changes in certain ways, subjects do not acquire beliefs, and their behaviour is not affected in the ways distinctive of belief. Thus, as an empirical matter of fact, the brain is where we should locate belief; and the same goes for mental states in general, as similar considerations apply across the board.

Three warnings are in order here. First, locating mental states in subjects’ brains does not mean that it is how things are with their brains alone that determines what kinds of mental states they are in. Sunburn, to take Davidson’s example, is on the skin; it is a state of the skin. 18 This is because the skin is where the sun has the effects that give rise to the distinctive results of sunburn, just as the brain is where our surroundings have the effects that give rise to the distinctive results of belief. But what makes a certain result on the skin a burn is its being of a kind with damage produced by heat, and what makes it sunburn is its having been produced by the sun, and these are both matters concerning how things are remote from the skin.

Second, although ‘state’ is a word for kinds, identifying mental states with brain states is compatible with denying that the properties we normally have in mind when we talk of mental properties are brain properties. Brain states, those kinds, have properties that are not specific to brain states—certain functional properties, for example—and we leave open the possibility of identifying mental properties with certain of these properties. In locating mental states in the brain, we are committing ourselves to a type-type identity theory, but not to the most controversial type-type theory.

Of course, many regard any commitment to a type-type

-91-



mind—brain identity theory as bad enough. One source of dissent is expressed by Colin McGinn as follows:

mental states are not, as brain states literally are, situated inside the head. A modal argument quickly puts paid to any hopes of identification here: you could have the same brain states in a world in which the objects of the corresponding mental states did not exist, but you could not have those mental states in such a world…. Intrinsic (in the head) physicalism cannot…do justice to externalism about the mind. 19

But this is like arguing that sunburn cannot be a state of the skin because you could have the same state of the skin in a world that lacked a sun, but you could not have sunburn in such a world. A second source of dissent turns on the claim that placing those mental states that fall into the category of intentional states inside the head makes it impossible to explain how they can be about objects in the world. The claim is that if we locate, say, belief inside the head, we make it impossible to capture adequately the way a belief may be about particular things outside us. We must, runs the argument, view my belief that there is a particular tree in front of me as literally including the tree itself. 20 It is hard, however, to see how this would help with the (hard) problem of intentionality. Obviously, my belief about the tree is not located in its entirety where the tree is. For it would not in that case be a state of mine at all. But then the issue remains of what links this state of mine, be it right to refer to it as ‘my belief that there is a particular tree in front of me’ or not, with the tree; that is, the ‘aboutness’ problem remains.

A further source of dissent is the famous multiple realisability point. It is argued that we cannot identify a given kind of mental state with some kind of brain state because that mental state might have been realised differently in the brain. 21 But we can identify the behavioural gesture of assent in the West with nodding the head, despite the fact that some different gesture, say, waggling the head (moving it from side to side instead of up and down), might have been the gesture of assent in the West (waggling is in fact the gesture of assent in India, I understand). Multiple realisability tells us that which kind of brain state some given mental state is may vary from case to case—after all, which kind of physical structure is the word for red varies

-92-



from country to country; it is ‘red’ in Australia and ‘rouge’ in France—but multiple readability does not teach us that mental states are not brain states.

The final word of warning concerns the sense in which token mental states are token brain states. It is sometimes argued (plausibly to my mind) that because the individuation conditions for mental state tokens differ from those for brain state tokens, the sense in which mental state tokens are brain state tokens is, strictly speaking, the ‘is’ of constitution. 22 Nothing in our discussion turns on this issue. For if mental state tokens are constituted by brain state tokens rather than identical with them, it remains true that mental state tokens are in the brain and that their causal powers are those of the relevant brain state or states. The (true) view that a table at any time is constituted by, not identical with, an aggregation of molecules does not alter the fact that the table is located where the aggregation of molecules is and has precisely the causal powers of the aggregation of molecules.

If mental states are states of the brain, then mental causation is causation by states of the brain. Should we then hand over the topic of mental causation to neuroscience? Ultimately yes, but before the handover some important issues much discussed in the literature need addressing. They will take up the rest of this chapter.

8.4 How could brain causation be mental causation? 8.4.1 Behaviour and movement

Typically, we appeal to mental states to explain intentionally characterised behaviour. We explain Mary’s reaching for the glass in terms of her desire for milk, and her belief that it contains milk; we do not explain her arm moving at so many metres per second in terms of what she believes and desires. But it is precisely an arm’s moving at so many metres per second and such like—‘raw’ behaviour, mere bodily movement—that brain states cause and causally explain. Thus, it is sometimes objected, we cannot think of causation by mental states as a species of causation by brain states. What gets causally explained is different. 23 But we explain footprints in terms of feet, and poisonings in terms of poisons. How do we reconcile this with the fact that feet cause depressions in the ground, and poisons cause illness?

-93-



By noting that when a foot causes a depression in the ground, it thereby makes it true that the depression counts as a footprint, and that when a poison causes illness, it thereby makes it true that the illness counts as a poisoning. In the same way, when a mental state that is a brain state causes a certain movement in a certain way, it thereby makes it true that the subject in question is acting. The fact that mental states causally explain behaviour proper, whereas brain states causally explain bodily movement or raw behaviour, does not mean that mental states and brain states are different, nor does it mean that what they cause is different; rather, the very fact that certain brain states which are mental states cause in the right way certain movements, entails that those movements constitute behaviour proper. There is a nice question in the philosophy of action cum philosophy of language as to what noun phrases like ‘Mary’s reaching for a glass of milk’ denote: a movement (though not described merely as such), a complex of movement plus mental cause, a mental cause alone, a complex including mental cause, movement, milk,…? 24 But this is by the way in the present context.

A similar point arises in connection with the common observation that classifications by behaviour proper cut across classifications by movement: many different movements may constitute signalling a taxi, and the very same bodily movement may constitute brushing away a fly, indicating dissent or signalling a taxi. The very same phenomena—raw behaviour—may be typed qua movement or typed in part by causal origin. But how we choose to type the phenomena does not affect what is being typed, nor what causes what is being typed. It would be extraordinary if how we chose to group things together for one or another purpose changed causal facts.

8.4.2 Internal causation of environmentally embedded behaviour

A second worry is that the behaviour that is explained by beliefs and desires (and intentional states in general) is typically changes in the way a subject is oriented with respect to the environment—movement towards milk, movement away from tigers, and so on. Indeed, only when we look at behaviour in this way do we discern the patterns that enable us to make sense of someone as an intentional agent, as Daniel Dennett has particularly emphasised. 25 Now it is easy to see how a neurophysiological state

-94-



might explain movement towards milk in the sense of explaining movement in direction D at so many metres per second, where direction D happens to be where the milk is. But then the fact that the motion is directed towards milk is a fluke. How does a neurophysiological state explain non-accidental movement towards the milk? After all, the neurological state is inside the subject, and the milk is outside the subject. As opponents of the kind of internalism represented by any view that identifies intentional states with brain states like to put it: how can internal states explain the essentially relational character of the behaviour that psychology explains? 26

The answer is by virtue of the fact, if it is a fact, that the milk impacted on the subject sometime before the reaching towards it took place. The reaching for the milk is causally downstream from a perceptual contact with the milk, or with someone who has told our subject where the milk is, or some such, and the contact provides information that modifies the internal neurological state so that it guides the arm towards the milk. Likewise, we can explain the way plants non-accidentally orient themselves towards the sun in terms of how their internal states get appropriately modified by the direction of the sun’s rays on the plant before the corresponding movement. When we open up the plant we will find the state that does the work, and also find out how it is sensitive to the sun’s direction in such a way that the plant orientates itself appropriately towards the sun. It would be a mistake for philosophers to write to biologists telling them that the internal states that they cite in their texts cannot, as a matter of principle, explain the relational nature of the movement of plants. The same goes for philosophers writing to animal biologists—and we are animals.

8.4.3 The causal role of mental properties

The amount of explosive, the timing of the lob, the height of the mountain, the sharpness of the curve in the road are among the many factors we cite as causally responsible for the size of craters, the winning of games of tennis, the exhaustion of climbers, and the occurrence of accidents. It is very often properties, how things are, that we cite as causally responsible. This is, I take it, a commonplace; what is a matter of lively debate is how to incorporate the commonplace into the events framework made familiar by Davidson for thinking about causation, and how we

-95-



should relate the properties of the commonplace to the issues in analytic ontology referred to earlier. 27 But even setting aside these thorny issues, our commonplace gives rise to a major problem, a problem that dominates much recent discussion of mental causation. It arises from a combination of plausible theses about the kinds of properties that can be causes, with plausible theses about what makes it true that one or another brain state is one or another mental state, as we will now detail. 28

Although multiple realisability does not, we argued, stand in the way of identifying mental states with brain states, it does tell us that what makes it true that some given brain state in some given kind of creature is such and such a mental state with such and such a content is not its neurological type. Analogy: we can identify poison kinds with chemical kinds—each and every poison is some chemical kind or other—but what makes it true that arsenic is a poison is not its chemical kind, but rather what that kind does, its functional role. We might, accordingly, think of what makes it true that some brain state is one or another mental state as that feature of the brain state that best qualifies as a properly mental property of it. For example, a functionalist about mental properties could allow that a certain kind of brain state is belief that snow is white in humans, just as a certain kind of chemical compound is the most dangerous poison for humans, while insisting that the properly mental property of the brain state is not the neurological kind that fills the role but is, rather, the functional role itself.

With this background, we can now identify the problem that dominates much current discussion of mental causation. It comes in two forms. The first form presupposes some kind of functionalist account—wide or narrow—of mental properties; the second presupposes some reasonably strong kind of externalism about intentional content in particular, and mental kind-hood in general.

8.4.4 Functional properties and causation

Just as the sharpness of the curve in the road may explain the accident by being part of what caused it, my believing that there is coffee in front of me may explain my moving in a certain way by being part of what causes me to so move, or so it very much seems. But if my believing that there is coffee in front of me is a functional property, saving the appearance requires that functional kinds be causes, and there are two good reasons to doubt

-96-



that functional kinds can be causes. Consider first a fragile glass that shatters on being dropped because it is fragile, and not because of some peculiarity in the way it is dropped. There will be, as it might be, a certain kind of bonding B between the glass’s molecules which is responsible for the glass being such that if dropped it breaks. Some say that there being such a ‘categorical’ basis is a matter of necessity; others say that ‘bare’ dispositions are possible. 29 All that matters here is that the dispositions, and more generally the functional properties that concern us, are all constituted ones, be this a matter of necessity or not. But then it is bonding B together with the dropping that causes the breaking; there is nothing left for the disposition itself to do. All the causal work is done by bonding B together with the dropping. Or take an even simpler example, the way a thermostat consisting of a bi-metal strip turns a refrigerator on and off. What causes the refrigerator to turn on and off is the bending and straightening of the strip, and this in turn is caused on any particular occasion by the differential expansion of the two metal strips: there is no causing left to be done by the relevant dispositional properties. I have made the point for two very simple kinds of functional property, but the point clearly extends to functional properties in general.

Some functionalists about the mind think of their functional properties as narrowly specified in the sense that the relevant inputs and outputs are at or within the skin, or may be at or within the brain’s surface, whereas wide or arm’s-length functionalists think of the inputs and outputs as at least sometimes specified in distal terms, in terms of glasses of water, tigers, sunsets and the like. The difference does not matter for the argument just given. Either way, it is the categorical bases of functional properties, and not functional properties themselves, that cause: the transitions between mental states, the transitions from inputs to mental states, and the transitions from mental states to outputs will be governed by the neurological or syntactical nature of the mental states (depending on which are best thought of as the relevant categorical bases), and not their functional properties.

The second reason for insisting that it is the categorical bases of dispositional and functional properties that do any causing relates to what is right about Hume’s thesis that causal connections are contingent. There are necessarily true statements asserting causal connections. Thus, if a certain chemical structure is highly poisonous, and thereby gets to be correctly described as

-97-



‘the killer structure’, the statement The killer structure typically kills’ is necessarily true (modulo worlds where there is no killer structure). 30 However, what Hume had in mind is untouched by such examples. Hume held (I take it) that what a state causes, or would cause were such and such to happen, is not an essential property of that state. His claim is that for any state it is true that that very state might not have caused what it did in fact cause, and might not have had the causal powers it in fact has. The killer structure does not kill in some possible worlds—namely, in those worlds where the laws of nature or the internal nature of humans differ from how they actually are in such a way that that very structure is harmless to humans. In these worlds it is wrong to call the property in question ‘the killer structure’, but it is the same structure nevertheless (at least on reasonable ways of counting structures). However, to allow that fragility itself, as opposed to its categorical basis, causes breaking on dropping, would be to allow that there are properties that have causal powers essentially. If fragility does or would do the causing in one world, then it does or would do the causing in any world. In every world the fact of being fragile is intimately connected with the fact of breaking on dropping, and if that intimate connection counts as fragility causing or having the power to cause breaking in any world, it will so count in every world. 31 Clearly, the same goes for functional properties in general—to allow them as causes is to violate a good Humean principle.

8.4.5 Two possible responses for functionalists

What should functionalists say in response to the result that, by their own lights, mental properties are not causes?

The first response starts by emphasising that it would be misleading to describe the result as committing them to epiphenomenalism about mental properties. 32 To be epiphenomenal is to be outside the causal story about how things come about. Dispositions and functional properties are not epiphenomenal in this sense. The causal story about how a fragile glass comes to break includes its fragility: by the time you have explained how the glass’s categorical nature gets together with the (distally characterised) impact to cause breaking, you have said inter alia that the glass is fragile. Although dispositional properties do not cause, they are an ineliminable and integral part of

-98-



many stories about what does cause. They are not an addition to the causal story.

This point has wider ramifications. The ‘causal problem for content’ is often raised independently of functionalist theories of content, most often in the context of various forms of externalism about content. If content is in part a matter of how things are outside a subject, concerning for instance the subject’s causal history, word usages in the subject’s language community, or, as on Ideological views, how things were with the subject’s distant ancestors, there is a question about how content can enter causal explanations. As McGinn puts it: ‘According to externalism, contentful states are identified by reference to entities that lie outside the subject’s body…but the causal powers of a state must be intrinsically grounded; they cannot depend essentially upon relations to what lies quite elsewhere.’ 33 Any view of content that insists that content properties are not intrinsic, ‘syntactical’, shape or neurological properties, or some such, is inconsistent with content being causally efficacious. The usual response on behalf of such theories is to grant that content is not causally efficacious while insisting that a property intimately connected with content is. 34 Something of the shape, syntactical or neurological category, encodes the relevant semantic property, and the encoding property does the causing—that is how content gets to be causally relevant without actually causing.

There are two standard objections to this response. One is that it makes content per se epiphenomenal. Talk of the coding property doing the work is a fig-leaf to cover this fact. The other is that it invites the elimination of content on Occamist grounds. If all the causal explaining is able to be done by the coding property, we should eliminate the content property. 35 But now we can see how to defend the coding response from the two standard objections. The coding response should be set in the context of a view that holds that a rich enough story about a property of subjects that belongs in the syntactical category, which tells us about how the property relates to a subject’s surroundings, about its causal history and connections to other states, about the way it mediates between inputs and outputs, and so on and so forth, necessitates that the subject has mental states with such and such contents. The situation is formally the same as the one obtaining when fragility’s categorical basis in an object causes the object to break (in the right way). This necessitates that the object is fragile; it is a mistake to think of the fragility as an addition to

-99-



the picture which might be dispensed with in the spirit of Occam. The same goes for contents encoded in syntax. Of course, to make good this defence of the coding response, its proponents need a theory of content, a theory about what kinds of relations between syntactical states and their surroundings necessitate that subjects in those syntactical states have mental states with such and such contents. But this they typically claim to have. 36 Whether their theories are good ones is outside our remit.

Moreover, to return to the main theme of this section, functionalists about content can note that we give causal explanations in terms of dispositions. Richard Feynman (with a little help from informants) explained the Challenger disaster in terms of the loss of elasticity at low temperatures of an O-ring. It is not for philosophers to tell him that this cannot be right. We must allow that dispositional and functional properties are causally relevant, even if they are not causally efficacious.

There is a nice question about what we are doing when we give a causal explanation in terms of a disposition. (I’ll stick to dispositions rather than functional properties in general throughout this discussion, but the points carry over to the more general case.) It is tempting to give a simple answer. When we explain in terms of a dispositional property, we are citing the disposition’s basis, which we may or may not know, as the cause. Thus, Feynman was right if and only if the basis of the O-ring’s loss of elasticity at low temperatures caused the Challenger disaster.

There are, however, clear counter-examples to this simple answer. Here is one due to David Lewis. I simplify the science, but in a way which does not damage the philosophical moral. 37 Opacity, electrical conductivity and thermal conductivity in metals have the same categorical basis. The way free electrons permeate metals that are opaque, are good thermal conductors, and are good electrical conductors is responsible for the input-output behaviours distinctive of these three dispositional properties. Nevertheless, an explanation in terms of a metal’s opacity is clearly not the same as one in terms of its electrical conductivity; nor is it the same as one in terms of its thermal conductivity. The transmission of a telephone call down a wire is explained by its electrical conductivity, not by the wire’s opacity. Good results from cooking in copper-based saucepans are explained by copper’s high thermal conductivity, not by its opacity or electrical conductivity.

We need, I suggest, to think of causal explanations by appeal to

-100-



dispositional properties as having a dual character. When we explain by citing a dispositional property, we make two claims together. First, we say that what was caused was caused by the disposition’s categorical basis. If the categorical basis of the O-ring’s loss of elasticity did not cause the Challenger disaster, Feynman was wrong. Second, we say that the kind of inputs and outputs distinctive of the disposition in question lie in the causal path to what happened. (We can think of cases where the distinctive outputs are what is being explained as a special case: we want fragility to explain, albeit in a rather uninformative way, breaking, as well as matters downstream from breaking.) One part of what Feynman was saying was that the categorical basis of the loss of elasticity caused the disaster, and the other, more interesting part, was that what happened resulted from the kind of output distinctive of loss of elasticity: the O-ring failed to expand after compression, and its failure to do so led to the disaster. Or consider the difference between explaining a death in terms of a metal’s opacity, as opposed to its thermal conductivity or its electrical conductivity. Although the operative property of the metal will be the same in each case, in the first it will, say, have led to the obstruction of someone’s vision at a crucial moment; in the second, the property will have caused, as it might be, a surge of heat that burnt someone to death; and in the third, the death will be by electrocution. In each case the causally operative internal property of the metal will be the same; the differences lie in how this property interacts with different inputs to cause different outputs, and the right disposition to cite in a causal explanation will be settled by the nature of the inputs and outputs.

Finally, followers of the first line of response can point out that they can preserve the ‘here-and-now’ intuition: the appealing intuition that when I reach for a glass because I want some milk, and believe that the glass contains milk, it is something about me pretty much at or just before the time of reaching that explains the reaching. Although functionalists who take the first line must admit that mental properties do no causing, they can allow that their causal relevance is a here-and-now matter. What I am thinking at a time is a local fact about me at the time in question that potentially causally explains, is causally relevant to, my behaviour, and to what else I may come to think. For although functionalisms of either the narrow or wide kind deny that mental properties are intrinsic features of subjects, or of

-101-



subjects’ internal states, they can allow that mental nature in some important sense is in common between doppelgängers (or doppelgänger). Although what makes some substance water soluble is a relation to water, being water soluble is common to doppelgängers. The reason is that the relations that matter for being water soluble are relations to actual and possible inputs, including especially water ones, without regard to whether they are actual or merely possible, and doppelgängers agree in what would happen, while possibly disagreeing in what did or will happen, or is happening.

The first line of reply on behalf of functionalism, then, is to distinguish causal efficacy from causal relevance in a wide sense that includes efficacy but much else besides, and to argue that content is causally relevant without being efficacious. 38 This reply is a special case of a widely discussed approach to the problem of mental causation. 39 According to it, we have properties picked out in the language of neuroscience, and distinct properties picked out in the language of psychology. The second set of properties is not caused by the first. Rather, as it is often put, the psychological properties are realised by the neurological properties, or the neurological properties are implementations of the psychological properties, though different writers give different accounts of what realisation and implementation here come to. 40 The theoretical challenge is then seen as that of showing how the relation between the neurological properties and the psychological ones is such that causal relations between the neurological properties induce (but not in a causal sense) appropriate causal relations, which may or may not count as causal efficacy, between the psychological properties. The first line of reply on behalf of functionalism is a special case in which the relation between the neurological properties and the psychological ones is that of categorical bases to functional properties, and the causal connections between the neurological properties induce causal relevance between the psychological properties. An important question raised by the general strategy is whether general accounts are available of how causal relations between one set of properties might induce causal relations between another set of ‘realised’ properties. In ‘Functionalism and Broad Content’, Jackson and Pettit argue that often the causal relevance of P to some result should be understood in terms of the causal efficacy that the various ways that P might be realised would have in producing the result; in such cases, P can be thought of as programming for

-102-



the result. Indeed, the account above of how dispositions get to be causally relevant to their manifestations is a special case of this style of account. The first line of reply on behalf of functionalism seems to me to be a viable one, but I now prefer a second line of reply. It is motivated by the appeal of the idea that what a subject believes and desires actively monitors what a person does. You are describing the movement of a fly as it crawls up the wall. It is very appealing that what you call out is controlled by the content of your perceptual experience. To do justice to this intuition and its many kin, we require that mental properties be causes of bodily movements (and of other thoughts, if it comes to that); we require that species of causal relevance that counts as causing simpliciter. 41 But functionalists who take this path must repudiate the earlier claim that mental properties are functional properties. Instead, they must hold that mental properties, or at least some mental properties, the ones that are causes, are the kinds that fill the functional roles. The distinction between the kinds that fill the roles, and the functional kinds that are the roles themselves is sometimes marked by calling the first realiser properties or states, and the second role properties or states. In these terms, the first line of reply on behalf of functionalism says that mental properties are role properties, and are causally relevant without being causes, literally speaking, of behaviour and other mental states; the second says that mental properties are realiser properties, and are literally causes of behaviour and other mental states. It is sometimes objected that even if we can identify the content property with a neurological, realiser property (or perhaps with some more structural but still relatively intrinsic, syntactic property), we do not achieve what we are after. ‘Isn’t it still the case that it is the nature of the role occupied, and not the nature of the occupier that matters for content, and the nature of the role occupied is causally irrelevant?’ But, of course, the nature of the role occupied is causally relevant in a wide sense. Indeed, one way to account for this is the story in terms of programming. A similar objection is sometimes put in terms of whether the content is causally efficacious qua the content that it is. ‘Isn’t it true that the neurological state N will cause what it causes qua the neurological state that it is, and not qua the content C that it is?’ But, on the functionalist picture, what N causes is part of what makes it the content C that it in fact is. And it seems a fair capturing of the intuition that a content property causes the

-103-



behaviour (and the other mental states) it distinctively explains in virtue of being C, that its causing the behaviour (and mental states) in question is part of what makes it C. Analogy: we distinguish between what the killer structure does, and what it does qua killer structure, and the distinction is made in terms of the functional roles that make something a killer structure—to do something qua being a K is to do something that is part of what makes for being a K.

8.4.6 Externalism and the causal problem

We observed that externalism is like functionalism in raising a causal problem for content. But the problem raised by externalism has a special feature. Functionalism can, as we saw, allow that doppelgängers are psychologically alike. 42 It can therefore satisfy the ‘here-and-now’ intuition, as we noted. When my reaching for the milk is explained by what I believe and desire, it seems that it is something about me (‘here’) at the time of reaching (‘now’), something that I would share with any doppelgänger of myself, which explains what I do. But, on just about any version of externalism, the content of belief and desire is broad; it involves how things are outside subjects concerning their causal history, surroundings, word usages in their language community, and the like; it is not in common between doppelgängers. 43 Externalism about content violates the here-and-now intuition. 44

One response to this problem has been to invoke a special notion of narrow content. On one way of fleshing the idea out, we should distinguish two notions of content: there is content that has truth conditions, semantic content, and there is a separate notion of content for explaining behaviour, causal explanatory content. 45 Causal explanatory content is narrow, and so we preserve the intuition that when we explain in terms of what subjects believe and desire, we explain in terms of a here-and-now feature of them. On some views, e.g., Fodor’s in ‘Methodological Solipsism Considered as a Research Strategy in Cognitive Psychology’ and Psychosemantics, the causal explanatory narrow contents are the coding states we discussed earlier, and so the narrow content is causally efficacious in addition to being here-and-now. On other views the causal explanatory narrow content is a narrow functional property and is here-and-now relevant but not efficacious. In any case, the postulated bifurcation of content is unattractive, or so it seems to me. Jones’s belief’s being that there is

-104-



a tiger nearby explains her running, precisely because to believe that there is a tiger nearby is to believe that it is true that there is a tiger nearby. If she had believed that it was false that there was a tiger nearby, she would not be running. Again, the role of her desire in getting her to run is tied to her desiring that it be true that she goes on living. Moreover, the way we use belief and desire to predict and explain behaviour presupposes that their contents are truth-valued. The behaviour that we regard as predicted and explained by subjects’ beliefs and desires is behaviour that would satisfy their desires, that is, would make true what they desire, if their beliefs were true.

A more promising response is to note the many cases where we causally explain in broad or relational terms. The tracking behaviour of a predator may be explained by its perceptual contact with its prey, and perceptual contact is defined in part in terms of causal origin. Causal explanations in economics are often relational in character: such notions as competition for resources, and poverty traps are clearly relational in character. And in astronomy the movements of the planets are explained in terms of their being planets, bodies having the relational property of being under the influence of a sun. 46 But it is important to note the limitations of this response. Clearly, knowledge of relational properties, and in particular matters of causal origin, can provide useful information about how things come about, and are thereby part of good causal explanations. (If information about causal origin is not causally explanatory, what is?) Moreover, the explanatory value of knowing the causal origin of the information-sensitive states inside a subject is clear. You then know which thing’s properties are being encoded inside the subject, even if you do not know which properties of that thing are being encoded. But the problem about here-and-now causal efficacy, as opposed to causal relevance in some wide sense, is left unaddressed by these observations. This latter problem concerns which properties of my internal states here-and-now are making my arm move, and surely the answer to that question must advert to local properties of my internal states. From this perspective, the example of the planets heightens the worry rather than laying it to rest. What operates on a planet is the gravitational field where the planet is at the time in question. We are not, it seems, being offered a reply to the worry about how what a subject believes and desires can have here-and-now causal efficacy. Moreover, the situation is worse than it is for functionalists who take the first line of reply to the causal

-105-



problem mentioned earlier. Although they must deny causal efficacy to content, they give it here-and-now causal relevance, while the property that externalism identifies with content is not a here-and-now property to start with.

A third response to the causal problem for externalism is, it seems to me, the most attractive, but it rests on a contentious premise. The case for broad content is a powerful one and will not be rehearsed here. But despite the tendency in some writers simply to equate truth conditional content with broad content, I think (and this is the contentious part) that we should accept that there is narrow, truth conditional content—intentional states per se are not broad—and, what is more, that whenever someone is in an intentional state with broad content, what makes it true that they are in such a state is the combination of the narrow intentional state they are in with certain connections between that narrow state and their surroundings, including especially the causal origin of the narrow state. This is, obviously, a fairly deflationary position on broad content, and is controversial. I will not try and defend it here. 47 I will simply note that it handles the here-and-now problem without excessive repudiation. For it means that whenever subjects act on what they believe and desire, there is a here-and-now fact about what they believe and desire that is causally relevant to what they do. When we cite what they believe and desire in terms that impute broad contents, we give information about, say, causal history, but, underlying these historical facts, there is some more or less current, local fact about them which settles what they believe and desire in the narrow sense, and which explains what they do. Whether this narrow local fact is causally efficacious in addition to being causally relevant is a further question, the answer to which depends on matters discussed above in connection with functionalism. If we can identify this narrow local fact with a relatively intrinsic feature, presumably a neurological property of some kind, or something of a more structural kind that supervenes on the neurology, then the narrow content that underlies broad content will be causally efficacious; if we cannot, then the narrow content will be causally relevant without being efficacious, but at least it will be local. 48

Notes 1 For a sketch of the reasons I favour for rejecting eliminativism about intentional

states, see Frank Jackson and Philip Pettit, ‘Folk

-106-



Belief and Commonplace Belief’, Mind and Language, 8 (1993): 298-305. 2 I talk in terms of (logical) supervenience rather than entailment, partly because it

is usual to do so, and partly to sidestep certain complications about infinite disjunctions.

3 On one reading (but others are possible), this is the message of Jerry Fodor, ‘Special Sciences, or the Disunity of Science as a Working Hypothesis’, Synthese, 28 (1974):97-115.

4 Donald Davidson, ‘Psychology as Philosophy’, reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980, pp. 229-39, p. 230.

5 ‘Psychology as Philosophy’, p. 229. See also William Child, Causality, Interpretation, and the Mind, Oxford, Clarendon Press, 1994, and John McDowell, ‘Functionalism and Anomalous Monism’ in Ernest LePore and Brian McLaughlin, eds, Actions and Events, Oxford, Basil Blackwell, 1985, pp. 387-98.

6 Donald Davidson, ‘Actions, Reasons and Causes’, reprinted in Essays on Actions and Events, pp. 3-19. We return to the issue of any potential clash between causal and conceptual connection later.

7 See, e.g., Ted Honderich, ‘The Argument for Anomalous Monism’, Analysis, 42 (1982):59-64. Davidson, ‘Thinking Causes’ in John Heil and Al Mele, eds, Mental Causation, Oxford, Clarendon Press, 1993, pp. 3-17, replies to a number of versions of this criticism.

8 See especially Child, Causality, Interpretation, and the Mind, and Stephen Schiffer, Remnants of Meaning, Cambridge, Mass., MIT Press, 1987, § 6.4.

9 For a critique of the idea that beliefs about the mental states of others should be viewed as hypotheses based primarily on their behaviour, see, e.g., McDowell, ‘Criteria, Defeasibility and Knowledge’, Proceedings of the British Academy, 68 (1982):455-79. For a recent version of the last line of argument mentioned in the text, see John Haldane, ‘Folk Psychology and the Explanation of Human Behaviour’, Proceedings of the Aristotelian Society, supp. vol. 62 (1988):223-54; for some opposition that accepts the claim about inference requiring the possibility of independent characterisation while arguing that the needed independent characterisation is available, see Jackson and Pettit, ‘Some Content is Narrow’ in Heil and Mele, Mental Causation, pp. 259-82; for some opposition that denies the need for independent characterisation, see Paul Churchland, ‘Folk Psychology and the Explanation of Human Behaviour’, Proceedings of the Aristotelian Society, supp. vol. 62 (1988):209-21. The force of the point about the frequent impossibility of describing behaviour without using psychological terms like ‘annoyed’ is obscure: we can properly call the passage from ‘This behaviour is of a kind to prompt in me the judgement that the person in front of me is annoyed’ to ‘The person in front of me is annoyed‘ an inference—and, of course, we are typically more confident that behaviour is of a kind to prompt the judgement that someone is annoyed than we are of the judgement that they are in fact annoyed.



10 This point is independent of whether good opinion about others’

-107-

intentional states is in part constitutive of their intentional states. Interpretationists are interested in interpretations causally sourced in subjects’ bodies, and not, say, in tea leaves.

11 See especially Child, Causality, Interpretation, and the Mind, but see also Schiffer, Remnants of Meaning, p. 165.

12 There is a tradition, sometimes associated with Wittgenstein and sometimes with very strong versions of the view that observation is theory laden, that insists that the relevant sense of ‘behaviour’ in this discussion can only be interpreted behaviour: there is no separating the causal source of opinion about intentional states from the intentional states themselves. But then the famous problem that the same behaviour is consistent with different hypotheses about intentional nature becomes a contradiction in terms.

13 For a survey of the options, see D.M. Armstrong, Universals: An Opinionated Introduction, Boulder, Westview Press, 1989; see also Alex Oliver, ‘The Metaphysics of Properties’, Mind, 105 (1996): 1-80.

14 For references and a recent version of interpretationism, see Child, Causality, Interpretation, and the Mind.

15 David Lewis, ‘Reduction of Mind’ in Samuel Guttenplan, ed., A Companion to Philosophy of Mind, Oxford, Basil Blackwell, 1994, pp. 412-31.

16 See Tim Crane, ‘Mental Causation’, Proceedings of the Aristotelian Society, supp. vol. 69 (1995):211-36, p. 232, where he says, ‘if we believe that mental and physical states are linked by psychophysical laws—a claim which is defensible on independent grounds—then overdetermination would not be a coincidence: it would be a matter of natural law that the mental and physical causes both bring about the effect’.



17 D.H. Mellor, The Facts of Causation, London, Routledge, 1991, pp. 31-2. The view that the problem has been overplayed in the literature is also prominent in Tyler Burge, ‘Mind-Body Causation and Explanatory Practice’ in Heil and Mele, Mental Causation, pp. 97-120, but his discussion of the supervenience of the mental on the physical makes it unclear to me where he stands in relation to Crane and Mellor. Proponents of the view that the problem has been overplayed sometimes object to being labelled kinds of dualist, and not just because it is bad publicity. Their point is that they are insisting that there are many kinds of things in our world; they are pluralists. However, the crucial division is between that which supervenes on the physical and that which fails to supervene on the physical; what is crucial is whether, out of these two classes, psychology belongs in the first.

18 Davidson, ‘Knowing One’s Own Mind’, Proceedings of the American Philosophical Society, 60 (1986):441-58.

19 Colin McGinn, Mental Content, Oxford, Basil Blackwell, 1989, p. 104. 20 Michael Lockwood, Mind, Brain and the Quantum, Oxford, Basil Blackwell, 1989,

pp. 143-8, considers at a little length views of this kind. The version he especially considers is McDowell’s in ‘Singular

-108-

Thought and the Extent of Inner Space’ in Philip Pettit and John McDowell, eds, Subject, Thought, and Context, Oxford, Clarendon Press, 1986, pp. 137-68.

21 There is of course a huge literature on this objection. My comment on it is a considerable distillation of Frank Jackson, Robert Pargetter, and Elizabeth Prior, ‘Functionalism and Type-Type Identity Theories’, Philosophical Studies, 42 (1982):209-25. For a recent reply to the objection by one of the principal architects of the type-type theory, see Lewis, ‘Reduction of Mind’.

22 For a recent version of this view, see Michael Tye, Ten Problems of Consciousness, Cambridge, Mass., MIT Press, 1995, ch. 2. Of course, many hold that constitution can be analysed in terms of identity via a temporal parts metaphysics. On p. 40, Tye extends the argument from tokens to types in a way that escapes me—I am not clear what constitution comes to in the case of types. Thus I cannot appeal to it, as some would want to, to provide a further reply to McGinn’s modal argument.

23 See, e.g., Jennifer Hornsby, ‘Which Physical Events are Mental Events?’, Proceedings of the Aristotelian Society, 81 (1980): 73-92, §3.



24 See, e.g., Davidson, ‘Agency’, reprinted in Essays on Actions and Events, pp. 43-61; Alvin Goldman, A Theory of Human Action, Englewood Cliffs, N.J., Prentice Hall, 1970; and Jennifer Hornsby, Actions, London, Routledge & Kegan Paul, 1980.

25 See, e.g., Daniel Dennett, ‘True Believers’ and ‘Reflections: Real Patterns, Deeper Facts, and Empty Questions’, reprinted in The Intentional Stance, Cambridge, Mass., MIT Press, 1989, pp. 13-35 and pp. 37-42, respectively.

26 See, e.g., Christopher Peacocke, ‘Externalism and Explanation’, Proceedings of the Aristotelian Society, 93 (1993):203-30.

27 I here move quickly over an area of much debate. For more on the matter from the perspective I sketch, see Jackson and Pettit, ‘Causation in the Philosophy of Mind’, Philosophy and Phenomenological Research, 50 (supplement, Fall 1990):195-214, Jackson, ‘Mental Properties, Essentialism and Causation’, Proceedings of the Aristotelian Society, 95 (1995):253-68. Although the position sketched is widely accepted in recent discussions, see, e.g., the references in Brian McLaughlin, ‘On Davidson’s Response to the Charge of Epiphenomenalism’ in Heil and Mele, Mental Causation, there is dissent. Two examples are Davidson, ‘Thinking Causes’, and Graham MacDonald, ‘The Nature of Naturalism’, Proceedings of the Aristotelian Society, Supp. vol. 66 (1992):225-44. Davidson grants, as is very well known, that properties figure in causal explanations; his opponents respond that they figure in causal explanations by virtue of being causes. Some who are uneasy about talking of properties as causes talk instead of the properties of events in virtue of which the events cause what they cause. The rephrasing leaves the substantial issues undisturbed, or so it seems to me.

28 Although I frame the discussion to follow under the assumption that mental states are brain states, many of the issues arise independently.

-109-



For example, most of the issues appear in one form or another in McGinn, Mental Content, pp. 132ff., and as we noted earlier, he denies the identity thesis.

29 For example, D.M. Armstrong, A Materialist Theory of the Mind, London, Routledge & Kegan Paul, 1968, argues that dispositions are necessarily constituted, whereas Gilbert Ryle, The Concept of Mind, London, Hutchinson, 1949, insists that dispositions may be bare.

30 Or see the discussion in Davidson, ‘Causal Relations’, reprinted in Essays on Actions and Events, pp. 149-62.

31 This argument from Hume only directly impinges on the claim that dispositions cause their own manifestations or typical outputs, and someone who thought that they had an answer to the first argument, the redundancy argument, against dispositions (and functional properties) as causes might be tempted to allow dispositions as causes of things other than their own manifestations. Thus, some philosophers of colour identify yellow with a disposition to reflect light differentially, not with the disposition to look yellow, and their reason is that they want looking yellow to be a typical response to yellow and believe that looking yellow could not be a response to the disposition to look yellow itself. But putative cases of dispositions causing are ones where what is putatively caused is downstream from the dispositions’ manifestations—the yellow object manifests its disposition to reflect light differentially, and the differentially reflected light then causes the object to look yellow, as it might be. But if the disposition does not cause its manifestation, how can what the manifestation goes on to cause count as something caused by the disposition?

32 Pace, I take it, Ned Block’s view in ‘Can the Mind Change the World?’ in George Boolos, ed., Meaning and Method: Essays in Honor of Hilary Putnam, Cambridge, Cambridge University Press, 1990, pp. 137-70.

33 Mental Content, p. 133. 34 This is, I take it, the response of Jerry Fodor in Psychosemantics, Cambridge,

Mass., MIT Press, 1987, ch. 2, and ‘Methodological Solipsism Considered as a Research Strategy in Cognitive Psychology’, Behavioral and Brain Sciences, 3 (1980):63-109.

35 For versions of this style of objection, see McGinn, Mental Content, p. 137, and Brian Leiter and Alex Miller, ‘Mind Doesn’t Matter Yet’, Australasian Journal of Philosophy, 72 (1994):220-32.

36 See, e.g., Fodor, Psychosemantics. 37 The philosophical moral simply turns on the possibility of a converse of multiple

realisability: just as different states can realise the same functional role, so the same state can realise different functional roles. I am indebted here to dissension from Ned Block.



38 The idea that we might approach the problem posed by the causal role of mental properties by distinguishing causing proper, or causal efficacy as it is sometimes called, from a wider notion of causal relevance is prominent in, e.g., Jackson and Pettit, ‘Functionalism and Broad Content’, Mind, 97 (1988):381-400, and Fred Dretske, ‘Mental Events as Structuring Causes of Behaviour’ in Heil and Mele, Mental Causation, pp. 121-36. Crane, ‘Mental Causation’,

-110-

p. 233, suggests that views of this kind involve denying that mental and physical causation are homogeneous, and wonders whether they are really physicalist. I think this is misleading. First, the wider notion of causal relevance is not confined to relations involving mental properties. It is not special to the mental domain. And, second, the wider notion is explained in terms of a univocal concept of causation. For one instance, the causal relevance of P is analysed by Jackson and Pettit in terms of the causal efficacy, the single notion, of various ways that P might be realised.

39 See, e.g., Jaegwon Kim, ‘Epiphenomenal and Supervenient Causation,’ reprinted in Supervenience and Mind, Cambridge, Cambridge University Press, 1993; Michael Tye, The Imagery Debate, Cambridge, Mass., MIT Press, 1992, ch. 2, and Gabriel Segal and Elliott Sober, ‘The Causal Efficacy of Content’, Philosophical Studies, 63 (1991):1-30.

40 And some say that the psychological properties supervene on the neurological properties, though they do not supervene in the logical sense that we have been using ‘supervene’ for. To get supervenience in our sense, you would need a much wider supervenience base.

41 For instance, if it turns out (as some hold) that causation is essentially a matter of energy transfer, then there must be an energy transfer from mental state to movement.

42 But it does not have to, see, e.g., Jackson and Pettit, ‘Functionalism and Broad Content’.

43 The literature started with Hilary Putnam, ‘The Meaning of “Meaning”’, reprinted in Mind, Language and Reality, Cambridge, Cambridge University Press, 1975, pp. 215-71, and Tyler Burge, ‘Individualism and the Mental’, Midwest Studies in Philosophy, 5 (1979):73-122. For references to the recent literature, see Andrew Pessin, ed., The Twin Earth Chronicles, London, Paragon House, 1995.

44 See, e.g., Terence Horgan, ‘Actions, Reasons, and the Explanatory Role of Content’ in Brian McLaughlin, ed., Dretske and His Critics, Oxford, Basil Blackwell, 1991, pp. 73-101. But some, e.g., Dretske in ‘Mental Events as Structuring Causes of Behaviour’, repudiate the intuition.



45 See, e.g., McGinn, ‘The Structure of Content’ in Andrew Woodfield, ed., Thought and Object, Oxford, Clarendon Press, 1982, and Mental Content, pp. 161-2.

46 For points of this kind, see, e.g., Burge, ‘Individualism and Psychology’, Philosophical Review, 95 (1986):3-45.

47 A view of this general kind has been defended by David Lewis in a number of places, see, e.g., ‘Reduction of Mind’, and On the Plurality of Worlds, Oxford, Basil Blackwell, 1986, §1.4. See also Jackson and Pettit, ‘Functionalism and Broad Content’, and David Braddon-Mitchell and Frank Jackson, Philosophy of Mind and Cognition, Oxford, Basil Blackwell, 1996, ch. 12. Fodor, Psychosemantics, and Michael Devitt, ‘A Narrow Representational Theory of Mind’ in W.G. Lycan, ed., Mind and Cognition, Oxford, Basil Blackwell, 1990, pp. 371-98, defend a similar view, but with the

-111-

crucial difference that their narrow content is not truth conditional. Incidentally, I understand that Devitt and Fodor no longer agree with the views there expressed.

48 I have omitted the final section of the originally published version as the material is more fully covered in Chapter 9, ‘Mental causation without the language of thought’. The somewhat ‘surveyish’ quality of the paper is explained by the fact that the paper was one of Mind’s ‘state of the art’ articles.

-112-

9 MENTAL CAUSATION WITHOUT THE LANGUAGE OF THOUGHT Here is a familiar kind of story. Harry believes that there is coffee to the right and beer to the left. His belief that there is coffee to the right explains, when combined with certain desires, his moving right; his belief that there is beer to the left explains, when combined with certain other desires, his moving left. The beliefs play different causal explanatory roles. Accordingly, we need to acknowledge individual beliefs conceived of as distinct, presumably internal, states of subjects that are thus able to play distinct



causal roles—or so it seems.

I will call any doctrine of this general kind, individualism about belief. It is a pity that the word ‘individualism’ is already used for another doctrine in the philosophy of mind, but I hope that explicitly noting the possible source of confusion is the best way of guarding against it.

The language of thought hypothesis gives us a way of fleshing out individualism about belief. According to it, beliefs are, or are relations to, stored sentences of mentalese, and storings of different sentences of mentalese will be distinct states. We then account for the difference between explaining behaviour in terms of a subject’s believing that P and their believing that Q in terms of the different causal powers of the distinct structures that are the storings of the relevant sentences in the head. Likewise, we can account for interconnections between what a subject believes—the way one belief may lead to another—in terms of causal interconnections between distinct stored structures.

This chapter is concerned with what to say should it turn out that individualism is false, that nothing like the language of thought hypothesis is true. How then, if at all, can we account for the difference between causal explanations of behaviour in

-113-

terms of subjects’ believing that P and their believing that Q? This chapter sketches an answer to this large question. I will not address the related and important question of what to say about explanations of subjects’ believing that P in virtue of believing that Q—the interconnection issue—should individualism turn out to be false, but, for the record, I believe that a similar treatment applies.

9.1 Belief, maps and holograms

It will be helpful to have before us a reasonably concrete example of an alternative to the language of thought. What I am about to describe is, I take it, what many supporters of F.P. Ramsey’s idea that belief is a map by which we steer have had in mind. 1

It is, or should be, common ground that belief states are states of heads that represent things as being one way or another, and crucial to their being representational states is that there is a correspondence between the relevant



possible head states and various possible world states. Otherwise they would not count as representational states. Now one way such a correspondence can obtain is by virtue of more piecemeal correspondences obtaining. Some part of a belief state might play a distinguished role in settling some part of how it represents things as being: there is, as it might be, a part of a subject’s belief state that is specially responsible for how she takes things to be in South America. By disturbing that part of her belief state we could change what she believes about South America while leaving everything else much the same.

The part of the belief state that has special responsibility for South America need not be localised. It might be spread through the state. Think of the part of a book responsible for what the book says about South America. The part might be made up of sentences and words scattered throughout the book. It is still the case that there is a part, a scattered part, that has special responsibility for South America.

However, the correspondence between what does the representing and what it represents might be radically holistic. Holograms provide an example. (Or so I understand—it is not crucial that what I am about to say be the truth of the matter, only that it might be.)

Holograms are ‘laser photographs’. When light from the laser is projected through the negative, the well-known, three-dimensional,

-114-

coloured array is produced. The negative can be thought of as representing things as being the way the coloured array depicts them. However, no part of the negative has special responsibility for some part of the array, and so for some part of how things are represented as being. Each part contains information about the whole array. In consequence, what happens if you damage a part (local or scattered) is a loss of detail, a blurring, of the three-dimensional array, not a loss in any particular part of the array.

This is not an example of representation without structure. We could have systems of representation that involved arbitrary correspondences between a finite number of representing states and a finite number of represented possibilities. The past use of flags to represent diseases in sailing ships may be an actual example. (Or close to one: the choice of a yellow flag for yellow fever was not totally arbitrary, obviously.) The correspondences would be arbitrary in that there would be no right way to go on when presented with new cases. We would have a correspondence between state S1



and possibility2 and P2,…and Sn and Pn, but there would be no answer to what Sn+1 represents flowing from this information. Hologramic representation, however, is not like this. There is a right way to go on. The same goes for maps and diagrams, of course. In all these cases, structural connections between what does the representing correspond to structural connections between what is represented. This is how a new map or hologram negative can be created to represent a totally new situation.

The fact that maps and hologram negatives are structured representations means that they possess the desirable properties of systematicity and productivity. It is a commonplace that a map (or globe) which can represent that more of the southern hemisphere than of the northern hemisphere is covered by water can also represent (falsely) that more of the northern hemisphere than of the southern hemisphere is covered by water. Likewise, maps can represent indefinitely many new possibilities: before the discovery of Australia, people speculated in maps as well as in words about the great southern continent.

It seems to me that our heads might represent in the way maps or holograms represent. It might, for all we now know, be the case that we find our way around the world by virtue of having an internal structure that represents how things are in broadly the way that maps or negatives of holograms do. If this is the case, then when Jones believes that snow is white, there is no individual structure in

-115-

her head that has assigned to it the job of representing that snow is white. Rather, inside Jones there is a structure that constitutes, as we might put it following David Lewis and intentional systems’ theory, a system of belief that represents, among a great many things, that snow is white: Jones believes that snow is white by having a system of belief according to which snow is white. 3 We can still talk of individual beliefs in the sense of aspects of the way heads represent things which are captureable in individual sentences, but there will be no underlying individual states that these belief sentences report. We might put it by saying that there are individual belief contents but no individual beliefs. A similar point is sometimes made in terms of maps. A map of the Earth might represent that the taller mountains are mostly near the deeper oceans, but there is no part of the map that says just that in the way that there may be a sentence that says just that—for instance, the sentence you heard (or saw) a moment ago. 4 The same goes for hologramic representation except that it is more radically holistic than map representation. In the case of maps, we can match up changes to parts of what does the representing to changes in parts of how things



are represented as being. Rubbing out one part of a map typically has especially drastic effects on what the map says about some particular part of the world, whereas damaging a part of a hologram negative blurs the whole hologram. We can now sharpen our initial question. If when Harry believes that there is coffee to the right, he has a head according to which there is coffee to the right, but there is nothing in his head which is the belief that there is coffee to the right, how do we make sense of causal explanations that advert to his believing that there is coffee to the right? Even if we reject explanations of behaviour by individual beliefs, we had better allow for explanations by individual belief contents. Our question is whether and how the radical anti-individualist can do this. But before we address it directly, it will help to have a more detailed account of the view of belief that reflection on maps and holograms suggests.

9.2 Radical holism

I will call the following view of belief radical holism. 5 It is not an analysis of belief in the traditional sense, for it rests on the empirical, contingent hypothesis that heads represent the way

-116-

things are in somewhat the way hologram negatives and maps do. But it is a piece of analysis in the sense that it delivers a truth-making story. The view tells us how a certain, possibly true account of how heads represent the way things are would make true certain claims about belief.

The first tenet of radical holism is that there are no individual beliefs. There are only systems of belief. We can say that Jones’s belief that snow is white is true but her belief that all swans are white is false. But what makes such a claim true is her having a system of belief according to which snow is white, and a system according to which all swans are white, and its being the case that the first claim is true and the second is false. 6 In the same way, a map might represent that more of the southern hemisphere than of the northern hemisphere is covered by water, and that the largest island is in the northern hemisphere. The first claim would be true and the second false. But for both head and map, it is false that there are two distinct parts, one of which represents truly and the other falsely.

Second, radical holism takes as fundamental not just the notion of a system of belief but that of a rich system of belief: a system that says a lot; a system that represents



how things are in considerable detail, including as a tiny part that, say, snow is white. This is not an accidental feature of radical holism. It is a consequence of the way radical holism approaches the problem of content.

In the case of language of thought views, a strategy worth serious attention is that of giving the content of belief in terms of causal links, or patterns of co-variation, or something along these general lines, between parts of individual beliefs and parts of the world. The belief that snow is white, for example, might be held to be about snow by virtue of its ‘snow’ part having the right kind of connection to snow. And, of course, this story can be supplemented by including as contributors to content interconnections between mentalese sentences and their parts, as in conceptual role semantics. Strategies of this kind are not available to radical holists. They must instead assign content, at least in large part, behaviourally. It will be the actual and potential way that the relevant head structure steers Jones around a putative world that is crucial for the content of that head structure, and so for the content of her system of belief. Behavioural approaches to content fit nicely with recent emphases on informational approaches to content. The head would have a hard time moving

-117-

the body through the world unless it recorded information about the world, unless, that is, its states were apt to co-vary systematically with its surroundings. (Also, those who want to can allow a role for evolutionary history in the determination of content provided that they retain a major place for behaviour.) 7

Now behavioural approaches to content inevitably deliver rich content. There is no behaviour that the belief that there is a mine near that tree as such points to. It is instead a rich system of belief to the effect, say, that there is a mine near the tree, that the mine is likely to be triggered by going near it, that moving one’s legs in such and such a way will or will not bring one near the tree, that there is not a bigger mine that can only be avoided by going close to the tree, that triggering mines tends to cause death, and so on and so forth, along with the appropriate desires that points to behaviour.

The problem is not merely that the behaviour subjects are disposed to manifest depends on a big part of how they take things to be. The way a particle moves under the influence of massive bodies depends on the totality of the component gravitational forces acting on it. Nevertheless, for each force, there is a distinctive contribution specifiable in terms of the motion the particle would undergo if that were the one and only force acting on it—a contribution that is explicitly represented in the



familiar diagrams for calculating resultant forces. Again, the action of a poison can be blocked by an antidote; nevertheless, there is a result that taking a poison typically points towards: the existence of antidotes simply means that typically pointing towards is not the same as ensuring. However, there is no such thing as the distinctive behaviour associated with too small an account of how an agent takes things to be or wants them to be. Jones’s belief that there is a mine over there has no direction whatever in and of itself.

Third, radical holism holds that there is no interesting distinction between core and derived beliefs. The problem of clutter means that language of thought theories must distinguish core from derived beliefs. If we allowed an internal sentence for everything Jones believes on an inclusive sense of ‘belief’, we could not fit the sentences into a head. Accordingly, language of thought theories only treat a small set of beliefs, the core beliefs, in terms of encoded sentences, and offer a separate treatment for beliefs which, it is argued, are in some sense derivative on the core beliefs. 8 For radical holism, however, there is no fundamen-

-118-

tally interesting distinction between core and derived belief. Although there are natural ways of carving up passages of prose at their representational joints, there are no natural ways of carving up maps or hologram negatives at their representational joints. For instance:

Jones lives in the north of England. The north of England is colder than the south of England

represents inter alia that Jones lives in the colder half of England. But we extract this content by cutting across the more natural way of itemising what the passage says. By way of contrast, consider a map of the world that represents the largest island as being in the hemisphere with the most water, the hemisphere with the most water as being the southern hemisphere, and the largest island as being the smallest continent. It equally represents the largest island as being in the southern hemisphere, the smallest continent as being in the hemisphere with the most water, and the largest island as being the smallest continent. There is no sense in which one way of itemising its content carves nearer to the joints than the other, or any other. At most, there is a reporting-relative distinction. Some things a map represents are more easily reported on in language by someone using the map than are others. A quick look enables us to say that a map of the world represents the southern hemisphere as having more water than the northern hemisphere; it takes more study to report that



it represents the largest island as being the smallest continent.

Fourth, the fact that there is a reporting-relative distinction to be drawn means that radical holism can distinguish a more inclusive from a less inclusive sense in which we may believe that P. According to radical holism, what makes it true that S believes that P is that S has a system of belief according to which P. This is the inclusive sense. But there is, I take, a pressure in our ordinary talk to tie belief that P to the ability to produce a relevant sentence on request—or at least there is when we focus on belief in humans rather than belief in animals. Radical holism can allow that there is a more exclusive sense of ‘belief in which an ability to produce the right kind of sentence is called for. The underlying representational reality is that there is a rich system that takes things to be thus and so, but different aspects of this rich content may vary in, as we might put it, how near they are to the linguistic surface. And anyone who wishes to is free to restrict

-119-

their use of the word ‘belief’ to aspects that are near to that surface.

Finally, we should note that any view based on the idea that the way belief represents is like the way maps or holograms represent cannot be accused of representational incompetence. It is sometimes suggested that maps and the like are good for representing geographical facts but not for representing the way house prices vary over time, or, more generally, the kinds of facts we typically use numbers to capture; nor are they good for representing rather abstract, theoretical beliefs. 9 However, we can be confident that nothing escapes the representational powers of a map-like structure.

This is because everything perfectly represents itself. A book or a film or a set of diagrams may get one or another detail about a battle wrong, but the battle itself gets it exactly right, down to where each and every soldier is located and when each and every order was made. Likewise, the world itself represents the way things are perfectly. It gets everything there is to get right exactly right down to the last sub-atomic particle. But the world itself is a huge four-dimensional map-like structure. So we know that there is nothing a map-like thing cannot get right.

9.3 Causal explanation by appeal to dispositions

The story I will tell about how radical holism accounts for the evident difference



between causally explaining behaviour by reference to believing that P and by reference to believing that Q arises from a certain view about explanation by appeal to dispositional properties. So this section of the chapter is about explanation by appeal to dispositional properties. It is, of necessity, short and somewhat dogmatic. 10

Properties stand in causal relation. The film The Way We Were is about how the way its protagonists were in their youth led to how they became in middle age. But the way its protagonists were is a matter of the properties they had back then, and what they became in middle age is a matter of the properties they acquired in middle age. The film makers simply took it for granted that properties can be causes. And so do we all. The remark that when the US sneezes the rest of the world gets a cold is nothing other than a claim that how things are in the US causally affects how the rest of the world is. I am, of course, aware that many insist that it is general causal laws that relate

-120-

properties, and that singular causal statements connect events, thought of as quite distinct from properties. But surely general causal laws deliver generalisations about what does or would cause what. It thus escapes me how general causal laws could be about properties without singular causal statements also being about properties. Likewise, I find strange the objection that properties pertain to causal explanation and not singular causation. To explain causally is to tell about what caused what, and so if explanation involves properties, as surely it does, that is support for properties being causes.

There remains the question of how to integrate the property view with the familiar events’ framework. Perhaps the best strategy is to think of events as property-like and view the term ‘event’ as reminding us that it is properties qua instances that cause. 11 In any case, I will talk of properties and state types as causes.

Nevertheless, although causes are properties, not all properties are causes. In particular, dispositions do no causing. Their categorical bases do all the causing. 12 However, it would be wrong to infer that we cannot give causal explanations in terms of dispositions. For example, Richard Feynman explained the Challenger disaster in terms of the loss of elasticity at low temperatures of an O-ring. What then are we doing when we explain in terms of dispositions?



It is tempting to say that a dispositional property explains something if and only if its basis causes it. There are, however, clear counter-examples to this simple answer. Here is one due to David Lewis. (I will simplify the science but in a way which does not damage the philosophical moral.) Electrical and thermal conductivity in metals have the same basis—the way free electrons permeate certain metals is responsible for the separate input-output profiles distinctive of these two dispositional properties. Nevertheless, an explanation in terms of a metal’s electrical conductivity is not the same as one in terms of its thermal conductivity: good results from cooking in copper-based saucepans are explained by copper’s high thermal conductivity, not by its high electrical conductivity; the coming on of a light is explained by a connecting wire’s electrical conductivity, not by its thermal conductivity.

Peter Menzies holds that we should account for the difference in these causal explanations by viewing the different dispositions as different causes. Thus he holds that we should abandon the

-121-

doctrine that dispositions are not causes; 13 for if they are different causes, they are ipso facto causes. I admit the prima facie force of this argument but reply by insisting that it is outweighed by the case against dispositions being causes.

I say instead that causal explanations that appeal to dispositional properties have a dual character. When you explain by citing a dispositional property, you make two claims together. First, you say that what was caused was caused by the disposition’s categorical basis. If the basis of the O-ring’s loss of elasticity at low temperatures did not cause the Challenger disaster, Feynman was wrong. Second, you say that the kind of inputs and outputs distinctive of the disposition in question lie in the causal path to what happened. 14 One part of what Feynman was saying was that the basis of the loss of elasticity caused the disaster, and the other, more interesting part was that what happened resulted from the kind of output distinctive of a loss of elasticity: it was the failure of the O-ring to recover from deformation at a crucial point that led to the disaster. Or consider the difference between explaining a death in terms of a metal’s thermal conductivity as opposed to its electrical conductivity. Although the operative property of the metal will be the same in each case, in the first it will, say, have caused, as it might be, a surge of heat that burnt someone to death, and, in the second, the death will be, say, by electrocution. In each case, the causally operative internal property of the metal will be the same; the differences lie in how this basis interacts with different inputs to cause different outputs, and the right disposition to cite in a causal explanation will be settled by the nature of the inputs and outputs.



9.4 Explanation by individual belief contents (back to the main plot)

We started by noting that we might explain one bit of behaviour by reference to Harry’s believing that there is coffee to the right, and another by reference to his believing that there is beer to the left. Radical holism cannot account for the difference by citing different individual beliefs. There are no such animals as individual beliefs according to radical holism. But the lesson of explanation by appeal to dispositions is that we can have distinct causal explanations without having distinct causes. The explanation of Harry’s right-moving behaviour by reference to his believing that there is coffee to the right can be importantly distinct from the

-122-

explanation of his left-moving behaviour by reference to his believing that there is beer to the left—as it had better be—without supposing that the two explanations involve distinct individual beliefs. Explanation by distinct belief contents need not be explanation by distinct individual beliefs. Both bits of behaviour can be caused by the one system of belief and yet the explanations be distinct. For radical holism must, we saw, take an essentially behavioural approach to content. The content of a system of belief is, somehow or other, the product of the total story about how the system moves a body through the world. 15 In particular, there will be the inputs and outputs that (along with other factors) make it true that Harry’s system of belief represents coffee to the right, and the distinct inputs and outputs that make it true that Harry’s system represents beer to the left. Radical holists can and should see the distinctness of explanations by appeal to different individual belief contents as deriving from the distinctness of the input-output pairings associated with the different contents. In particular, what makes it true according to radical holism that Harry’s right-moving behaviour is explained by his believing that there is coffee to the right is the conjunction of: (a) his right-moving behaviour is caused by how his system of belief represents things as being; (b) how his system of belief represents things as being includes that there is coffee to the right; and (c) his right-moving behaviour is, or is caused by, part of the pattern of inputs and outputs that makes it true that his system of belief represents that there is coffee to the right. The difference between an explanation in terms of his believing that there is coffee to the right, and one in terms of his believing that there is beer to the left is now transparent. It lies in the difference between the pattern of inputs and outputs responsible for his system of belief representing that there is coffee to the right, as opposed to the pattern responsible for



the very same system representing that there is beer to the left. It lies, that is, in the same place as the differences between explanations in terms of the thermal and electrical conductivity of metals. What causes Harry’s right-moving and what causes his left-moving behaviour is the very same complex internal structure according to radical holism; the difference is in whether what gets caused is, or is caused by, something definitive of the structure representing inter alia that there is coffee to the right, or its representing that there is beer to the left.

There are many questions that might be asked about this story on behalf of radical holism. I will conclude by offering answers to

-123-

two of the more obvious ones. The first concerns the type-type identity theory of mind; the second concerns certain overdetermination cases.

9.5 Radical holism and the type-type identity theory

Although radical holism must, it seems to me, give some kind of dispositional cum functional treatment of content, the most plausible version does not treat the content of a subject’s system of belief as a dispositional property. For then content would not be a cause, and I—and thousands—take it to be obvious that how a subject takes things to be literally moves that subject around the world. True, I must admit that the observation that dispositional properties are not causes opens up an interesting alternative position on content. It is obvious that dispositional properties can causally explain. This is so whether or not you accept the two-stranded account I gave of how they do so. We must, therefore, acknowledge a notion of causal relevance that includes, but goes beyond, causal efficacy or production or, as we are calling it most simply, causing. Dispositional properties are causally relevant to various outcomes without actually causing them. In putting it this way, I am taking a stand against the view that if dispositional and functional properties are not causes then they are, as Ned Block, for instance, sometimes puts it, epiphenomenal. 16 As I put it in Chapter 8, ‘Mental Causation’, §8.4.5, to be epiphenomenal is to be outside the causal story about how things come about. Dispositions are not epiphenomenal in this sense. The causal story about how a fragile glass comes to break includes its fragility: by the time you have explained how the glass’s categorical nature gets together with the impact to cause breaking, you have said inter alia that the glass is fragile. Although dispositional properties do not cause, they are an ineliminable and integral part of many stories about what does cause. This is why they should be counted among the



potentially causally relevant properties of one or another happening, including behaviour.

The interesting, alternative position on content then is that it is causally relevant to behaviour without actually causing behaviour. All the same, I think we should regard this position as a fall-back one. The clear intuition is that the content of belief drives behaviour. 17 If this is right, belief content is a neurophysiological

-124-

property that drives behaviour and gets to be belief content, and gets to be the belief content that it is in virtue of how it does the driving—in virtue, that is, of how it moves a subject around the world. Thus, radical holism is best thought of, it seems to me, as a style of type-type identity theory. But instead of making a claim like ‘Belief that there is beer to the left in Harry=neuro-state B’ it makes a claim like ‘A rich taking of things to be thus and so in Harry=huge, complex neuro-state N’. There are no identity claims for individual beliefs for the good reason that there are no individual beliefs according to radical holism.

The identity claims made by radical holism can only be contingently true. Although radical holism is very different from more individualistic views of belief, the powerful case for multiple readability applies equally to both. In consequence, it must be allowed that Harry’s taking things to be thus and so might have been a different neuro-state, or indeed not a neuro-state at all, just as the most dangerous poison might have been a different structure from the one it in fact is. The obvious question, then, is what I say against common objections to contingently true type-type mind-brain identities. In outline—all that is possible here—I say the following. 18

True, the early identity theorists offered as illustrations of what they had in mind when they said that mental states are brain states, scientific identities like ‘Lightning=electrical discharge’, identities that are now widely held to be necessary a posteriori. But that was because it was at that time widely believed that these identity sentences are contingent. They could have made the points they wished to make in terms of indubitably contingent type-type identity statements like ‘The most dangerous poison for humans=plutonium’ (which is, I understand, true on some ways of measuring toxicity).

True, many early identity theorists left out qualifications like ‘in Harry’ that I included in the two examples just given, and this gave the impression that it was an essential part of their view that the same neuro-kind was the same mental state type in



everything. It may even be that some early identity theorists thought that this was an essential feature of their view. But it was not and is not.

True, many on being presented with examples of what seem to me—and thousands—to be clear examples of contingent type-type identities respond by saying that we must be reading phrases like ‘the most dangerous poison’ as definite descriptions, and

-125-

Russell teaches us that definite descriptions are not singular terms. But, they then say, the examples cannot really be identity statements. This seems to me to reflect an unfortunate terminological decision in the philosophy of language. ‘Singular term’ is a good phrase for a bunch of words intended by the conventions of the language to apply to one thing, as W.V. Quine says; 19 and this is true independently of the success of Russellian accounts of definite descriptions. But, in any case, at most what is shown is that type-type identity theorists chose a bad name for their theory.

9.6 The Ramsey, Stich and Garron (RSG) question

William Ramsey et al. have pointed out that the following kind of situation is possible. 20 Harry believes that P.Harry believes that Q.Harry acts in a way that might be rationalised by reference to his believing that P, but equally it might be rationalised by his believing that Q. But only his believing that P actually causally explains what he does. Perhaps Harry wants access to a phone and also to a computer. He believes, truly let’s suppose, that the best way to access a phone is to enter room 107, and that the best way to access a computer is to enter room 107. He intentionally enters room 107. Clearly, either belief rationalises his entering the room, but consistently with this, it might be that his belief that entering room 107 is the best way to access a computer is what in fact causally explains his action.

Let’s call a situation of this kind a RSG situation. They are clearly possible. The RSG question for any theory of belief is how it accounts for the possibility of RSG situations.

The individualists’ answer to the RSG question for our example is that Harry’s belief that entering room 107 is the best way to access a phone is a different state from his belief that entering room 107 is the best way of accessing a computer; where what is meant is not just that the contents entering room 107 is the best way of accessing a



phone and entering room 107 is the best way of accessing a computer are distinct, but that there are distinct states with those contents. And, as there are two states, it is possible that one and not the other does the causing. Radical holism cannot give this answer.

Radical holism has to answer the RSG question by insisting that the difference between Harry’s action being explained by his believing that entering room 107 is the best way of accessing a

-126-

phone and its being explained by his believing that entering room 107 is the best way of accessing a computer will show up, somehow or other, in the relevant, actual or possible, inputs and outputs. If, as we supposed, it is his believing that entering room 107 is the best way of accessing a computer which explains his entering the room, his body will be oriented towards where he takes the computer to be in the room rather than where he takes the phone to be; or perhaps his movements will be that bit more sensitive to information concerning computers than to information concerning phones; or whatever. A fair and obvious question is whether we can describe a RSG case where there is no relevant difference whatever in terms of inputs and outputs. Perhaps, it might be urged, there could be a determinate fact of the matter as to whether Harry’s behaviour is causally explained by the computer belief content or by the phone belief content, and yet there be no difference in actual and possible behavioural responses. I think the jury is still out on this possibility and that the verdict it will return will depend on how representation in the head takes place. The possibility is posterior to whether individualism or holism is the right story about mental representation. It is not a prior datum with which to assess theories of mental representation. 21

Notes 1 A view I learnt about from D.M. Armstrong, Belief, Truth and Knowledge,

Cambridge, Cambridge University Press, 1973, part I. 2 Possibility, not possible world, as the state will not represent how things are in

full detail. The metaphysics of possibilities (and possible worlds) is an important issue but one we can fairly set aside here. There had better be some way talking about representational content, and the role of possibilities in this chapter is to be that way.

3 See, e.g., David Lewis, On the Plurality of Worlds, Oxford, Basil Blackwell, 1986, §1.4.



4 The maps we use often contain words, and to the extent that they do, they represent in the way language represents. In what follows it is perhaps best to think of ‘wordless’ maps.

5 Not to be confused with holistic doctrines about how some system of representation—belief, language or whatever—gets to represent what it does represent.

6 As Lewis and Robert Stalnaker have emphasised, it is no part of the systems’ view that there is at most one system for each head. Thus, we need to distinguish two ways we might give an account of when S believes that P: we might say that S believes that P if every system of belief in S’s head is one according to which P; or we might say that S believes that P if some system of belief in S’s head is one according to

-127-

which P.See the discussion in On the Plurality of Worlds, p. 31, and ch. 5 of Robert Stalnaker, Inquiry, Cambridge, Mass., MIT Press, 1984. We will not be concerned with this complication in what follows.

7 For why I would not want to do this, see David Braddon-Mitchell and Frank Jackson, ‘The Teleological Theory of Content’, Australasian Journal of Philosophy, 75 (1997):474-89.

8 See, e.g., Hartry Field, ‘Mental Representation’ in Ned Block, ed., Readings in Philosophical Psychology, vol. II, Cambridge, Mass., Harvard University Press, 1981, pp. 78-114, and Jerry Fodor, Psychosemantics, Cambridge, Mass., MIT Press, 1987, ch. 1.

9 For instance, Stephen Stich, The Fragmentation of Reason, Cambridge, Mass., MIT Press, 1990, p. 109, asks—in a highly sceptical tone—how the metaphor of beliefs as maps of extra-mental reality can handle examples like ‘I believe that if Robert Kennedy had not been assassinated, he would have become President of the United States’, and ‘I believe that we do not now have a good theory of consciousness’. However, the way he words his criticism suggests that he may be thinking of the map analogy as illuminating the topic of individual beliefs, whereas in fact it is put forward (here anyway) as a way of thinking about systems of belief.

10 What follows is even shorter than in the original publication as the material is covered in Chapter 8, ‘Mental causation’.

11 I am here indebted to Graham MacDonald’s dissent.



12 Or at least they do when they are there to do the causing. We will not be concerned with ‘bare’ dispositions. Perhaps they are possible, perhaps not. It does not matter for anything I say here. For reasons why it is the categorical bases that do the causing, see §8.4.4 of Chapter 8, ‘Mental causation’.

13 Peter Menzies, ‘Against Causal Reductionism’, Mind, 97 (1988): 551-74. 14 We can think of cases where the distinctive outputs are what is being explained

as a special case: we want fragility to explain (albeit in a rather uninformative way) breaking as well as matters downstream from breaking.

15 Part of the story about how the system moves the body through the world will be that it does not do it the way a blockhead does. If it did, there would be no content at all. But I set aside the important question of precisely how radical holism should rule out blockhead. Also, we could, as noted earlier, insist that the history of how structures of the kind in question moved bodies around in the past is important if we wished to incorporate proper function considerations into the story about content.

16 Ned Block, ‘Can the Mind Change the World’ in G. Boolos, ed., Meaning and Method: Essays in Honor of Hilary Putnam, Cambridge, Cambridge University Press, 1990.

17 In Frank Jackson and Philip Pettit, ‘Functionalism and Broad Content’, Mind, 97 (1988):381-400, we are neutral as between the view that content drives behaviour and the view that it is causally relevant to behaviour without driving it. My reasons for now pre-

-128-

ferring the driving position are spelt out in a little more detail in ‘Mental Properties, Essentialism, and Causation’, Proceedings of the Aristotelian Society, 95 (1995):253-68; see also the last few paragraphs of §8.4.5 of Chapter 8, ‘Mental causation’.

18 For a fuller account, see David Braddon-Mitchell and Frank Jackson, Philosophy of Mind and Cognition, Oxford, Basil Blackwell, 1996, ch. 6.

19 W.V. Quine, Word and Object, Cambridge, Mass., MIT Press, 1960, §20. 20 William Ramsey, Stephen Stich and Joseph Garon, ‘Connectionism,

Eliminativism and the Future of Folk Psychology’, Philosophical Perspectives, 4 (1990):499-533.

21 I am much indebted to philosophical exchanges, some supporting and some dissenting, with Michael Devitt, Georges Rey, Karen Neander, and, most especially, David Braddon-Mitchell and David Lewis.

-129-




-130-

Part III METHOD IN METAPHYSICS

-131-


-132-

10 METAPHYSICS BY POSSIBLE CASES 10.1 Introduction

We often do metaphysics by intuitions about possible cases. An example is the argument for functionalism about belief and desire. The argument starts from the premise that, intuitively, it is not possible for belief and desire to vary independently of functional nature (spelt out in some suitable way)—functional duplicates are necessarily belief-desire duplicates—and concludes that belief and desire are functional states. An equally famous example is the argument against functionalism for sensory qualities. The argument starts from the premise that, intuitively, it is possible for sensory nature to vary independently of functional nature—functional duplicates might be sensory inverts, for example—and concludes that sensory states are not functional states.



My primary concern in this chapter is with a simple and influential example of metaphysics by possible cases which proceeds from an intuition about laws and flukes to the conclusion that we have to acknowledge something radically non-Humean, some kind of necessary connection, something that cannot be captured in terms of regularities or sophistications of regularities, in our picture of the world. The argument can be set out as follows.

Pr. 1

Any pattern of matters of particular fact, no matter how all-encompassing and no matter how great its display of the virtues we associate with law-hood like simplicity and strength, could be an entirely flukey matter. (Intuition about possibility)

Pr. 2 Being entirely flukey is inconsistent with exemplifying a law. (Common ground)

-133-

Con. Being a law is not identical with exemplifying any pattern, no matter how all-encompassing and no matter how great its display of the virtues we associate with law-hood like simplicity and strength.

Premise 1 expresses a widely shared intuition—one, indeed, that is more widely shared than the intuitions we adverted to concerning functionalism. It is the intuition, as it applies to the patterns manifested in our world, that the Hume world—the world where nothing is causally connected to anything, and nothing happens by law, and yet the passing parade of matters of particular fact is exactly as it is in our world—is a possible world. 1 Premise 2 is, as noted, common ground. The conclusion, when combined with the fact that there are laws, yields the overall conclusion that we need to supplement our picture of what there is with some non-Humean ingredients. This chapter is about why the argument fails, or so I will argue. Although the argumentation will centre on the argument from the Hume world, I will note at the end the implications of the discussion for a number of arguments of the same general type as our target, directed to conclusions about motion and ungrounded identity, and personal identity.

The problem I see for the argument arises from a certain view of what is going on



when we do conceptual analysis, or at least conceptual analysis of concepts like law-hood, personal identity, and change. If this view is right, there are certain circumstances, to be specified later, in which intuitions about possibilities need to be treated with considerable caution. Most of the chapter will be directed to this lemma about what is going on when we do conceptual analysis. But, first, there are two preliminary matters to address.

Are there Ks? is a distinct question from, What does it take for there to be Ks? Even if conceptual analysis delivers that, say, in order for there to be laws there must be more than Hume allowed—there must, we might suppose, be relations of necessitation between universals as David Armstrong, Michael Tooley and Fred Dretske have argued 2 —it does not follow that there actually are relations of necessitation between universals. Perhaps the right conclusion is that there are no laws. 3 It might be urged that the better the philosophical arguments for the necessitation analysis of laws, the better the case for scepticism about laws. After all, on the analysis, the relations of necessitation do no causal work, so nothing counts as a sign of their presence.

-134-

It is tempting to dismiss this worry. There have always been philosophers of a sceptical bent, but for most of us it is a Moorean fact that there are laws, a fact more obvious than most things philosophers say qua philosophers. This attitude need not be a mind-less rejection of scepticism as a topic for serious attention, but rather an insistence that when scepticism is not the topic under discussion, we can take such matters as that there are laws; that things, including persons, remain numerically the same through time; that things move, and so on and so forth, for granted. I think, nevertheless, that this dismissal would be a mistake. If conceptual analysis is claimed to reveal that belief in laws requires believing in relations of necessitation between universals, then surely it is open to us to deny that there are any laws according to the official concept as explicated, while granting that there are laws on a Humean regularity account, or some refinement of it. Again, if conceptual analysis reveals that belief in identity through time requires believing in ungrounded identities through time, identities which raise at least prima facie worries concerning how we are supposed to detect them, then surely it is open to us to deny that there are any identities through time according to the official concept as explicated, while granting there are identities through time according to, for example, W.V. Quine’s, J.J.C. Smart’s and David Lewis’s concept. 4 There is, it seems to me, a real problem in reconciling the Moorean nature of, for instance, the existence of laws and the existence of identity through time with the unMoorean things philosophers often say about laws and



identity through time. A noteworthy feature of the account of analysis I will be defending is that it explains how it can be Moorean that there are laws, and that there is identity through time, at the same time as it is true that what makes something a law, or what makes something the very same thing through a period of time, is a highly controversial matter and, in consequence, a decidedly unMoorean one.

The second preliminary concerns a quick and dismissive objection to the very idea of metaphysics by intuitions about possible cases. The quick and dismissive objection urges that intuitions about possible cases are doubly irrelevant to metaphysics—because they concern possibilities, and because they are intuitions. Metaphysics is about how things actually are, and it had better have a more secure grounding than intuition.

This sounds admirably tough-minded but is, I think, profoundly mistaken. The question, Are there Ks? just is the conjunction of

-135-

two questions: What does it take (or, what needs to be the case) for there to be Ks? and Is what it takes in fact the case? There are motor cars in the twentieth century because what it takes for there to be motor cars in the twentieth century is the case; there were no motor cars in the seventeenth century because what it takes for there to be motor cars was not the case during the seventeenth century. Moreover, the question: What does it take for there to be Ks? just is the question: Under what conditions do we have a K, and under what conditions do we fail to have a K? That is, it is a question about possibilities. It follows that when addressing the question of what it takes for there to be Ks, we have no alternative but to be guided by our intuitions about whether possible cases, described in some set of terms or other, are or are not cases where there are Ks. For what else should we do? Be guided by intuitions about cases described in no terms at all? That is impossible. Say what is counter-intuitive? That is absurd. There is, I think, nothing wrong in principle and in general with giving intuitions about possible cases a prominent role when we do metaphysics. The objection we will be raising to the argument from the Hume world is not an objection to the very idea of appealing to intuitions about possible cases when we do metaphysics, but rather an objection to the effect that we need to treat our intuitions very carefully in certain cases.

As heralded, we need first to address the question of what is going on when we do conceptual analysis. I now turn to this task. Rather than starting with a characterisation of the view in the abstract, I am going to approach it via a discussion



of an actual example: the ongoing debate between three-dimensionalism and four-dimensionalism about change and identity over time. I take it that the terms of this debate are by now well understood, and this makes it a good example for our purposes. The next section of this chapter, accordingly, is about the debate between three-dimensionalism and four-dimensionalism; and although my purpose in entering that debate is to enable me to draw some general morals about analysis for subsequent application to our target in §10.3, what I say in §10.2 is largely independent of the use I make of it in §10.3. You could agree with what I say in §10.2 while disagreeing with much of what I say in §10.3.

-136-

10.2 Three-dimensionalism and four-dimensionalism

I am watching a strongman at a circus bend a poker. I take it that I am watching numerically the same object, the very same poker, throughout, and that this object changes as I observe it. What features of the world am I picking out when I say—aloud, or to myself—that it is the same poker, and that it is changing? We might look to the world for an answer. After all it is some way the world is which is prompting the thought that it is the same poker throughout, and that it is changing. On this approach, it would be an a posteriori matter what constitutes its remaining the (self-)same poker, and what constitutes its changing. It would be a matter for total science. It would not of course be a totally a posteriori matter. Not any old property of the poker which prompted the thought that it was changing would thereby count as the poker changing. For instance, if the density of the poker prompted the evil demon to cause in me the experience I describe as seeing the poker’s shape change, that would not make density into change. There are a priori constraints on which features of our world could count as change and remaining the same object. We know a priori that density is neither identity through time nor change. If you think that change might turn out to be density, your understanding of change and density is thereby defective, or anyway it is not change and density as understood by us.

The alternative view is that what constitutes change and what constitutes remaining numerically the same is an entirely a priori matter. Our very understanding of phrases like ‘remained the self-same object throughout’ and ‘underwent a change in shape’ determines which properties we are picking out, or at least purport to be picking out, when we use these phrases. I will defend the view that it is an a posteriori matter (in part) what constitutes change and what constitutes remaining the same object: there are, in particular, two prime candidates for what change might turn out to be, and two prime candidates for what remaining numerically the same object might turn out to



be, and the choice between them is an a posteriori one.

Before we start on the argument for this view, it will be helpful to have some now more or less standard terminology before us. 5 We have been talking of an object remaining the same object throughout a period of time. It is convenient to describe this as an object persisting through time. Our poker, I took it, persisted

-137-

throughout the period it changed from being straight to being bent. The dispute between three-dimensionalism and four-dimensionalism, or more precisely, that part of the dispute we will be concerned with, concerns what persistence, and correlatively, what change, comes to. Three-dimensionalism holds that an object exists at a time by being wholly present at that time, and, accordingly, that it persists if it is wholly present at more than one time. For short, it persists by enduring. Four-dimensionalism holds that an object exists at a time by having a temporal part at that time, and it persists if it has distinct temporal parts at more than one time. For short, it persists by perduring.

We have change, or change over time, if we have a persisting object with a given attribute at one time, and a distinct attribute at another time. 6 Thus, we have change according to three-dimensionalism just if we have an enduring object which has different attributes at different times; and we have change according to four-dimensionalism just if we have a perduring object which has different attributes at different times. In other words, we have change according to three-dimensionalism—call it ‘3D change’—just if we have an object with different attributes at different times at which it is wholly present; and we have change according to four-dimensionalism—call it ‘4D change’—just if we have distinct temporal parts of a single temporally extended object possessing different attributes.

My contention about the debate between four-dimensionalism and three-dimensionalism can now be put thus: whether change is 3D change or 4D change, or, equivalently, whether persistence is endurance or perdurance, is an a posteriori matter. The usual opinion appears to be precisely the opposite. Many three-dimensionalists declare the notion of a temporal part of an object to be unintelligible. And they argue that 4D change is not change properly speaking. Thus P.T. Geach urges that on the four-dimensionalists’ view ‘the variation of a poker’s temperature with time would simply mean that there were different temperatures at different positions along the poker’s time axis. But this, as McTaggart remarked, would no more be a change in temperature than a variation of temperature along the poker’s



length would be.’ 7 It is more common for four-dimensionalists to see the issue between four-dimensionalism and three-dimensionalism as an a posteriori one, for they often rely on a posteriori considerations deriving from the picture relativity theory gives of what our world is like. But they also appeal to a priori considerations. Thus, some

-138-

return the compliment to three-dimensionalists by complaining that the notion of an object being wholly present at different times is unintelligible. Or they argue that the problem of temporary intrinsics shows that change must be construed as 4D change and, as we will see, this problem arises from a claim, clearly regarded as a priori knowable, to the effect that shape is an intrinsic property.

I have three arguments for heterodoxy.

If what constitutes change and persistence are a priori matters, then sufficiently acute reflection on our, that is, the folk, concepts of change and persistence (the latter under its folk title of remaining the self-same or numerically the same object throughout a period of time, of course) can in principle reveal what constitutes change and persistence. But if change is 3D change, and persistence is endurance, this is hard to believe, for the simple reason that being fully present is not a folk notion. Indeed, it is so far from being a folk notion that many highly educated folk insist that they do not understand it (at least as applied to objects). True, we are familiar with the idea that a priori reflection can reveal unobvious connections. That is part of the reply to the paradox of analysis. The most famous illustration is what happened after Gettier. It took his examples to make us realise that our concept of knowledge was linked to more than truth, belief and justification. But all the plausible candidates for the other concepts that need to be brought into the picture—cause, reliability, tracking, non-defeasibility, and so on—are familiar concepts (though not always under these names). What we are not familiar with, and what it is hard to believe, is that a priori reflection can reveal a connection between folk notions like change and persistence, on the one hand, and a notion like being fully present, on the other, which requires special explanation. The same goes, of course, if change is 4D change, and persistence is perdurance. The notion of a temporal part is not a folk notion. Indeed, it is so far from being a folk notion that many highly educated folk insist that they do not understand it (at least as applied to objects).

You might be tempted at this point to argue that the elucidation of the folk notions of change and persistence is a boring question. But what then is the question of interest



in the area? The elucidation of Geach’s and David Wiggins’s concepts of change and persistence? But the answer to that question is too easy. They tell us plainly and in so many words (to all intents and purposes)

-139-

that their concepts of change and persistence are 3D change and endurance, respectively. 8 Or is the interesting question the elucidation of Quine’s, Smart’s and Lewis’s concepts of change and persistence? 9 But the answer to that question is also too easy. They tell us plainly and in so many words (to all intents and purposes) that their concepts of change and persistence are 4D change and perdurance, respectively.

The argument just given is not decisive. It may be that someone will come up with accounts of change and persistence which are spelt out entirely in folk terms, and which, when presented for examination, strike us as clearly correct. But the leading candidates, with their dependence on their respective terms of art, have been the leading candidates for long enough to justify scepticism about this possibility.

My second argument to the conclusion that it is a posteriori what change and persistence are is directed to three-dimensionalists. There is an argument which, it seems to me, shows that three-dimensionalists cannot declare the four-dimensionalists’ picture of what the world is like, and in particular of what change and persistence come to, a priori false. They have to see the issue as an a posteriori one. The argument is a development of a point made by David Armstrong. 10

Suppose that God entertains Himself by creating very short-lived objects which remain unchanged throughout their brief lives. Suppose that He does this in such a way that what is in reality a sequence of short-lived, unchanging objects differing at most in small, systematic ways, one from another, appears to us to be a single, changing object. God is a kind of cartoonist working in three-dimensions, and working with such skill that the ‘flicker’ is too fast for us to see. Suppose, further, that God decides after a time to vest his power of creation in the objects themselves. He gives the n-th object in the sequence the causal power to produce a similar (n+1)-th object. How similar depends on the degree of outside interference, and, indeed, if the outside interference is great enough, there may be no (n+1)-th object produced at all. Finally, He eliminates the gap between the objects; the moment one ends is the moment the next is brought into existence. But now we have described, in terms wholly acceptable to the three-dimensionalist, the way the four-dimensionalist thinks



that objects in general are. On the four-dimensionalists’ view, all objects are like the descendent of our flickering object. I conclude that the three-dimensionalist must

-140-

admit that the four-dimensionalists’ picture of the world might be true. 11

This does not, of course, mean that three-dimensionalists must hold that the sentences ‘Change is 4D change’ and ‘Persistence is perdurance’ might be true. They can hold (a) that the four-dimensionalists’ conception of what the world is like is false, but it is contingent that it is false—this respects the conclusion we have just reached, and (b) that ‘change’ and ‘persistence’ are rigid designators of 3D change and endurance, respectively. If they are right, ‘Change is 3D change’ and ‘Persistence is endurance’ are both necessarily true, and ‘Change is 4D change’ and ‘Persistence is perdurance’ are both necessarily false. The argument shows that three-dimensionalists should concede the a posteriori nature of their view that change is 3D change, and persistence is endurance, but it leaves it open to three-dimensionalists to hold that it is necessary that change is 3D change, and that persistence is endurance.

A three-dimensionalist might seek to draw a line between allowing that the world might be the way the four-dimensionalist thinks it is, and it being a posteriori that, for example, change is 3D change, by urging that it is part of what we folk mean by ‘change’ that, should it turn out that the four-dimensionalist picture of the nature of our world is correct, there is no change. This is hard to believe. It makes it out that when the folk say that things change, they are, by virtue of the meaning they give the word ‘change’, taking a position on a tricky metaphysical issue, the very stating of which involves notions with which they are unfamiliar. 12

My third and final argument for the a posteriori nature of the choice between three-dimensionalism and four-dimensionalism is directed to four-dimensionalists. The four-dimensionalists who take the view that the story modern physics tells us about our world forces the four-dimensionalist picture on us do not need convincing that the issue between four-dimensionalism and three-dimensionalism is an a posteriori one. But, as we observed earlier, some four-dimensionalists appeal to the problem of temporary intrinsics, and this problem appears to provide an a priori reason in favour of four-dimensionalism. 13 I will argue, however, that inasmuch as it provides a good reason, it cannot be regarded as providing an a priori one.



The argument from temporary intrinsics, as applied to our poker, turns on three claims: (i) that, in some sense, being bent

-141-

and being straight are incompatible properties; (ii) that we therefore require an explanation of how one and the same object—our poker, say—can be both straight and bent; and (iii) that the four-dimensionalist can, and the three-dimensionalist cannot, provide a satisfactory explanation. The four-dimensionalist explains how one and the same poker can be straight at one time and bent at another by construing this as one temporal part of the poker being bent, and another temporal part of the poker being straight. There is only one poker because the two temporal parts are parts of a single, temporally extended poker, but this is of course consistent with the two parts having incompatible properties.

Why is the three-dimensionalist unable to provide a satisfactory explanation? The question here is one about the metaphysics of the situation, and on this topic, as opposed to the topics of conceptual priority and of the semantics of tensed discourse, it seems that the three-dimensionalist has just one thing to say—namely, that being bent and being straight are relations to different times. There is, three-dimensionalists can say, no more a contradiction between our poker being bent and our poker being straight than there is between our poker being heavier than that object and lighter than this object. What is supposed to be wrong with this reply? The advocate of the argument from temporary intrinsics bangs the table at this point and declares that being bent and being straight are intrinsic properties, not relations, and there’s an end on it! It is, that is, a priori that being bent and being straight are intrinsic properties, and so, we have, it seems, an a priori argument for four-dimensionalism. But how are we supposed to determine whether or not a feature is intrinsic or relational?

The standard way into the question of whether a feature F is intrinsic or not is via the question of whether or not F can be possessed in isolation. Now, intuitively, being straight and being bent can be possessed in isolation from any other contingent beings—a world might consist of a single straight thing, or of a single bent thing—but it is quite another question whether the single straight thing or the single bent thing could exist outside time. A protagonist of the view that being straight and being bent are relations allows of course that they are less relational than, for example, being taller than. The contention is that being taller than is really a three-place relation: when Tom is taller than Dick, a three-place relation holds between Tom, Dick, and the time in question; whereas when Tom is bent, a two-place relation holds



-142-

between Tom and the time in question. Thus, a protagonist of the view that being bent and being straight are relations can grant the intuitions that lie behind the view that being bent and being straight are intrinsic. They are as intrinsic as it is possible for such features to be. Moreover, these protagonists might well point out, the view offers an explanation of the difficulty we have in imagining something bent or something straight existing outside time. The difficulty is caused by the fact that we are trying to imagine the holding of a relation in the absence of one of its relata. 14

My conclusion, accordingly, is that the issue between three-dimensionalism and four-dimensionalism is an a posteriori one, and that three-dimensionalists and four-dimensionalists should agree with this conclusion. And if it is a posteriori whether the three-dimensionalists or the four-dimensionalists are right, I do not see how we could do better than leave the decision between the two positions to total science.

This conclusion brings with it a problem, and I suspect that this problem is the underlying reason so many have taken the issue to be an a priori one. If it is a posteriori whether or not change is 3D change or 4D change, what is this feature called ‘change’ concerning which it is a posteriori whether or not it is 3D change or 4D change? We must be able to give an account of how the term ‘change’ picks out change which leaves it open whether change is 3D change or 4D change. If the best account we can give of change is that it is a matter of a difference between two temporal parts of the one object, then it will be a priori that change is 4D change; equally, if the best we can do to explain what change is is to tell the story about the fully present having different properties at (or better, in view of the problem of temporary intrinsics, different relations to) different times, then it will be a priori that change is 3D change. There must be a way of saying what change is without actually saying which of 3D or 4D change it is—a way which leaves that as a decision to be made in the light of total science rather than one made by the very setting up of the question. The same goes for persistence. It must be possible to identify persistence without actually saying that it is endurance, or saying that it is perdurance, otherwise the choice between endurance and perdurance will be, contrary to the conclusion we have just argued for, an a priori one.

The solution is, I think, to appeal to the folk theory of change and persistence. We can think of ‘change’ and ‘persistence’ as

-143-



theoretical terms in the Lewis/Carnap/Ramsey manner, in the way best known for its application to mental state terms under such headings as the ‘theory theory’ and ‘folk psychology’. 15 We have a network of interlocking notions: those notions which get deployed in the only too familiar debates over personal identity, the ship of Theseus, the fission of amoeba, the propagation of properties, motion, counting by identity, and so on and so forth. This network constitutes the folk theory of change and persistence. We can think of it as the conjunction of all the intuitively appealing claims about change and persistence, typical members being ‘A car which has had a new steering wheel fitted is still numerically the same car as it was before, but a car which has had ninety percent of its parts replaced overnight is not’; ‘Motion involves change of position’; ‘If I mark this tyre with a piece of chalk, then, other things being equal, the tyre with the chalk mark on it in two hours time will be this tyre and not some other tyre’; ‘A person is more worried by the prospect of their being audited than they are by the prospect of someone else being audited’; ‘Tomatoes change colour when they ripen’; and so on and so forth. Change, then, is what occupies the ‘change’ place in the folk theory, and persistence is what occupies the ‘persistence’ place in the folk theory.

The folk theory gives us a ‘first fix’ on change and persistence. Change and persistence are whatever properties make the folk theory come out true (or near enough true—the folk theory may need some massaging). The philosophical trouble started with what we might call the second fix. Metaphysicans found two eligible solutions to the folk equations. There are two internally consistent stories to be told about how the equations can come out true. One, told for instance by Wiggins, has 3D change in the folk place for change, and endurance in the folk place for persistence; and the other, told for instance by Quine, has 4D change in the folk place for change, and perdurance in the folk place for persistence. 16 There is room for debate over which story is the best solution, all things considered. There is room for debate over whether someone might come up with a brand new solution to be judged against the two familiar competitors. And there is room for debate over whether the two familiar competitors are indeed both eligible solutions; perhaps there is an as yet undiscovered, but a priori detectable, fatal flaw in one or the other—in which case we were wrong to say that the choice between the two is an a posteriori one. But the historical fact that very able philosophers,

-144-



well versed in articulating the folk theory and in assessing the merits of possible solutions to it, have nearly all ended up advocating one or other of the two standard solutions, and have ended up in stalemate—neither side has been able to convince the other—strongly suggests that there are just the two solutions, and that each is good enough to count as an acceptable solution. None of this means, of course, that there is no way for us to make the choice between four-dimensionalism and three-dimensionalism. It means rather that, as we have already urged, the choice has to be made on straightforwardly empirical grounds, in terms of total science. 17

10.3 The network view of conceptual analysis

The story I have just told in a little detail about the debate between three-dimensionalism and four-dimensionalism illustrates a general picture of what is going on when we do analysis in philosophy, or at least what is going on very often when we do analysis in philosophy. It is a picture we find most clearly in the writings of David Lewis, although, if the theory is correct, it is implicit in just about everyone’s writings when they do analysis. We might call it the ‘network’ view of conceptual analysis. It breaks the analytical task into two parts. One consists in the assembling of the folk theory (or the physicist’s or biologist’s or economist’s or…theory, if we are analysing a concept special to physics, or biology, or economics, or…), thought of as a network of principles teasing out the connections between concepts, which would typically include the circumstances in which the various concepts are instantiated, and what characteristically follows from the instantiation of the concepts. If it is free action we are seeking to analyse, it will be the mass of connections between free action, moral responsibility, causal explanations of various kinds, the justifiability of punishment, personal identity, and so on, along with a catalogue of those cases most obviously judged to be of free action. Or in the case of ethics, we have what we might call, by virtue of its structural resemblance to folk psychology, folk morality. It is a body of folk opinion encapsulated in: input clauses telling us what sorts of situations naturalistically described typically warrant what sorts of moral descriptions (‘If an action increases suffering and has no other effects, it is wrong’, for example); internal role clauses telling us about the internal connections between matters ethically

-145-



described (‘Rights generate duties of respect’, ‘One ought to promote what is good’, for example); and output clauses telling us about the actions and motivations associated with moral judgements (‘If someone believes that they ought to do something, then they will typically be motivated to some extent to do it’, for example). 18 If it is laws of nature we are trying to analyse, it will be: the mass of connections between being a law, the holding of subjunctive conditionals, and causal explanations; the role of simplicity in the selection of laws; how laws figure in predicting the future; which generalisations are generally agreed to be laws; and so on and so forth. It is, I take it, a commonplace that major parts of books and articles on free action, ethics, and laws consist in the assembling, grading, refining, making explicit, and rendering as coherent as possible what reflection on actual and possible cases tells us about these networks.

The other part of the analytical task consists in testing various possible solutions to the equations so assembled. Thus, we might—many have—set the property of an act’s being free equal, roughly, to the property of an act’s being such that had the agent decided to do otherwise, the agent would have done otherwise, and then consider whether that assignment, along with appropriate assignments elsewhere, makes the relevant principles come out true, or near enough true. Again, we might set the property of being a law equal to being a regularity of such and such a kind, or set it equal to being a regularity following from a necessitation between universals, or set it equal to being a regularity true in all accessible worlds, and then see how much of the network associated with laws we can recover from that assignment. 19

In all these cases, it seems to me, the question addressed in the second part of the analytical task is a posteriori. I have argued for this in a little detail in the case of three-dimensionalism versus four-dimensionalism. But even without the detail, the claim is plausible for most of the hoary chestnuts of philosophical dispute. Scientists know what they mean by ‘law’—better than philosophers most likely—and yet they often have no firm, definite opinion, and sometimes little interest, in whether what makes for law-hood is being a certain kind of regularity, or being a necessitation between universals, or being true at every accessible world. How can this be? After all, I would not count as knowing what the word ‘square’ meant if I did not know what property it stood for. The obvious answer is: because grasping the concept of a law, the a priori part of the story, requires knowing the network.

-146-



It does not require knowing exactly which property occupies the ‘law’ position in the network (they will, though, know that some properties are non-starters). Again, members of a first-year philosophy class may come to very different conclusions about what makes an action free, and yet their agreement on the place of free action in the network allows them to understand what it is that they are disagreeing about.

We can now answer the question raised in the first section. How can it be Moorean that there are laws, and yet it be a matter of so much controversy what this property of law-hood is? If we are so divided about what the property is, and given that some seriously entertained suggestions as to what the property is are properties whose instantiation might well be doubted, how can we be so confident that there are laws? The same query can be raised for free action, moral rightness, persistence and change, and so on and so forth. The answer is that what is Moorean is that there is a solution to the equations for law-hood, moral rightness, and so on; what is possibly unMoorean is what the instantiated solution actually is.

We are now, finally, in a position to return to our target. For the network view of analysis enables us to put our finger on where the argument from the widely shared intuition that universal flukiness is possible to the conclusion that law-hood cannot be identified with any Humean property, goes wrong.

On the network view, what is a priori, what follows from our grasp of the concept of a law, is that ‘All F are G’ is a law iff (a) there is a property, L, which plays the law-hood role, and (b) ‘All F are G’ has L. But it is both a posteriori and contingent what L is. 20 There are many properties which might play the law-hood role. They would plausibly include following from nomic necessitation between universals F and G, suitably spelt out; for the suitable spelling out would be precisely giving to nomic necessitation between universals the right features for it to play the lawhood role. The candidates for law-hood would also plausibly include being true in all accessible worlds, for suitable spelling out of ‘accessible worlds’. This time the suitable spelling out would be giving truth in all accessible worlds the right features for it to play the law-hood role. And plausibly the candidates for L would include ‘is a regularity of type R,’ for suitable spelling out of R. This means that what is possible for law-hood is not always an a priori matter, because it may depend on what L turns out to be. If L turns out to be being a regularity of type R, it is not

-147-



possible for any and every regularity to be an entirely flukey matter. The widely shared intuition that it is possible will be a confused apprehension of the following possibility: that the property which plays the law-hood role could have been absent consistent with any amount of regularity. The latter is indeed possible. As we noted, following from nomic necessitations between universals might be the property which plays the lawhood role; and if it is, any and every regularity is consistent with the absence of law-hood, because any and every regularity is consistent with the absence of nomic necessitations between universals. 21

The point here is a general one. It is that if the network view of analysis is correct, a priori intuitions about possibilities have to be treated with considerable caution. For what is possible depends on which property it is that occupies the place in the network, and which property that is may not be an a priori matter. Motion is, it seems to me, a good illustration of this point. Consider a particle x whose position s at time t is given by: at t 0, s=0; at t> 0, s=t. Is x moving at t=0? It is, it seems to me, a mistake to investigate this question a priori. Rather, we need first to ask what motion is: that is, what property plays the motion role in the network. If it turns out to be change of position over time, then it is indeterminate whether or not x is moving at t=0. For there is no way to choose between the answer that it is, and the answer that it is not. 22 However, if it turns out—say, physics tells us—that motion is an instantaneous vector which explains change of position over time, and so is not identical with it, then there may well be an answer one way or the other concerning whether x is moving at t=0. 23

My objection, then, to the argument from the widely shared intuition that universal flukiness is consistent with any amount and kind of regularity to the conclusion that law-hood cannot be identified with any Humean property, is that it starts in the wrong place. We should start with the network into which laws fit; proceed to identify as best we can which property fills law-hood’s place in that network; and then, armed with an account of which property law-hood is, address the question of whether law-hood could be absent no matter how much regularity of whatever sort was present.

A similar point applies to arguments from undoubtedly appealing modal intuitions about motion and personal identity to conclusions about the nature of motion and personal identity. Take motion first. We are sometimes invited to admit that the following

-148-



is possible. There are two completely homogenous, and everywhere and every time self-sustaining discs that differ in that one is rotating and the other is stationary. But if this is possible, motion cannot be reduced to anything concerning how things are at times, even when supplemented with facts about dependencies between how things are at times. For how things are at times are exactly the same for the two discs, and as the discs are everywhere and every time self-sustaining, there is trivially no difference in dependencies. But, on the network view, there is an explanation of the appeal of this intuition that does not require admitting that what seems to be possible really is possible. The network that gives us our first fix on motion does not fully determine what motion is. It leaves the final determination of what motion is to a posteriori considerations. But then it will inevitably seem possible that there be two completely homogenous, and everywhere and every time self-sustaining discs that differ in that one is rotating and the other is stationary. For it really is possible that the property which plays the motion role should be a property which could be possessed by one of the discs and not possessed by the other. But it is quite another question whether the property which actually plays the motion role could be possessed by one of the discs and not possessed by the other. That question awaits a determination of what the property in fact is.

Again, each of us has the intuition that we could fall asleep and wake up in a new body, with no memory, actual or recoverable, of our past, and a quite different personality. The imaginative exercise is a simple, if radical, extrapolation from such facts as: we can get new body parts, we sometimes forget things, and we sometimes dream of improving our personalities. And indeed, there is something that really is possible, namely, that the property which plays the role in the network associated with personal identity over time should survive a new body, a new set of memories and a new personality. If we gave Cartesian egos suitable properties, the property of being associated with a given Cartesian ego would have what it takes to play the required role, and that property could survive getting a new body, a new set of memories and a new personality. But, of course, the property of being associated with a given Cartesian ego is not instantiated; it is not the property that in fact plays the role. The property that in fact plays the required role is a quite different property, and one that, on most reckonings of its nature, makes it impossible to remain the self-same person after changes as radical as those imagined.

-149-



I close with a comment on a general objection to the network approach to analysis. The objection is that the network is not well enough known to give even a first fix on the property we are seeking to identify. The point here would be like the point urged against common-sense functionalism by, for instance, Stephen Schiffer. 24 In that context, the objection is that various mental state concepts cannot be specified by their place in the commonsense network simply because there is no such thing as the common-sense network. We folk can say a bit about how the various mental states interconnect, but not nearly enough to do the required job. In the same way, it might be objected that we can say something about how for example law-hood interconnects with prediction, explanation, subjunctives, and all the rest, but not enough for it to be plausible that position in the network so specified captures the concept of law-hood. It would be nice to answer this objection by writing down the network. All I can do is say that, even without doing the job, we know that it can be done. Even if Armstrong et at. are right that law-hood is nomic necessitation between universals, it is clear that we and the scientists do not divide the generalisations that are laws from those that are not by detecting nomic necessitations through our senses or through instruments. There are no ‘nomic necessitationometers’. But nor do we make the division by magic: something about the information to hand guides us, or scientists or whoever, in making the division. But if it is explicable, it can be explicated—and to do that would be to write down the network. The same goes, it seems to me, for the other concepts we have alluded to: personal identity, change, motion, and so on. 25

Notes 1 I discuss the Hume world in Frank Jackson, ‘A Causal Theory of

Counterfactuals’, Australasian Journal of Philosophy, 55 (1977): 3-21. What I say in the current paper explains why I am no longer happy with what I say there concerning the Hume world. The main thrust of that paper is unaffected, though.

2 See, e.g., D.M. Armstrong, What is a Law of Nature?, Cambridge, Cambridge University Press, 1983; Fred I. Dretske, ‘Laws of Nature’, Philosophy of Science, 44 (1977):65-72; and Michael Tooley, ‘The Nature of Laws’, Canadian Journal of Philosophy, 7 (1977):667-98.

3 The suggestion that the right response might be to deny that there are any laws is mentioned by Armstrong, What is a Law of Nature?, p. 5, as one drawn to his attention by Peter Forrest.

-150-



4 See, e.g., W.V. Quine, ‘Identity, Ostension and Hypostasis’, in From a Logical Point of View, New York, Harper and Row, 1963; J.J.C. Smart, Philosophy and Scientific Realism, London, Routledge & Kegan Paul, 1963; and David Lewis, On the Plurality of Worlds, Oxford, Basil Blackwell, 1986.

5 See, e.g., Mark Johnston, ‘Is There a Problem about Persistence?’, Proceedings of the Aristotelian Society, supp. vol. 61 (1987):107-35, and David Lewis, ‘Rearrangement of Particles: A Reply to Lowe’, Analysis, 48 (1988):65-72.

6 If we want to rule out Cambridge change as change, we would need to note that having an attribute is more than merely satisfying a predicate, but this issue will not concern us here.

7 ‘Some Problems about Time’, in Logic Matters, Oxford, Basil Blackwell, 1972, pp. 302-18; see p. 304.

8 See, e.g., P.T. Geach, ‘Some Problems about Time’, p. 304, and David Wiggins, Sameness and Substance, Oxford, Basil Blackwell, 1980.

9 See, e.g., W.V. Quine, ‘Identity, Ostension and Hypostasis’, in From a Logical Point of View, Harper and Row, New York, 1963; J.J.C. Smart, Philosophy and Scientific Realism; and David Lewis, On the Plurality of Worlds, Oxford, Basil Blackwell, 1986.

10 D.M. Armstrong, ‘Identity Through Time’, in Peter Van Inwagen, ed., Time and Cause, Dordrecht, Reidel, 1980, p. 76; see also Mark Johnston’s remarks on flickering objects in ‘Is There a Problem about Persistence?’; and David Lewis, postscript B to ‘Survival and Identity’, in Philosophical Papers, vol. I, New York, Oxford University Press, 1983.

11 Some three-dimensionalists allow that events have temporal parts and perdure. Their objection is to the view that objects and substances have temporal parts and perdure. See, e.g., D.H. Mellor, Real Time, Cambridge, Cambridge University Press, 1981. For these three-dimensionalists, the point could be put by saying that they should allow that the entities they take to be objects and substances might in fact be events.

12 And, of course, causal theorists of reference will point out that if the four-dimensionalists’ picture is correct, 4D change will be the right sort of causal origin of our use of the word ‘change’.

13 See, e.g., David Lewis, ‘Rearrangement of Particles: A Reply to Lowe’, and D.M. Armstrong, ‘Identity Through Time’, p. 68.

14 Cf. the familiar point made in favour of adverbial accounts of sensory experience that they explain why sensory states cannot exist unowned; see, e.g., R.M. Chisholm, Perceiving, Ithaca, Cornell University Press, 1957, ch. 8.

15 See, e.g., David Lewis, ‘How to Define Theoretical Terms’, in Philosophical Papers, vol. I, New York, Oxford University Press, 1983, pp. 78-95.

16 See, e.g., David Wiggins, Sameness and Substance, and W.V. Quine, ‘Identity, Ostension and Hypostasis’.

17 Speaking for myself, the four-dimensionalists’ solution seems the one supported by total science. For a defence of the view that total



-151-

science supports the three-dimensionalists’ view, see, e.g., Sally Haslanger, ‘Persistence, Change, and Explanation’, Philosophical Studies, 56 (1989):1-28.

18 For more on the folk morality view of ethics, see Frank Jackson, critical notice of Susan Hurley, Natural Reasons, Australasian Journal of Philosophy, 70 (1992):475-87. Added note: see also Frank Jackson and Philip Pettit, ‘Moral Functionalism and Moral Motivation’, Philosophical Quarterly, 45 (1995):20-40, and chs 5 and 6 of Frank Jackson, From Metaphysics to Ethics, Oxford, Oxford University Press, 1998.

19 See Robert Pargetter, ‘Laws and Modal Realism’, Philosophical Studies, 46 (1984):335-7, for a version of the accessibility account.

20 That is, it is a posteriori and contingent which property plays the law-hood role. I am assuming that law-hood is the same property in all possible worlds, and so that when we find out which property plays the law-hood role, we find out that a sentence of the form ‘Law-hood is…’, where the ellipsis is filled with a rigid designator of whatever the property turns out to be, is necessarily true. I make the same assumption for motion below. This is the usual assumption to make: ‘If law-hood is not what laws have in common, and motion is not what moving objects have in common, what is it that they have in common?’ is the thought, I take it. But of course the network account of analysis opens the possibility of giving an answer to what laws have in common other than law-hood; and likewise, an answer for what moving objects have in common other than motion. You can say that law-hood is a different property in different possible worlds, and that what unites the laws in the different possible worlds is that they have what plays the law-hood role in their worlds. Likewise, you can say that motion is a different property in different possible worlds, and that what unites the moving things in the different worlds is that they have what plays the motion role in their worlds.

21 If you favour the view, mentioned in Note 20, that law-hood is a different property in different worlds, then the mistake is that of inferring from law-hood not being being-a-regularity-of-type-R in some world to its not being being-a-regularity-of-type-R in our world.

22 For some detail on this, see Frank Jackson and Robert Pargetter, ‘A Question about Rest and Motion’, Philosophical Studies, 52 (1987): 21-6. Added note: this paper pre-dates my realising that the motion-network might have a solution other than that provided by the at-at theory.



23 For a defence of the vector view, see John Bigelow and Robert Pargetter, Science and Necessity, Cambridge, Cambridge University Press, 1990, ch. 2. They (rightly, according to our argument) advert to a posteriori considerations in support of their view.

24 Stephen Schiffer, Remnants of Meaning, Cambridge, Mass., MIT Press, 1987, ch. 2.

25 A version of this chapter was read to a conference on metaphysics at

-152-

the Australian National University, November 1992. I am grateful to many comments before, during and after this conference, and must mention John Bigelow, Caroline West, Lloyd Humberstone, Denis Robinson, David Lewis, Alex Miller, David Armstrong, and Kevin Mulligan.

-153-

11 ARMCHAIR METAPHYSICS What role, if any, is there for conceptual analysis in metaphysics? On the face of it, very little. Metaphysics is to do with what is in the world and what it is like, not with concepts and semantics. 1 We would expect science in the wide sense to be highly relevant, but not the armchair deliberations of the philosopher concerned with the analysis of concepts. However, traditionally metaphysicians have paid at least as much attention to questions of conceptual analysis, and to related questions of logical interconnections (to what entails or fails to entail what) as they have to what science tells us about the world. David Armstrong, for example, while rightly and famously insisting that what is said in the philosophy of mind must be answerable to what science tells us about the role of the brain in the causation of behaviour, spends most of A Materialist Theory of the Mind doing conceptual analysis. 2 It is understandable that recently many philosophers writing under the banner of ‘naturalism’ have declared the traditional preoccupations of metaphysicians with such armchair matters as conceptual analysis and entailments to be a mistake.



Naturalism comes in an extreme and in a moderate form. The extreme form rejects conceptual analysis and its ilk altogether. The extremists sometimes describe the history of conceptual analysis as the history of failure. I think that they are forgetting about biased samples. True, the well-known and much discussed examples of putative analyses in the philosophy journals are highly controversial, but that is why they are much discussed. The more moderate form of naturalism accepts that there is a legitimate activity properly called ‘conceptual analysis’: for example, conceptual analysis is what mathematicians did when they elucidated the notions of infinity, convergence on a limit, irrational numbers, and so on, and that was a good thing to do.

-154-

The moderate naturalist view is that conceptual analysis and its ilk are all very well in their place, but their place is not in metaphysics (qua speculative cosmology), or at least not at the heart of metaphysics. Questions of analysis and of what entails what belong to semantics and logic, not to ontology and metaphysics. Here is a characteristically straightforward statement by Kim Sterelny of moderate naturalism as it applies to the metaphysics of mind

My approach [to the mind] is not just physicalist, it is naturalist…. Naturalists are physicalists…but naturalists have methodological views about philosophy as well; we think philosophy is continuous with the natural sciences. On this view, philosophical theories are conjectures whose fate is ultimately determined by empirical investigation…. An alternative conception is to see philosophy as an investigation into conceptual truths proceeding by thought experiments that probe the way we understand our own concepts…. There very likely are ‘conceptual truths’; truths that depend only on the way we understand concepts and thus depend not at all on how the world is. But I doubt that there are any very interesting conceptual truths about the mind, or about thinking. 3

This chapter is a reply to moderate naturalism. I will argue that issues to do with conceptual analysis broadly conceived are inevitably central to any serious metaphysics. I will start by explaining what I mean by a ‘serious metaphysics’ and why any such metaphysics brings with it a problem I will call the placement problem. I will then argue that there is only one way to solve that problem, namely, by embracing a doctrine I will call the entry by entailment doctrine. But this doctrine, I will argue, inevitably makes matters of conceptual analysis central in metaphysics.



11.1 Serious metaphysics and the placement problem

Some physical objects are true. For example, if I were to utter a token of the type ‘Snow is white’, the object I would thereby bring into existence would be true. The object I would thereby bring into existence would also: have a certain mass, be caused in a certain

-155-

way, be of a type the other tokens of which have characteristic causes and effects in my mouth and from my pen, and in the mouths and from the pens of my language community, have a certain structure the parts of which have typical causes and effects, and so on and so forth. How is the first property, the semantic property of being true, related to the host of non-semantic properties of the sentence? How can we find a place or location for the semantic in the physical story about the sentence?

Some who have puzzled about this question have been tempted by a sceptical position on truth, and, correspondingly, on meaning and reference. They feel that the list of non-semantic properties—suitably expanded, of course—is an in principle complete one. Sentences are, when all is said and done, a species of physical object, and we know that science can in principle tell us the whole story about physical objects. And though we are not yet, and may never be, in a position actually to give that whole story, we know enough, as of now, to be able to say (a) that it will look something like the story I gave a glimpse of, and (b) that, in any case, it will not contain terms for truth, reference and meaning. But if the whole story does not contain truth, reference and meaning, then so much the worse for truth, reference and meaning. 4

A quite different response is to distinguish what appears explicitly in a story from what appears implicitly in a story. I tell you that Jones weighs 70 kilos and Smith weighs 80 kilos. That is something I tell you explicitly. Do I also tell you that Jones weighs less than Smith? Not in so many words, but, of course, it is implicit in what I said in the following sense: what I said entails that Jones weighs less than Smith. Likewise, runs the alternative response, truth, reference and meaning are implicit in the story completed science will tell about our sentence and the world in which it is produced: that story will entail that the sentence is true, that it has a certain meaning, and that its parts refer to certain things, including snow. This response locates the semantic properties of sentences within the picture completed science tells about sentences and the world by arguing that they are entailed by that story. The semantic gets a



place in the physical story by being entailed by the physical story.

I have just described a familiar example of the placement problem, and two responses to it. But we can generalise. Metaphysics, we said, is about what there is and what it is like. But of course it is concerned not with any old shopping list of what there

-156-

is and what it is like. Metaphysicians seek a comprehensive account of some subject matter—the mind, the semantic, or, most ambitiously, everything—in terms of a limited number of more or less basic notions. In doing this they are following the good example of physicists. The methodology is not that of letting a thousand flowers bloom but rather that of making do with as meagre a diet as possible. Some who discuss the debate in the philosophy of mind between dualism and monism complain that each position is equally absurd. We should be pluralists. Of course we should be pluralists in some sense or other. However, if the thought is that any attempt to explain it all, or to explain it all as far as the mind is concerned, in terms of some limited set of fundamental ingredients is mistaken in principle, then it seems to me that we are being in effect invited to abandon serious metaphysics in favour of drawing up big lists. But if metaphysics seeks comprehension in terms of limited ingredients, it is continually going to be faced with the problem of location. Because the ingredients are limited, some putative features of the world are not going to appear explicitly in the story. The question then will be whether they, nevertheless, figure implicitly in the story. Serious metaphysics is simultaneously discriminatory and putatively complete, and the combination of these two facts means that there is bound to be a whole range of putative features of our world up for either elimination or location.

What is it for some putative feature to have a place in the story some metaphysic tells in its favoured terms? One answer, already mentioned, is for the feature to be entailed by the story told in the favoured terms. I take it that few will quarrel with this as a sufficient condition. One good way to get a place in the story is by entailment. I will argue that not only is it one way, it is the only way. The one and only entry ticket into the story told in the preferred terms is by being entailed by that story.

In order to focus and make familiar the discussion, I am going to develop the argument for the entry by entailment view in terms of the particular example of physicalism and the psychological. I will be arguing, that is, that the psychological appears in the physicalists’ story about our world if and only if that story entails the



psychological. Similarly, when I discuss the connection between entry by entailment and conceptual analysis, I will focus on the example of physicalism and the psychological. It may help in following the argument that is to come to assume the truth of physicalism. Although the considerations will be deployed in this

-157-

familiar and particular context, it will be clear, I trust, that they would apply generally.

11.2 Placing the psychological in the physicalists’ picture

Physicalism is the very opposite of a ‘big list’ metaphysics. It is highly discriminatory, operating in terms of a small set of favoured properties and relations, typically dubbed ‘physical’; and it claims that a complete story, or anyway a complete story of everything contingent, about our world can in principle be told in terms of these properties and relations alone. It is miserly in its basic resources while being as bold as can be in what it claims.

A fair question is how to specify precisely the notion of a physical property. I am not going to answer this fair question. Roughly, I will mean what is typically meant: the kinds of properties that figure in, or are explicitly definable in terms of, those that figure in physics, chemistry, biology, and neuroscience. This rough characterisation leaves it open why those sciences, rather than say psychology or politics, are chosen to settle the favoured class, and it says nothing about how committed this approach is to those sciences being roughly right in the kinds of properties they need for their own internal purposes. Nevertheless, I think that the rough characterisation will do for our purposes here. As far as I can see, nothing in what follows turns on the answers to these two controversial matters. 5 What is important here is that there is a favoured list, not how a property or relation gets to be on that favoured list. 6

What will be important is the notion of a complete story. Our argument for the entry by entailment thesis, as it applies to physicalism and the psychological, will be that it is the physicalists’ claim to have a complete story about the nature of our world which commits them to our world having a psychological nature if and only if that nature is entailed by the world’s physical nature.

Physicalists variously express their central contention as: that the world is entirely physical; that when you have said all there is to say about physical properties and



relations, you have said all there is to say about everything, or anyway everything contingent including psychology; that the world is nothing but, or nothing over and above, the physical world; that a full inventory of the instantiated physical properties and relations would be a full

-158-

inventory simpliciter; and so on and so forth. What does all this come to?

We can make a start by noting that one particularly clear way of showing incompleteness is by appeal to independent variation. What shows that three co-ordinates do not provide a complete account of location in space-time is that we can vary position in space-time while keeping any three co-ordinates constant. Hence, an obvious way to approach completeness is in terms of the lack of independent variation. Four co-ordinates completely specify position in space-time because you cannot have two different positions with the same four co-ordinates. But, of course, lack of independent variation is supervenience: position in space-time supervenes on the four co-ordinates. So the place to look when looking for illumination regarding the sense in which physicalism claims to be complete, and, in particular, to be complete with respect to the psychological, is at various supervenience theses. 7

Now physicalism is not just a claim about the completeness of the physical story concerning the individuals in our world. It claims completeness concerning the world itself. Accordingly, we need to think of the supervenience base as consisting of worlds, not merely of individuals in them, Intra-world supervenience theses will not capture what the physicalists have in mind. We need to look to inter-world global supervenience theses, an example of which is

(A) Any two possible worlds that are physical duplicates (physical property, relation and law for physical property, relation and law identical) are duplicates simpliciter. 8

(A), however, does not express the physicalists’ thesis. Physicalism is a claim about the nature of our world. It is perfectly consistent with physicalism that there be a possible world exactly like ours which contains as an addition lots of mental life sustained in non-material stuff. This would simply be one of the worlds where physicalism was false. So physicalism does not say that every physical duplicate of our world is a duplicate simpliciter, and, therefore, does not say that all worlds that



are physical duplicates are duplicates simpliciter. For the same reason

(B) Any world that is a physical duplicate of our (the actual) world is identical simpliciter with our world.

-159-

does not capture physicalism. But it points us in the right direction. For it makes it clear that the trouble is being generated by the fact that (B) makes claims about worlds which are ‘bigger’ than our world, whereas physicalism is simply a claim that the physical nature of our world exhausts the nature of our world. It is not saying that there cannot be bigger worlds than ours which are bigger partly in virtue of having some non-physical ‘stuff. Accordingly, I suggest

(C) Any world which is a minimal physical duplicate of our world is a duplicate simpliciter of our world

as expressing physicalism’s essential claim, where we can think of a minimal physical duplicate in terms of what happens when we follow recipes. A recipe tells you what to do, but not what not to do. It tells you to add butter to the flour but does not tell you not to add bat wings. Why doesn’t it? Part of the reason is that no one would think to add them unless explicitly told to. But part of the reason is logical. It is impossible to list all the things not to do. There are indefinitely many of them. Of necessity, when we follow recipes, we take for granted an implicitly included ‘stop’ clause. A minimal physical duplicate of our world is what we would get if we—or God, as it is sometimes put—used the physical nature of our world as a recipe with an understood stop clause for making a world.

We noted earlier the physicalists’ talk of everything being nothing over and above, nothing more than, the physical; of the full physical story being complete, being the full story simpliciter, and so on. The hope has sometimes been that the key notion of, as it is sometimes called, ‘nothing buttery’ might be susceptible of a precise, non-circular definition in terms of supervenience. If (C), or something like it, is the best we can do by way of specifying what physicalism is committed to in terms of supervenience, then this hope has been dashed. The notion of minimality that features in (C) is too close to nothing buttery. We have, I trust, made matters clearer, but we have not reached what might have been hoped for.



We reached (C) by objecting to (A) and (B). But here is a positive argument to the result that (C) captures the physicalist’s claim. Suppose, first, that (C) is false; then there is a difference in nature between our world and some minimal physical duplicate of it. But then either our world contains some nature that the minimal phy-

-160-

sical duplicate does not, or the minimal physical duplicate contains some nature that our world does not. The second is impossible because the extra nature would have to be non-physical (as our world and the duplicate are physically identical), and the minimal physical duplicate contains no non-physical nature by definition. But if our world contains some nature that the duplicate does not, this nature must be non-physical (as our world and the duplicate are physically identical). But then physicalism would be false, for our world would contain some non-physical nature. Hence, if (C) is false, physicalism is false. We now show that if physicalism is false, (C) is false. If physicalism is false, our world contains some non-physical nature. But that nature cannot be present in any minimal physical duplicate of our world, as any such duplicate is a minimal duplicate. But then any such world is not a duplicate simpliciter of our world, and hence (C) is false.

There is a debatable step in this little proof that physicalism is true if and only if (C) is true. At the last stage, I assumed that if our world contains some non-physical nature, there is a minimal physical duplicate which lacks that nature. But suppose that some complex of physical properties in this world is necessarily connected in the strongest sense of ‘necessary’ to some quite distinct, non-physical property. In that case the assumption will be false. One response to this objection would be to invoke the Humean ban on necessary connections between distinct existences, but a less controversial response is available. I think that we know enough about what the physical properties are like—despite the fact that we ducked the question of how to specify them precisely—to be confident that the connection in question is not possible. The kinds of properties and relations that figure in the physical sciences do not have mysterious necessary connections with quite distinct properties and relations. In any case, this is what physicalists should say. To say otherwise is to abandon physicalism (in favour of something far more mysterious than interactionist dualism). So physicalists must, it seems to me, accept (C).

11.3 A side issue: supervenience and singular thought

To accept (C) is ipso facto to accept



(C*) Any world which is a minimal physical duplicate of our world is a psychological duplicate of our world.

-161-

Consider a minimal physical duplicate of the actual world. It will contain a duplicate of Bush. It might be urged that our Bush’s psychology, while being very similar to his doppelgänger’s, will not be quite the same as his doppelgänger’s. Their singular thoughts will be different by virtue of being directed to different objects. Only our Bush is thinking about our Clinton. Thus, if physicalism is committed to (C*), physicalism is false. 9

One response to this putative disproof of physicalism would be to challenge the view about singular thought that lies behind it, but I think we can steer clear of that issue for our purposes here. 10 The disproof as stated is trading on the counterpart way of thinking about objects in possible worlds: the way according to which no object appears in more than one world, and what makes it the case that an object which is F might have failed to be F is the fact that its counterpart in some possible world is not F. 11 However, the duplicate of our Bush is thinking about the very same person as our Bush in the only sense that the counterpart theorist can take seriously. If I had scratched my nose a moment ago, I would still have had the very same nose that I actually have. Noses are not that easy to remove and replace. The counterpart theorist has to say that what makes that true are certain facts about the nose of my counterpart in a world where my counterpart scratched his nose a moment ago. If that is good enough for being the very same nose, then the corresponding facts about Bush’s counterpart are good enough for it to be true that he is thinking about the very same person, and hence having the same singular thought.

The putative disproof might be developed without trading on the counterpart way of looking at matters. A believer in transworld identity typically holds that whether an object in our world is literally identical with an object in another is not a qualitative matter. Such a believer might well hold that though Bush, our Bush, and Clinton, our Clinton, eyeball each other in more worlds than this one, it is nevertheless true that in some minimal physical duplicates of our world, our Bush thinks about a qualitative duplicate of our Clinton who nevertheless is not our Clinton. On some views about singular thought, Bush will count as having a different thought in such a world from the thought he has in our world. In this case the steering clear requires a modification



of (C). The physicalist will need to require that minimal physical duplicates of our world be ones which, in addition to being physical property and relation identical with our world, are

-162-

haecceity-associated-with-physical-property-and-relation identical with our world.

11.4 From (C) to entry by entailment

Given that (C) follows from physicalism, there is a straightforward and familiar argument to show that if physicalism is true, then the psychological story about our world is entailed by the physical story about our world. (C) entails that any psychological fact about our world is entailed by the physical nature of our world.

We can think of a statement as telling a story about the way the world is, and as being true inasmuch as the world is the way the story says it is. Let Φ be the statement (an infinite disjunction of very long conjunctions) which tells the rich, complex and detailed physical story which is true at the actual world and all the minimal physical duplicates of the actual world, and false elsewhere. Let Ψ be any true statement entirely about the psychological nature of our world: Ψ is true at our world, and every world at which Ψ is false differs in some psychological way from our world. If (C) is true, every world at which Φ is true is a duplicate simpliciter of our world, and so a fortiori a psychological duplicate of our world. But then every world at which Φ is true is a world at which Ψ is true—that is, Φ entails Ψ.

We have thus derived the entry by entailment thesis for the special case of physicalism and the psychological. A putative psychological fact has a place in the physicalists’ world view if and only if it is entailed by some true, purely physical statement. Any putative psychological fact which is not so entailed must be regarded by the physicalist as either a refutation of physicalism or as merely putative. Although the argument was developed for the special case of physicalism and the psychological, the argument did not depend crucially on matters local to that special case. We could have argued in the same general way in the case of physicalism and the semantic, or in the case of Cartesian dualism and the semantic, or in the case of Berkelean idealism and physical objects. 12 Our argument essentially turned on just two facts about any serious metaphysics: it is discriminatory, and it claims completeness. It is those two features of serious metaphysics which mean that it is committed to views about what entails what.



How does entry by entailment show the importance of conceptual analysis? If Φ entails Ψ, what makes Φ true also makes Ψ

-163-

true (at least when Φ and Ψ are contingent). But what makes Φ true is the physical way our world is. Φ needs nothing more than that. Hence, the physicalist is committed to each and every psychological statement being made true by a purely physical way our world is. The analytical functionalist has a story about how this could be. It comes in two stages. One stage—the most discussed stage—is about how certain functional-cum-causal facts make it true that a subject is in one or another psychological state. The other stage is about how certain physical facts make it true that the appropriate functional-cum-causal states obtain. The story is a piece of conceptual analysis. Analytical functionalism is defended by the a priori methods characteristic of conceptual analysis. For us the important point is that the physicalist must have some story to tell; otherwise how the purely physical makes psychological statements true is rendered an impenetrable mystery. But it is the very business of conceptual analysis to explain how matters framed in terms of one set of terms and concepts can make true matters framed in a different set of terms and concepts. When we seek an analysis of knowledge in terms of truth, belief, justification, causation and so on, we seek an account of how matters described in terms of the latter notions can make true matters described in terms of the former. So far we have been unsuccessful (I take it, others disagree). When we seek a causal account of reference, we seek an account of the kinds of causal facts which make it true that a term names an object. When and if we succeed, we will have an account of what makes it true that ‘Moses’ names Moses in terms of, among other things, causal links between uses of the word and Moses himself. Compatibilists about free will seek accounts of what it is to act freely in terms of facts about the agent’s abilities and the causal role of the agent’s character. If they have succeeded, they have told us how the way things are described in terms of abilities and causal origins make true the ways things are described in terms of free action. And so on and so forth.

I take the understanding of conceptual analysis just outlined to be pretty much the standard one, but there seems to be another understanding around. For example, some who discuss, and indeed advocate, some version or other of the causal theory of reference also say that they oppose conceptual analysis. But the causal theory of reference is a piece of conceptual analysis in our sense. It is defended by a priori reflection on, and intuitions about, possible cases in the manner distinctive of conceptual analysis. And



-164-

it is hard to see how else one could argue for it. The causal theory of reference is a theory about the conditions under which, say, ‘Moses’ refers to a certain person. But that is nothing other than a theory about the possible situations in which ‘Moses’ refers to that person, and the possible situations in which ‘Moses’ does not refer to that person. Hence, intuitions about various possible situations—the meat and potatoes of conceptual analysis—are bound to hold centre stage simply because it is better to say what is intuitively plausible than what is intuitively implausible. Or consider the book by Sterelny from which we quoted the passages saying that ‘an alternative [to the methods of this book] is to see philosophy as an investigation into conceptual truths proceeding by thought experiments’ and ‘I doubt that there are any very interesting conceptual truths about the mind, or about thinking’. Later in the book we find the following passage:

Representational notions play a central role in psychology. Yet representational kinds are not individualistic. There is a standard parable that makes this point. Imagine that, far away in space and time, there is a world just like ours, Twin Earth. Twin Earth really is a twin to Earth; the parallel is so exact that each of us have a doppelgänger on Twin Earth, our twinself. A person’s twinself is molecule-for-molecule identical to himself or herself …you and your twin are individualistically identical, for what is going on inside your heads is molecule-formolecule the same. But the representational properties of your thoughts are not always the same. For consider your thoughts about your mother. These are about your mother. But your twinself has never seen, heard, or thought of your mother. That individual’s motherish thoughts are about another individual. An individual amazingly similar to your mother, but none the less a different person…. So the representational properties of thoughts about particular people don’t supervene on brain states; they are not individualistic. Variations of the Twin Earth theme have been run by Putnam and Burge to show that thoughts about natural kinds and socio-legal kinds are not individualistic either, (pp. 82-3)

Surely in this passage we are being offered a defence of a conceptual claim about the representational properties of our

-165-



thoughts, namely, to the effect that they are not individualistic, and moreover the defence is via a thought experiment (a ‘parable’)? Clearly Sterelny, and at least some other critics of conceptual analysis, must mean by conceptual analysis something different from what we mean here.

Many will feel that there is a major objection to the argument given from (C) to the conclusion that Φ entails Ψ. The objection is that the argument neglects a crucial distinction between entailment and fixing which arises from the recognition of the necessary a posteriori. 13

11.5 Entailment and the necessary a posteriori

‘Water=H2 O’ is necessarily true (modulo worlds where there is no water) despite being a posteriori. 14 This means that conditionals which say that if H2 O has a certain distribution, so does water—for example, ‘If H2 O is L-located then water is L-located’—are necessarily true though a posteriori. Does it follow that ‘H2 O is L-located’ entails ‘Water is L-located’? You might say ‘yes’ on the ground that every world where H2 O is L-located is a world where water is L-located. You might say ‘no’ on the ground that the conditional is not a priori knowable. The second response might be bolstered by noting that we would not normally assess the argument

H2 O is L-located.

Therefore, water is L-located.

as valid. The second response might be spelt out by insisting on a distinction between fixing and entailment proper. 15 If every P-world is a Q-world, then P fixes Q, but in order for P to entail Q it must in addition be the case that ‘If P then Q’ is a priori.

The decision between the two responses turns, it seems to me, on the decision between two different ways of looking at the distinction between necessary a posteriori statements like ‘Water =H2 O’ and necessary a priori ones like ‘H2 O=H2 O’. You might say that the latter are analytically or conceptually or logically (in some wide sense not tied to provability in a formal system) necessary, whereas the former are metaphysically necessary, meaning by the terminology that we are dealing with two senses of ‘necessary’ in somewhat the way that we are when we contrast

-166-



logical necessity with nomic necessity. 16 Thus the class of nomic possibilities is a proper subset of the class of logical possibilities—every nomic possibility is a logical possibility, but some logical possibilities (for example, travelling faster than the speed of light) are not nomic possibilities. Similarly, the idea is that the class of metaphysical possibilities is a proper subset of the class of logical possibilities. Every metaphysically possible world is logically possible, but some logically or conceptually possible worlds—for instance, those where water is not H2 O—are metaphysically impossible. On this approach the reason the necessity of water’s being H2 O is not available a priori is that though what is logically possible and impossible is available in principle to reason alone given sufficient grasp of the relevant concepts and logical acumen, what is metaphysically possible and impossible is not so available.

I think, as against this view, that it is a mistake to hold that the necessity possessed by ‘Water=H2 O’ is different from that possessed by ‘Water=water’ (or that possessed by ‘2+2=4’). Just as Quine insists that numbers and tables exist in the very same sense, and that the difference between numbers existing and tables existing is a difference between numbers and tables, I think that we should insist that water’s being H2 O and water’s being water are necessary in the very same sense. The difference lies, not in the kind of necessity possessed, but rather where the labels ‘a priori’ and ‘a posteriori’ suggest it lies: in our epistemic access to the necessity they share.

My reason for holding that there is one sense of necessity here relates to what it was that convinced us that ‘Water=H2 O’ is necessarily true. What convinced us were the arguments of Saul Kripke and Hilary Putnam to the conclusion that the terms ‘Water’ and ‘H2 O’ are rigid designators, and so that it follows from the fact that ‘Water is H2 O’ is true that it is necessarily true. This was the discovery of a semantic fact about certain terms, not the discovery of a new sort of necessity. Kripke and Putnam taught us something important about the semantics of certain singular terms which means that we must acknowledge a new way for a statement to be necessary, a way which cannot be known a priori; but it is a new way to have the old property. 17

If the relevant statements are necessary in the same sense but differ with regard to the epistemological status of their shared necessity, there is an obvious response to the question, Does ‘H2 O is L-located’ entail ‘Water is L-located’? It is to say ‘yes’, but

-167-



hasten to distinguish a priori from a posteriori entailment, and hold that the entailment in this case is a posteriori. In short, the response is to adopt the following definitions:

P a priori entails Q iff (a) every P-world is a Q-world, and (b) the conditional ‘If P then Q’ is a priori.

P entails (or fixes) Q iff every P-world is a Q-world.

P a posteriori entails Q iff (a) P entails Q, and (b) P does not a priori entail Q.

Hence it seems to me that our result that physicalism is committed to Φ’s entailing Ψ stands. (From here on we will think of Ψ as some arbitrary true psychological statement; it might even be the huge one that says all there is to say about our world’s psychology.) But there are two objections that moderate naturalists might well make all the same.

First, they might object that our argument assumed that there are psychological facts. The entailment result was obtained from the combination of (C) with the assumption that there are true psychological statements. A physicalist of the eliminativist persuasion can reject one of our premises. This is right, but does not affect the overall issue. Eliminativism about the mental is implausible. Eliminativism across the board about matters described in any terms other than the austerely physical is incredible. We could have developed the argument in terms of the relationship between the physical story and any true statement S about our world not framed in the austere terms, in order to show that the physicalist is committed to Φ’s entailing S. The general point is that a serious metaphysics is committed to there being entailments between the full story told in its favoured terms and each and every truth told in other terms. The only physicalistic eliminativism which would avoid commitment to entailments from Φ to statements in other terms would be one which said that there are no truths tellable in other terms. But such a metaphysics would no longer be a serious one in our sense. For it would not effect a partition of the truths acknowledged by it into those pertaining most directly to the properties and relations viewed as the fundamental ingredients of everything, and the others; it would not be discriminating. Also, it would be unbelievable. 18

Second, they might very reasonably object that the result that Φ entails Ψ is not a result which supports the importance of

-168-



conceptual analysis in anything like the traditional sense. Conceptual analysis in the traditional sense, as we noted earlier, is constituted by a priori reflection on concepts and possible cases with an aim to elucidating connections between different ways of describing matters. Hence, it might be objected, if we allow that some entailments are a posteriori, to demonstrate an entailment is conspicuously not to demonstrate the importance of conceptual analysis. In order to reply to this objection, I need to say more about the necessary a posteriori.

11.6 The a priori part of the story about the necessary a posteriori

We urged that the explanation of the a posteriori nature of the necessary a posteriori does not lie in the special necessity possessed; where then does it lie? I think that the answer is best approached if we have before us from the beginning two central facts about the issue. First, the issue is an issue about sentences or, if you like, statements or stories in the sense of assertoric sentences in some possible language, and not about propositions, or at least not propositions thought of as sets of possible worlds. 19 By the argument lately rehearsed, the set of worlds where water is water is the very same set as the set where water is H2 O; so by Leibnitz’s Law, there is no question of the proposition that water is water differing from the proposition that water is H2 O in that one is, and one is not, necessary a posteriori. Second, the puzzle about the necessary a posteriori is not how a sentence can be necessary and yet it takes empirical work to find this out. Russians utter plenty of sentences which are necessarily true and yet it takes, or would take, many of us a lot of empirical work to discover the fact. The puzzle is how a sentence can be necessarily true and understood by someone, and yet the fact of its necessity be obscure to that person. And the reason this is a puzzle is because of the way we use sentences to tell people how things are.

Consider what happens when I utter the true sentence ‘There is a coin in my pocket’. I tell you something about how things are, and to do that is precisely to tell you which of the various possibilities concerning how things are is actual. My success in conveying this simple bit of information depends on two things: your understanding the sentence, and your taking the sentence to be true. We have here a folk theory that ties together understanding, truth and

-169-



information about possibilities; and the obvious way to articulate this folk theory is to identify, or at least essentially connect, understanding a sentence with knowing the conditions under which it is true, that is, knowing the possibilities at which it is true and the possibilities at which it is false. 20 Indeed, if we could not connect understanding a sentence, accepting it, and knowing about possibilities in this kind of way, there would be little point in asserting sentences to one another. There are many possibilities concerning when the bus leaves. Your uttering a sentence I understand and accept, namely, ‘The bus leaves at six’, is exactly what I need to tell me which of the many possibilities is actual, and that fact constitutes the essential point of your utterance. But it seems, then, that understanding a necessarily true sentence should, at least in principle, be enough to reveal its necessary status. For understanding it would require knowing the conditions under which it is true, and how could you know them and yet fail to notice that they are universal? The puzzle is particularly pressing when, as in the cases we are concerned with, the statements are about relatively accessible, contingent features of our world.

I think—unoriginally 21 —that the way out of our puzzle is to allow that we understand some sentences without knowing the conditions under which they are true, but to do this in a way which retains the central idea of the folk theory that the three notions of: understanding a sentence, of a sentence’s being true, and of the information carried by a sentence are very closely connected. It is just that the connection is not always so simple as that understanding requires knowledge of the truth conditions of what is expressed.

Here is an illustrative example, familiar from discussions of two-dimensional modal logic, of understanding without knowing truth conditions. 22 Suppose I hear someone say ‘He has a beard’. I will understand what is being said without necessarily knowing the conditions under which what is said is true, because I may not know who is being spoken of. As it is sometimes put, I may not know which proposition is being expressed. If I am the person being spoken of, the proposition being expressed is that Jackson has a beard, and what is said counts as true if and only if Jackson has a beard; if Jones is the person being spoken of, the proposition being expressed is that Jones has a beard, and what is said counts as true if and only if Jones has a beard. Hence, if I don’t know whether it is Jackson, Jones or someone else altogether, I don’t know which proposition is being expressed,

-170-



and I don’t know the conditions under which what is said is true. But obviously I do understand the sentence. However, granting this does not require us to abandon the folk theory that understanding, truth and information are very closely connected. For I am much better placed than the Russian speaker who hears ‘He has a beard’. Unlike the Russian speaker, I know how context determines the conditions under which what is said is true. I know how to move from the appropriate contextual information, the information which in this case determines who is being spoken of, to the truth conditions of what is said. Thus, uttering the sentence has for me, but not for the Russian speaker, the potential to convey information about which possibilities are realised, and it has this potential precisely because I understand the sentence. 23 A similar point can be made about ‘water’ sentences. The doctrine that meanings ‘ain’t in the head’ means that the truth-conditions of our ‘water’ sentences depend on matters outside our heads that we may be ignorant of. But someone who does not know that it is H2 O that falls from the sky and all the rest, may, as we have already noted, understand ‘water’ sentences. Hence, understanding, say, ‘There is water hereabouts’ does not require knowing the conditions under which it is true, though it does require knowing how the conditions under which it is true depend on context, on how things are outside the head. 24 But this means that although understanding alone does not give truth conditions, understanding alone does give us the way truth conditions depend on context, and that fact is enough for us to move a priori from, for example, statements about the distribution of H2 O combined with the right context-giving statements, to information about the distribution of water. For example, suppose that the right account of the semantics of ‘water’ is that it is a rigidified definite description meaning roughly ‘stuff which actually falls from the sky, fills the oceans, is odourless and colourless, is essential for life, is called ‘water’ by our experts,…or which satisfies enough of the foregoing’, and consider the inference

(1) H2 O is L-distributed. (2) H2 O is the stuff which falls from the sky, fills the oceans, etc. (3) Therefore, water is L-distributed.

-171-



Although, as noted earlier, the passage from (1) to (3) is a posteriori—(2) being the a posteriori, contingent fact that needs to be known in order to make the step from (1) to (3)—the passage from (1) and (2) to (3) will be a priori. And it will be so because although our understanding of the word ‘water’ does not determine its reference-conditions, it does determine how the reference-conditions of the word depend on context, and (2) gives that context. It gives the relevant fact about how things are outside the head. 25 Our understanding of the relevant terms plus logical acumen is enough to enable us to go from (1) combined with (2) to (3). Indeed, this is the a priori fact which reveals the a posteriori entailment. We did not know that (1) entailed (3) until we learnt (2). But as soon as we learnt (2), we had the where-withal to move a priori to (3). We could put it this way. Conceptual analysis tells us that we can move from the combination of (1) and (2) to (3). For conceptual analysis—what is in principle possible a priori from understanding plus logical acumen—tells us how truth-conditions depend on context, and that is all we need to go from the combination of (1) and (2) to (3).

Earlier I drew the connection between the entry by entailment thesis which commits physicalists to holding that Φ entails Ψ, and the centrality of conceptual analysis to physicalism, by saying (a) that physicalism, on pain of making mysteries, had better tell us how physical nature makes true, in the sense of necessitating, psychological nature, and (b) that conceptual analysis tells us how matters framed in one set of terms can make true matters framed in different terms. One way of putting the objection under discussion is that H2 O being L-distributed makes it true that water is L-distributed, but conceptual analysis cannot tell us about this. That H2 O facts make true water facts is a posteriori. Our reply is that there is an a priori story to be told about how H2 O facts make true water facts.

More generally, the two-dimensional modal logic way of looking at the necessary a posteriori shows that even if the entailment from Φ to Ψ is a posteriori, there is still an a priori story tellable about how the story in physical terms about our world makes true the story in psychological terms about our world. Although understanding may not even in principle be enough to yield truth-conditions, it is enough to yield how truth-conditions depend on context. But of course the context is, according to the physicalist, entirely physical. Hence, the physicalist is committed to there

-172-



being an a priori story to tell about how the physical way things are makes true the psychological way things are. But the story may come in two parts. It may be that one part of the story says which physical way things are, Φ1 , makes some psychological statement true; and the other part of the story, the part that tells the context, says which different physical way things are, Φ2 , makes it the case that it is Φ1 that makes the psychological statement true. What will be a priori accessible is that Φ1 and Φ2 together make the psychological statement true. 26

Notes 1 I mean metaphysics in the sense of speculative cosmology rather than analytic

ontology, see D.C. Williams, Principles of Empirical Realism, Springfield, Ill., Thomas, 1966.

2 D.M. Armstrong, A Materialist Theory of the Mind, London, Routledge & Kegan Paul, 1968. Of course, part of the reason that he spends so much of the book doing conceptual analysis is that he thinks (reasonably) that the relevant empirical results are all but in.

3 Kim Sterelny, The Representational Theory of Mind, Oxford, Basil Blackwell, 1990, p. xi. I suspect however that Sterelny means (must mean) something different by ‘conceptual truths’ and ‘conceptual analysis’ from what we will be meaning by them. I return to this matter later.

4 This line of thought is very clearly presented, but not endorsed, in Michael Devitt, Realism and Truth, Oxford, Basil Blackwell, 1984, ch. 6.

5 I discuss some of questions that the notion of a physical property raises in Frank Jackson, From Metaphysics to Ethics, Oxford, Oxford University Press, 1998, ch. 1.

6 Physicalists who think that there are irreducibly de se truths—for instance, about what a person thinks—will of course include the appropriate physical specification of them in their favoured list. I am indebted here to Richard Holton and David Lewis.

7 What follows is one version of a familiar story, see, e.g., Terence Horgan, ‘Supervenience and Microphysics’, Pacific Philosophical Quarterly, 63 (1982):29-43, and David Lewis, ‘New Work for a Theory of Universals’, Australasian Journal of Philosophy, 61 (1983):343-77. I am much indebted to their discussions. Added note: I have sharpened and shortened the discussion of the supervenience theses in what follows.

8 As far as I can see, it does not matter for what follows what view precisely is taken of non-actual possible worlds: perhaps they are concrete entities of the same ontological type as our world, as David Lewis, On the Plurality of Worlds, Oxford, Basil Blackwell, 1986, holds; perhaps they are abstract entities as Robert Stalnaker, ‘Possible Worlds’, Nous, 10 (1976):65-75, holds; perhaps they are structured universals as Peter Forrest, ‘Ways Worlds Could Be’, Australasian



-173-

Journal of Philosophy, 64 (1986):15-24, holds; perhaps they are nothing at all, but talk of ‘them’ is understandable in terms of combinations of properties and relations, as D.M. Armstrong, A Combinatorial Theory of Possibility, Cambridge, Cambridge University Press, 1989, holds. Added note: in the original version of this chapter I omitted the needed reference to law in (A).

9 I am indebted to Rae Langton and David Lewis in what follows. 10 For a survey of various views about singular thought, see Simon Blackburn,

Spreading the Word, Oxford, Oxford University Press, 1984. 11 See Lewis, On the Plurality of Worlds, Oxford, Basil Blackwell, 1986. 12 A traditional objection to idealism has been that no congeries of actual and

hypothetical facts about sense impressions, no matter how complex and detailed, entails that there is a table in the next room; and so idealists must embrace eliminativism about tables. A contemporary criticism of this traditional objection is that it is confusing the ontological thesis of idealism with the analytical thesis of phenomenalism. If we are right, the traditional objection is very much to the point.

13 The possibility of this objection was forcibly drawn to my attention by Peter Railton, Lloyd Humberstone and Michaelis Michael. I am not sure where they stand on it however. The terms in which I characterise it are closest to those of Peter Forrest, ‘Universals and Universalizability’, Australasian Journal of Philosophy, 70 (1992): 93-8.

14 I am taking for granted the view that the term ‘water’ is a rigid designator of (as we discovered) H2 O. I am in fact agnostic on the question of whether ‘water’ as used by the person in the street is a rigid designator. But clearly it is a rigid designator in the mouths and from the pens of many philosophers alive today, and for the purposes of the discussion here we will give it that meaning. It is clear in any case that ‘water’ could have been a rigid designator.

15 Stephen Yablo, ‘Mental Causation’, Philosophical Review, 101 (1992):245-80 distinguishes fixing from conceptual entailment, see especially pp. 253-4. His conceptual entailment is, I take it, our a priori entailment below.

16 See, e.g., Peter Forrest, ‘Universals and Universalisability’. Saul Kripke, Naming and Necessity, Oxford, Basil Blackwell, 1980, p. 125, says that ‘statements representing scientific discoveries about what this stuff is…are…necessary truths in the strictest possible sense’. This suggests that he does not hold the view in question, though he does not, as far as I know, address the matter explicitly.

17 Saul Kripke, Naming and Necessity, Oxford, Basil Blackwell, 1980, and Hilary Putnam, ‘The meaning of “Meaning”’ in Mind, Language and Reality, Cambridge, Cambridge University Press, 1975, pp. 215-71.



18 I am indebted here to a discussion with John O’Leary Hawthorne. 19 At the end of the preface to Naming and Necessity, Oxford, Basil Blackwell,

1980, Kripke insists that his concern is with sentences, not propositions; see pp. 20-1.

-174-

20 This kind of theory in its philosophically sophisticated articulations is best known through the work of David Lewis, see, e.g., On the Plurality of Worlds, Oxford, Basil Blackwell, 1986, and Robert Stalnaker, Inquiry, Cambridge, Mass., MIT Press, 1984. But it would, I think, be wrong to regard the folk theory as being as controversial as these articulations. The folk theory is, it seems to me, a commonplace. The sports section of any newspaper is full of speculations about possible outcomes conveyed by sentences that discriminate among the outcomes in a way we grasp because we understand the sentences. There is, of course, a major problem about what to say concerning mathematical statements within this framework, but our concern will be with sentences about relatively mundane items like water, and with entailments between sentences putatively representing the way things are as a matter of empirical fact.

21 I take it that the account which follows is a thumbnail sketch of the approach naturally suggested by the two-dimensional modal logic treatment of the necessary a posteriori in, for instance, Robert C. Stalnaker, ‘Assertion’ in P. Cole, ed., Syntax and Semantics, vol. 9, New York, Academic Press, 1978, pp. 315-32 and M.K. Davies and I.L. Humberstone, ‘Two Notions of Necessity’, Philosophical Studies, 38 (1980):1-30. They should not be held responsible for my way of putting matters.

22 The example is a variant on one discussed by Stalnaker in ‘Assertion’. 23 It might be objected that I do know who is being spoken of: I know something

unique about the person being spoken of, namely that he is being spoken of and is designated by a certain utterance of the pronoun ‘I’. But this is ‘Cambridge’ knowing who. It is no more knowing who than Cambridge change is change. I do not know who will win the next election, but I know something unique (and important) about that person, namely, that he or she will be the person who will win the most votes.



24 Thus, there is a sense in which understanding ‘There is water hereabouts’ requires knowing truth-conditions, for it requires knowing a whole lot of things of the form ‘If the context is thus and so, then the sentence is true in such and such conditions’. What is not required for understanding is knowledge of the truth-conditions of the proposition expressed. I am indebted here to David Lewis and to Pavel Tich y, ‘Kripke on Necessity A Posteriori’, Philosophical Studies, 43, 1983:225-41. Added note: this issue is addressed in a little detail in From Metaphysics to Ethics, ch. 3.

25 Other views about how the word ‘water’ gets to pick out H2 O would require appropriately different versions of (2). The view I have used by way of illustration is roughly that of Davies and Humberstone, ‘Two Notions of Necessity’. If you prefer a causal-historical theory, you would have to replace (2) by something like ‘It is H2 O that was the (right kind of) causal origin of our use of the word “water”’. Of course, any view about how ‘water’ gets to pick out what it does will be controversial. But it is incredible that there is no story to tell. It is

-175-

not a bit of magic that ‘water’ picks out what it does pick out. Added note: moreover, when someone tells us a possible world ‘parable’, we are able to say what is, and what is not, water according to it (if we couldn’t, the parables would not have had the enormous impact that they manifestly have had). Therefore, the needed story is one we, at least implicitly, know. I discuss the implications of all this for the description theory of reference in ‘Reference and Description Revisited’, Philosophical Perspectives, vol. 12, forthcoming.

26 Versions of this chapter were read at Monash University and the University of New South Wales. I am much indebted to the ensuing discussions.

-176-



Part IV ETHICS AND ACTION THEORY

-177-


-178-

12 WEAKNESS OF WILL There is a long tradition which views the problem of weakness of will as the problem of how agents can intentionally do what they consider wrong. And given that intentional acts are ones in accord with wants and desires, the problem becomes how we can simultaneously judge something wrong and yet want to do it. Some say, ‘Nothing easier—unfortunately.’ Suppose they are right and there is no philosophical problem about desiring to do what you judge to be wrong. Would there then be no philosophical problem of weakness of will at all?

In this chapter I argue that there would still be a problem, and seek to solve it. I start, in §12.1, by locating the problem and distinguishing it from the problem (if any) about desiring to do wrong. Donald Davidson, in ‘How is Weakness of Will Possible?’, 1 separates out the problem of weakness of will from the problem of intentionally acting contrary to one’s judgement of the morally best thing to do. In his view, that is only a special case of the general problem of intentionally acting contrary to one’s judgement of what is best simpliciter. The aim of §12.1 is to justify taking the extra step of entirely separating out the problem of weakness of will from that of acting contrary to what is judged best. I should, however, point out that the arguments of the later sections do not depend crucially on §12.1. I have, naturally, largely formulated them on the assumption that the conclusion of §12.1 is correct, but they can be reformulated without this assumption.



Section 12.2 is concerned with certain essential preliminaries, principally about how the objects of desire can be viewed, and with how this view solves some puzzles about desiring and wanting, quite apart from its role in our solution to the problem of weakness of will. In §12.3 I derive from the central idea of §12.2, an

-179-

account of how desires can evolve in accord with the agent’s reason: weak-willed behaviour being behaviour springing from desires that do not evolve in this way.

12.1 Separating out the problem

Cases of intentionally acting against one’s judgement of what is right can be divided into many sub-cases. In particular, a contrast can be drawn between those we naturally describe in terms of a conflict between reason, on the one hand, and passion, bodily appetites and the like, on the other; and those where appetite et al. have no special role. My contention is going to be that we can identify a problem about understanding conflict between reason and appetite independently of whether one or the other (reason most likely) is directed towards what is judged good. Here is an example where appetite etc. have no special role. While backing out of a parking bay I dent the bodywork of someone else’s car, causing what I estimate to be $200 worth of damage. My car is unmarked. No one sees me. I know that I ought to leave my name and address under the windscreen wiper of the damaged car; but I also know that if I don’t, no one will know that I haven’t. I succumb to temptation and drive away quickly. In this case I intentionally do what is wrong because I want to hold on to my $200 more than I want to do what is right. I was not, though, overwhelmed by appetite, passion, emotion, feeling, or anything like that. I wanted to hold on to my $200, but I (perhaps unlike some misers) do not thirst or lust after money.

The contrast is with cases where agents are led to act against their better judgement by something that can be broadly classed as a bodily feeling. Cases of this kind are only too familiar to those who have struggled to give up smoking, or to stay on a diet. What leads me to succumb to a cigarette or a chocolate is different from what leads me to succumb to the temptation to hold on to my $200. It is the cases essentially involving feelings, bodily appetites, and so on that we will concentrate on.

Why is there a philosophical problem about them, regardless of whether there is one about desiring to do what is judged wrong? We naturally describe cases of weakness of will as cases where appetite plays an unduly significant role in causing action. The problem is to specify undue. Appetite can play a causal role without the action



thereby being incontinent; otherwise everyone who drinks because they are thirsty acts incontinently. Of course,

-180-

it is a good thing that on the appropriate occasions one feels thirsty to the appropriate degree. But that is not why drinking because one feels thirsty is, typically, not incontinent. Plenty of smokers decide that the unpleasantness associated with resisting their desire to smoke outweighs the improved chance of long life if they stop. (I have in mind smokers who face up to the issues honestly, not those who pretend—maybe half-successfully—that the chances of early death are less than they are.) Perhaps they make the decision to continue smoking after their family tells them just how unpleasant they were to live with the last time they tried to give up. These smokers need not be acting incontinently in smoking. Yet they smoke because of the craving, and grant that it would be better if they did not have the craving; and, no doubt, wish that they didn’t have the craving and so wish that they did not want to smoke. Nevertheless, their smoking need not be incontinent. They may simply have concluded, continently, that the increased chance of long life is not worth the certainty of increased misery for themselves and their family.

Also, undue influence cannot be analysed as influence over and above the ordinary. We may suppose that Casanova was influenced over and above the ordinary by sexual appetite, without thereby supposing him to be incontinent. For (as has been widely recognised) there is a distinction between being passionate or intemperate and being weak-willed. Two hypotheses about Casanova (as we are supposing him) are possible. According to one, he experienced greater sexual urges than most of us; according to the other, his urges were the same as ours, but he valued their satisfaction more highly than most of us by comparison with other things. Similarly, an overweight gourmet may be overweight because he experiences greater pleasure when eating than most of us, or because, though his pleasure is the same as ours, he values it more highly than, say, we value remaining slim. On neither hypothesis does it follow that our Casanova or our gourmet is weak-willed. Neither their strength of appetite nor their difference in values is sufficient for their being weak-willed. They may be weak-willed, but if they are, something else is involved as well.

At this stage one may be tempted to declare the attempt to identify a problem about weakness of will, separate from the whole question of desiring what is wrong, a failure. ‘It is obvious what the required sense of “undue” influence is: it is influence that leads agents to intentionally act contrary to their better judgement.’



-181-

But, on the contrary, it seems that causing agents to act contrary to better judgement is neither necessary nor sufficient for weak-willed action.

First, it is not sufficient, because there are three kinds of case where appetite leads to acting contrary to better judgement, only one of which is a case of weakness of will. Consider someone offered ‘one for the road’ at the end of a party, and suppose that he is in no doubt that the best thing for him to do is to say ‘No’—he has to drive home, he needs to be up bright and early next morning, and so on. We distinguish three cases, which shade into each other, where he nevertheless says ‘Yes’. The first is where he simply does not care enough about the dangers to others of driving home under the influence or about his duties on the morrow. He doesn’t care enough about what he acknowledges to be what it is best to do. The second kind of case is the weak-willed. He gives in because his will is weak in this particular case (and perhaps in general as well). The third case is where the desire for a drink is irresistible; his behaviour is neither reckless nor weak-willed, but rather is compulsive. In each of these three cases, the agent intentionally acts against better judgement because of the influence of appetite; but only in one, the second, do we have weak-willed behaviour—the first is recklessness, and the third is compulsion. 2

Second, it is not necessary. Weakness of will can be the explanation of your failure to follow out a course of action you judge to be the wrong one. It does not have to be a course of action you judge to be right, though typically it is. Imagine a committed catholic who is raped and becomes pregnant. She is certain that it would be wrong for her to have an abortion, but she determines nevertheless to have one. When the time comes, however, her feelings of guilt overcome her resolve to have the abortion. Cannot such a case be one of weakness of will? Had she been more strong-willed, she would have gone through with the abortion; hence her failing must count as one of weakness of will. She acts incontinently, while doing what she judges best.

I have sought to raise our problem about weakness of will by pointing to the problem about specifying the undue influence we intuitively assign feeling in cases of weak-willed action. We have seen that the undue influence definitive of incontinence does not amount to some influence, nor to influence that is not desired, nor to influence above the norm, and nor to influence that leads to acting intentionally against better judgement. These difficulties

-182-



may engender scepticism about the viability or significance of our ordinary concept of weakness of will. After all, we invoke weakness of will as an explanation in conflict situations. Something about a contemplated course of action appeals; something else about it repels. If our intuitions about undue influence prove so hard to clarify, perhaps the right response is to abandon weakness of will as a possible explanation of how the agent comes to choose as he or she does. Perhaps all there is to say is that if an agent adopts the course of action, this proves their desire for the appealing feature is greater than their desire to avoid the unappealing feature; while if they refrain, this proves their desire for the appealing feature is less than their desire to avoid the unappealing one.

The attraction of such a procrustean position is obvious: explanation solely in terms of relative strength of competing desires is explanation in comparatively well-understood terms; and so invoking weakness of will as a possible factor can be seen as muddying the waters with pre-scientific and pre-Humean notions about the existence and relative independence of various faculties of the mind. What are reason and desire? And, supposing we have clear answers to this question, how can they conflict, given Hume’s dictum about reason being the slave of the passions? 3

Nevertheless, I think that clear sense can be made of weakness of will in terms of agents’ acting against the dictates of their reason; and that this can be done without becoming enmeshed in the faculties of the mind, and without denying what is right about Humean views about reason and desire. My starting point is, in fact, a Humean position about reason and desire.

12.2 Preliminaries concerning wanting

Hume said that ‘reason alone can never be a motive to any action of the will’. 4 I am going to assume what I take to be a noncontroversial element in this and related views of Hume’s about action. According to the non-controversial element, every action (or at least every action of the kind we are concerned with—maybe idle tongue clickings and the like are exceptions) involves both beliefs and something like wants or desires. 5 When I sign a cheque, there is something I want or desire—say, to pay a bill—and also something I believe about how what I am doing will achieve what I want; and both are involved.

There is a whole host of terms—‘want’, ‘desire’, ‘preference’,

-183-



‘value’, ‘pro- and con-attitudes’, to name a few—used for the thing which is not belief but which must be there along with belief, in intentional action. I am going to assume that there is some unified notion lying behind all these terms; and I will use the various terms more or less interchangeably for it, but without, I hope, making controversial presumptions about how it is to be analysed, and without presuming that in other, related contexts, it might not be important to distinguish one from another. Also I will, a little later, assume that we can measure an agent’s wants, desires, etc. with a value function which takes a higher positive value the more something is wanted, and a lower negative value the more something is disliked; and which is such that if A is wanted as against an alternative B, V(A) > V(B). This last assumption is one of convenience, considerable convenience admittedly, but dispensable nevertheless. The calculation sketches given later could all have been replaced by a purely qualitative and comparative discussion entirely in words.

There is one respect in which what I will be assuming is weaker than Donald Davidson’s thesis in ‘Actions, Reasons and Causes’. 6 I will not be assuming that wants cause actions. For all I assume, it is (always or sometimes) beliefs alone among psychological states that do the causing; the reason that wants are always involved being that wanting in the relevant sense is a consequence of the fact that belief does the causing. According to such a position, what made it true that I wanted to pay my debts may have been nothing more than the fact that my belief that a certain action would have that result was what caused the action. The want did no causing, rather that I had it followed from the fact that the belief did the causing. 7 Alternatively, for all I need, you might hold that wants do the causing, and the fact of belief follows from that. All I need is the common ground that both wants and beliefs are involved.

The main preliminary concerns how I will treat the objects of wanting, valuing, etc. I am going to borrow from the treatment characteristic of decision theory (though not from those parts embroiled in the dispute over Newcomb’s paradox). I will talk of states of affairs as being what is wanted, desired, valued, and so on, and regard them as propositional in that they are subject to the standard logical operations—conjunction, negation, and so forth. (One way of articulating this treatment is by identifying a state of affairs with the set of possible worlds in which it obtains and treating, for instance, conjunction as intersection.) We will be

-184-



discussing such questions as how the value placed on a state of affairs A relates to that placed on AB̅ and AB; and we will be treating A as one and the same as AB or AB̅, and as one and the same as ABC or AB̅C or…; and so on. Incorporated in this treatment of wanting, desiring, et al. is the intuitively appealing idea that how much I want something depends in part on what I think is likely or probable if that something obtains. I want a holiday in the sun. I know that one way this might obtain is by my having a holiday in the sun along with my getting skin cancer obtaining; another is by my having the holiday along with returning refreshed to work obtaining. It is clearly vital to my wanting a holiday in the sun that I judge the second much more likely than the first, given I take the holiday.

The justification for this treatment is partly that it leads naturally to my account of weakness of will, but also that it enables a relatively straightforward account of features of wanting that otherwise are hard to understand. I think that this fact has not received the publicity it deserves. I will consider, briefly and in turn, the account we get of the distinction between derivative and ultimate desires, of why certain inference patterns involving desires are invalid, and of the relationship between conditional and unconditional desires.

(a) Suppose I come to believe that having cold showers will strengthen my moral fibre; then, assuming I am not a masochist, the value I give having cold showers will derive from that I give having them and strengthening my moral fibre. (And if I am a masochist, the value will still be derivative, but on the value I give suffering.)

(It can be tempting to say that I don’t want, really want, to have a cold shower, C, at all. What I really want is the moral uplift, M. I don’t want C, I do want M. But this is to restrict one’s use of ‘want’—or of ‘really want’—to what one wants as an ultimate end. And this will not serve the purpose of understanding our ordinary talk of wanting, valuing, et al. For the vast majority of things we cite as desirable, including those cited in potential cases of weakness of will, are not ultimate ends. And, after all, derived value is derived value.)

What makes it true that the (positive) value I give C derives from that I give CM? The answer seems to be the combination of the following facts: (i) I value CM—the state of affairs of having both cold showers and strengthened moral fibre. It would not be

-185-



enough just to value M; the game has to be worth the candle, (ii) I disvalue CM̅. The cold shower without the uplift is of no interest to me. (iii) I judge CM much more likely than CM̅, given C. In sum, the value of C derives from that of CM because—although C can obtain by either CM or CM̅ obtaining, and I disvalue CM̅— I value CM, and I judge it to be much the most likely given C; to be, that is, much the most likely way of C’s obtaining. In general, we can use any mutually exclusive and jointly exhaustive set of states of affairs {B1 ,…, Bn }—(exactly one Bi obtains at every possible world)—to partition a state of affairs, A, into {AB1 ,…, ABn }. And where Pr is the agent’s probability function, the value of A is related to that of its sub-cases ABi by

(P) V(A)=Σi V(ABi ).Pr(ABi /A)=Σi V(ABi ).Pr(Bi /A).

Intuitively, {ABi } is a set of ways of realising A, and the value of A is the value of each of the ways weighted by how likely it is to be the actual way given A obtains. 8 What we have just been doing, in effect, is looking at a particular instance of (P), namely V(C)=V(CM).Pr(M/C)+V(CM̅).Pr(M̅/C); and noting that the derived nature of C’s value is reflected in the high values of V(CM) and Pr(M/C), together with the low values of V(CM̅) and Pr(M/C).

(A variant on (P) which also is intuitively appealing is

.

However, a discussion of the relative merits of (P) and (P*) would involve inessential complications: the points I wish to make could be made with either. We will work with (P) as—for reasons related to the example about blood types to be given shortly—it happens to be the one I accept.) 9

The information about the structure of an agent’s values that a partition yields and explicates varies. Partitioning having cold showers into CM and CM̅ told us that my value for C derived from my value for CM. But partitioning can equally reveal—and explicate—irrelevance. Consider a utilitarian who holds that ultimately value is nothing but a matter of the total amount of happiness. Any and every way of partitioning the state of affairs of the total happiness equalling n will give sub-cases all having exactly the same value, because for her everything else is irrelevant when set beside total happiness. For instance, her V(total

-186-



happiness=n & the happiness is distributed equitably)=her V(total happiness=n & the happiness is not distributed equitably). And if she balks at this, she reveals herself as a lamb in wolf’s clothing, as one who tempers utilitarianism with a principle of justice. Ultimate value is a matter of V(AX)=V(AY) for any X, Y. There are, of course, complications. For instance, the value of many states of affairs is neither ultimate nor wholly derivative. Thus, the value I give being happy now derives partly from the fact that I believe happiness breeds happiness, that is, from the fact that I think the most likely way that happiness now will be realised is by happiness now and later being realised; but also the value I give happiness now and later partly derives from that I give happiness now. Also, a person may have a number of ultimate values, which may or may not form a hierarchy. 10 In such a case, the fact that, say, A, B, C are all the states of affairs that are ultimately valued will be reflected in facts like that V(ABCX)=V(ABCY), V(A̅B̅C̅X)=V(A̅B̅C̅Y), and so on, for any X, Y. But, without going into these complications, I trust enough has been said to show how a decision-theoretic approach can illuminate the nature of derived versus ultimate value.

Incidentally, although what value derives from is often a causal result—we pretended that I believe cold showers cause strengthened moral fibre—this is not essential. Suppose I read in a medical journal that a certain blood type is highly correlated with long life, not because it causes long life but because it is a reliable sign of genes which cause long life. Then I would want to have that blood type, and my valuing having it would derive from my wanting long life. (Of course, what I would want would be to have that blood type, not to bring it about that I have it: bringing that about, even if it were possible, would achieve nothing, because it wouldn’t change my genes.) 11

(b) Davidson in ‘Intending’ 12 raises the problem that you can desire something sweet yet not desire something sweet and poisonous. How so? After all, eating something sweet and poisonous is a sure-fire means to eating something sweet. Our answer is that the relationship between the desirability of a state of affairs and its sub-cases is mediated by probability. Provided that eating something sweet and poisonous is a very improbable way of realising the eating of something sweet, the undesirability of the former does not make the latter undesirable.

Of course, we frequently argue on the pattern: A is desirable, B

-187-



is a known sure-fire means to A, therefore B is desirable. (And ‘desirable’ may be replaced by ‘ought to be’, etc.) Now we have, from (P), V(B)=V(AB).Pr(A/B)+V(A̅B).Pr(A̅/B). If B is a known sure-fire means to A, Pr(A̅/B) ≈ 0 and Pr(A/B) ≈ 1, so V(B)=V(AB). Therefore, provided only that B does not unduly detract from A, so that V(AB) and V(A) are close, B must be desirable if A is. This is typically the case when we use the pattern in question: not getting polio is desirable, taking the Sabin vaccine is a known virtually sure-fire means for avoiding polio; therefore, taking the Sabin vaccine is desirable. We ought to be able to vindicate this inference. And we can, because the value of avoiding polio and taking the vaccine is nearly as high as that of avoiding polio.

As well as wanting A without wanting AB, you can want AB, yet not want A. I don’t want to smoke, but only because of the risk to health (let’s say). That is, I do want to smoke and have a long life; my reason for not wanting to smoke is only that I believe it is not that conjunctive state of affairs which would most likely obtain if I smoked, but rather the highly undesirable one of smoking and short life. Hence I want smoking and a long life, but not smoking.

Perhaps the failure of wanting AB to entail wanting A is most transparent when looked at in terms of the contrapositive. It is clear that not wanting A does not entail not wanting AB. Take most things you do not want, can’t they be conjoined with something—a million dollars, two million dollars,…—to give something you do want? (You may want the money by itself more, but we are talking about wanting, not most wanting.)

Clearly, we can explain the failure of wanting AB to entail wanting A in the same sort of way that we explained the failure of wanting A to entail wanting AB. From (P), it is perfectly possible for V(AB) to have a high value while V(A) has a low one; either due to the effect of a low Pr(A/B) diminishing V(AB)’s contribution to V(A), or due to V(AB̅) having a significant negative value, or to a combination of both. Intuitively, the typical case will be where AB is an unlikely way for A to obtain, and AB̅ is bad. In the same general sort of way, we can explain why wanting A and wanting B is not equivalent to wanting A-and-B. (Wanting neither agglomerates nor distributes.)

The theme for the past few paragraphs has been that whether a person wants A, or wants A more than A̅, depends in part on what they hold likely to be the case given A is the case, and that when

-188-



you bear this in mind, you can see why the various inference patterns are invalid. It might be objected that what is desirable depends not at all on the person’s opinions, but rather on the facts. Smoking was as undesirable before the health risks became known, as it was afterwards. The facts alone, be they known or not, are enough. It would be absurd to say, ‘What a pity we know about the health risks; for it is only the knowledge, not the risks themselves, that makes smoking undesirable.’ But this is more the making of a distinction than the making of an objection. I am talking about wants, desires, etc. in the sense in which they feature in explaining choice, decisions, etc.; and in that sense they do depend in part on opinions—or, more exactly, subjective probabilities. Before the health risks became known, I did desire to smoke, and that fact is important in explaining my choosing to smoke. I do not, though, deny that there is another notion, for which terms like ‘desirability’ or ‘objective value’ might be used; and I grant that this is independent of opinion, and consequently that a discussion of, for example, whether ‘AB is objectively desirable, therefore A is objectively desirable’ is valid would have to follow different lines from those above. 13

(c) Also in ‘Intending’, Davidson raises a problem about how conditional intentions relate to unconditional ones (see especially p. 100). A similar problem can be raised for wants and desires. For just about anything I want, it will be true that I do not want it if…I want breakfast tomorrow, but I don’t want breakfast tomorrow if I’m not hungry. Similarly, I value telling the truth, but not if it leads to World War III. The only wants, desires, values, obligations, intentions, etc. for which this is not true are the ultimate ones. Provided she did not backslide, our utilitarian wanted an increase in total happiness, be it equitably distributed or not. The temptation, then, is to hold that all the non-ultimate wants are conditional. It is literally false that I want breakfast tomorrow; all that is literally true is that I want breakfast if I’m hungry.

We need a way of resisting this conclusion, for consider the following sequence: I want breakfast; I do not want breakfast if I am not hungry; I do want breakfast if I am not hungry but need nourishment; I do not want breakfast if I am not hungry but need nourishment though not of the kind provided by breakfast; …If the second member shows that the first is not strictly true, then equally the third must show that the second is not strictly

-189-



true, and the fourth must show that the third is not strictly true, and so on and on. We won’t only lose a host of unconditional wants but a host of conditional ones as well.

The observations already made show us how to resist the undesirable conclusion. We are dealing with derived wants, for, as we saw, the problem of wanting A but not wanting A if (given) C does not arise with ultimate ones. Let {ABi } be a partition that reveals (at least in part, whether it goes ‘all the way down’ to the ultimate wants does not matter) the derivative nature of my want for A. From (P), we have V(A)=Σi V(ABi ).Pr(Bi /A). Now the obvious treatment of my want for A if C is to look at what happens to this expression for V(A) if we conditionalise the probabilities in it on C; we look, that is, at Σi V(ABi ).Pr(Bi /AC). Hence I can want A yet not want A if C, in exactly the same way that A can be probable yet not be probable given C. It follows that if C is certain according to me, my want for A if C must be the same as my want for A; because if C is certain, conditionalising on C cannot change the probabilities. This is the right result. The more sure I am that I will be hungry tomorrow, the closer my want for breakfast will be to my want for breakfast if I am hungry. (As Davidson in effect observes in the case of intention.)

12.3 The dynamics of value

The previous section was about the statics of value, about how a person’s wants at a time for various states of affairs are inter-related; but the discussion naturally suggests one way that a person’s wants may evolve over time. (P) yields a whole host of truths (one for each way of partitioning) of the form V(A)= Σi V(ABi ).Pr(Bi /A), about a person’s values at a time. When the value an agent gives A changes, these sums will have to change too. The obvious speculation is that sometimes, for some of these summations, the change will be in the probability terms alone. More specifically, it is that sometimes the value an agent gives A changes on acquiring new information, I, due not to any change in the V(ABi ) for some sets of ABi , but due to the change in the probabilities from Pr(Bi /A) to Pr(Bi /AI). (I’m going to assume, for concreteness, that probabilities evolve by conditionalisation; and I will sometimes describe the evolution of value consequent upon change in probability alone as evolution by conditionalising.)

I think exactly this often happens. At the beginning of I967, I wanted to smoke more than not smoke, and so did smoke (when I

-190-



felt like it, not all the time of course). During 1967 the health risks associated with smoking were drawn to my attention. By the end of 1967 I no longer wanted to smoke more than not smoke, and accordingly stopped smoking. My pro-attitude towards, my preference for, my desire for the value I gave smoking went down sharply, and the job was done entirely by information—by my reason, if you like. How?

The explanation, in simplified form, can be represented as follows. The state of affairs of my smoking, S, can be partitioned into my smoking and dying young, SD, and my smoking and not dying young, SD̅. My attitude towards each sub-case remained unchanged throughout 1967. The high value of SD̅ remained constant, as did the low value of SD; the information acquired did not change these values. The crucial changes were in my opinion about the chance of D given S—that went up sharply; and in my opinion about the chance of D̅ given S—that went down sharply. Now, from (P), we have V(S)=V(SD).Pr(D/S)+ V(SD̅).Pr(D̅/S). My V(SD) and V(SD̅) remained as before, and the change in my V(S) was due solely to those in my Pr(D/S) and Pr(D̅/S) to, respectively, Pr(D/SI) and Pr(D̅/SI), where / is what I learnt about the dangers of smoking.

This change in value, therefore, was determined by my reason alone. Nothing but a change in my probabilities was called for. Let’s say, then, that a change in an agent’s value, want, etc., for A on learning / is in accord with the dictates of an agent’s reason if there is a partition of A (in general, if there is one, there will be many) by {Bi ,…, Bn } such that each V(ABi ) is the same before and after, the change in V(A) being due to the change in the Pr(Bi /A) to Pr(Bi/AI); that is, the agent’s wants evolve by conditionalisation. The germ of this chapter will now be clear. There is a long tradition according to which weak-willed action is action not in accord with the agent’s reason—‘passion pushes reason from its throne’; my suggestion is that weak-willed action is action arising from wants and desires that have not evolved according to the dictates of the agent’s reason, as just defined. 14

It may be objected that thirst leads people to want water, and so to drink water, yet thirst is not a change in subjective probability; it is a bodily feeling. Hence it may seem that I am forced to judge anyone who drinks water because they are thirsty as thereby acting incontinently, because they are not acting from a want evolved in accord with the dictates of reason. This would be directly contrary to our earlier observation that you can act out

-191-



of bodily feeling, passion, appetite and so on, without it having to be the case that you are displaying weakness of will—and would, anyway, be absurd.

Thirst can, however, make your wants change so that you choose to drink, and it still be the case that your wants have evolved in accord with reason alone. For although thirst is not a belief (simpliciter, anyway), it typically changes belief. When I get thirsty, I acquire beliefs like that my body has a physiological need for water, and that I am experiencing a vaguely unpleasant feeling. Given what I know about how drinking meets the need and removes the feeling, these beliefs in themselves can be more than enough to explain why my wants change so that I come to want to drink, and so choose to drink. For simplicity, I will detail this simply for the impact of acquiring the belief that I have a need for water.

Let D stand for my drinking in the near future, and U for my having an unsatisfied need for water; then from (P) we have:

V(D)=V(DU).Pr(U/D)+V(DU̅).Pr(U̅/D),

V(D̅)=V(D̅).Pr(U/D̅)+V(D̅U̅).Pr(U̅/D̅)

Suppose that V(DU), V(DU̅), V(D̅U), V(D̅U̅) all remain the same, before and after I get thirsty; what I have to show is that information consequent on becoming thirsty may change the probability factors in a way which is in itself sufficient to explain why I want to drink in the near future. Before I get thirsty, I know that I don’t have an unsatisfied need for water, whether I drink or not. Thus, Pr(U/D) and Pr(U/D̅) will be almost nought, and so Pr(U̅/ D) and Pr(U̅/D̅) will be almost one. Hence, before I get thirsty, V(D)≈V(DU̅) and V(D̅)≈V(D̅U̅). As a result V(D) and V(D̅) will be much the same. After I get thirsty, I know that I will have an unsatisfied need for water unless I drink in the near future; thus Pr(U/D̅) will be close to one and Pr(U̅/D̅) close to nought. Pr(U/D̅) will remain near nought, because I know that water will relieve the need, and so Pr(U̅/D) will stay near one. Consequently, after I get thirsty, V(D) stays near V(DU̅), but V(D̅) drops to the low value I give V(D̅U)—for it is bad to have unsatisfied physiological needs. Hence we have the desired explanation of how probability changes alone can make V(D̅) less than V(D), and so make me want to drink water in the near future.

The crucial point here is, put very simply, that agents may

-192-



value removing thirst—in itself and by virtue of the undesirability of not satisfying physiological needs—even when they are not actually thirsty. Not every traveller in the desert is experiencing thirst when he packs his water bag. At some fundamental level, the evaluations of the strong-willed are not changed by the onset of feeling, appetite, emotion or whatever. Delve deep enough, and you will find a relevant set of evaluations unchanged by the onset of feeling. The change appears on the surface, so to speak, by virtue of changes in the probabilistic way that evaluations of more specific states of affairs feed into evaluations of more general states of affairs.

This treatment of how thirst can change your wants so that you choose to drink, without departing from the dictates of reason, confirms our observation at the very beginning that feelings may have some influence without thereby having the undue influence characteristic of weak-willed behaviour. Similarly, the influence may be greater than normal without being undue in the relevant sense. According to our account, it is irrelevant to weakness of will whether or not the agents’ values, wants, etc. are like ours or are close to the norm. What matters is, rather, whether feelings and the like have influence over and above that which the change in probability they cause has.

Consider another example, smoking—where weakness of will is notoriously prevalent. Suppose, for simplicity, that what value smoking, S, has derives entirely from its role in eliminating a craving, C. (I understand that this has come to be its only attraction for some.) From (P), we have:

V(S)=V(SC).Pr(C/S)+V(SC̅).Pr(C̅/S),

V(S̅)=V(S̅C).Pr(C/S̅)+V(S̅C̅).Pr(C̅/S̅).

Now consider smoker Fred’s situation when he is not experiencing the craving. Given plausible assumptions, both Pr(C/S) and Pr(C/S̅) will then be negligible, both Pr(C̅/S) and Pr(C̅/S̅) will be close to one; hence V(S)≈V(SC̅), V(S̅)≈V(S̅C̅), and so V(S) V(S̅). And this, of course, is the right answer. Smokers don’t want to smoke when they are not experiencing the craving (given our simplifying assumption). What happens when the craving comes on? Pr(C/S̅) will jump to nearly one, Pr(C̅/S) will stay near one (smoking always works to get rid of the craving), and Pr(C/S) and Pr(C̅/S̅) will be near nought. Hence, if Fred’s prior V(SC̅) is enough above his prior

-193-



V(S̅C), we can explain his V(S) rising above his V(S̅), and so his choosing to smoke, without supposing any change in V(SC̅) or V(S̅C). He is following the dictates of his reason, and so not displaying weakness of will. On the other hand, his prior V(SC̅) may not be enough above V(S̅C). When the craving is not on him, he may value SC̅ less than S̅C (presumably because of the health risks associated with smoking). In this case, the explanation for his choosing to smoke when the craving is on him lies not just in the craving’s effect on his probability function, but in its effect on his value function too. The craving must have lifted V(SC̅) above V(S̅C). In this case he is weak-willed: the mark of his incontinence being the disparity between his prior and his posterior underlying values consequent upon the craving.

We can express the difference between the strong-willed smoker and the weak-willed one in terms of the distinction between conditional wants (discussed earlier) and conditionals about wants. Neither smoker has the conditional want to smoke if he has no craving; both are such that if they have the craving, then they want to smoke. But they differ in that when they have no craving, the strong-willed smoker nevertheless has the conditional want to smoke if he craves to (because his wants evolve by conditionalisation), whereas the weak-willed smoker does not. He wants not to smoke if he craves to—although when the time comes, he will want to smoke.

This joint consequence of our account of weakness of will and of our earlier account of conditional wants seems plausible. Isn’t it characteristic of dieters who fear that their wills may not be up to the task, that they are clear that they want not to eat cheesecake when the urge is upon them, while at the same time fearing that when the urge is upon them, they will want to eat cheesecake? They want not to eat when the testing time comes, but fear that when the testing time comes, they will want to eat.

All this assumes that we can talk of how much an agent values eliminating a craving, avoiding thirst, and so on when he is not actually experiencing them. But I don’t think there is any real problem here. People usually buy aspirin in anticipation of a headache, not while they are actually experiencing one; picnic hampers are packed before hunger sets in; and so on and so forth. People’s behaviour now can indicate the attitudes they take now to experiences in the future (and the past), which attitudes may or may not remain the same when the time comes. Consider the behaviour characteristic of dieters who fear that they will be

-194-



weak-willed. At a time when they are not feeling hungry, they take steps to make it hard to get fattening food when they are hungry—they throw out the cheesecake, they arrange to lunch at special restaurants, they make sure they won’t have time to stop at the cakeshop. These actions indicate their attitudes when not hungry. They are doing things which they know will ensure that a state of affairs of their being hungry and slim rather than one of being not hungry and fat will obtain. Hence, at the time they value the former more than the latter. But these actions also indicate their knowledge about how these attitudes may change later, when they get hungry; about their tendency to weakness of will. For if their underlying attitudes remained constant, there would be no need to throw out the cheesecake or whatever; if when the hunger struck, they still valued being hungry and slim over being not hungry and fat, then their wants would not be such as to lead them to choose eating.

We can summarise our problem and our answer, as follows. Before the onset of appetite, passion, craving, yearning, or whatever, Fred does not want to smoke, eat, drink, make love or whatever. After the onset, he does; and does. In saying just this much, I have not said enough to determine whether Fred was weak-willed. The problem is what to add to my description in order to make it true that the explanation of Fred’s action is that his will was weak. Our answer is that it depends on what Fred’s values were before the onset. If his prior values, when taken along with the change in his subjective probabilities consequent upon the onset, are enough—as revealed by some suitable instance of (P)—to make him want to smoke, or want to smoke more than not smoke, his action is not a manifestation of weakness of will. On the other hand, if the onset does more than can be got from prior values and posterior probabilities, he is rightly said to succumb to passion or whatever, in the way characteristic of weakness of will.

Thus we explain why it is right to say that the weak-willed act against their reason, and show how to meet the sceptical challenge mooted early in this chapter. Consider a fairly typical someone who values remaining slim and also satisfying hunger. The sceptical challenge was what more is there to say than that such a person will frequently find themselves in a conflict situation where one or the other value—not necessarily the same one on different occasions—will prove itself stronger by winning out. What more there is to say concerns the agent’s values when not hungry. 15

-195-



Notes 1 Reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980. 2 I am indebted to Gary Watson, ‘Skepticism about Weakness of Will’,

Philosophical Review, 86 (1977):316-39. 3 A subtle sceptical position is put by Watson in ‘Skepticism about Weakness of

Will’; also Davidson, ‘How is Weakness of Will Possible?’, rightly points out how reflection on weakness of will can easily lead to ‘absurd’ pictures of what is going on.

4 A Treatise of Human Nature, bk II, pt III, §3. 5 Added note: I included the parenthesis in the spirit of avoiding unnecessary

controversy; I happen to think that idle tongue clickings are typically caused by the belief that they will make one that little bit more at ease, along with the desire to be that little more at ease, though of course there is no conscious reviewing of either the belief or the desire.

6 Reprinted in Essays on Actions and Events. 7 A position of this kind is put by Don Locke, ‘Reasons, Wants and Causes’,

American Philosophical Quarterly, 2 (1974):169-79, and Thomas Nagel, The Possibility of Altruism, Oxford, Oxford University Press, 1970.

8 I have taken (P) from Richard C. Jeffrey, The Logic of Decision, New York, McGraw-Hill, 1965; see also David Lewis, ‘Causal Decision Theory’, Australasian Journal of Philosophy, 59 (1981): 5-30, see §2.

9 Incidentally, the choice between (P) and (P*) is a separate matter from the choice between classical and causal decision theory.

10 As in certain prima facie duty theories. 11 This distinction between wanting A to be the case, and wanting to bring about A,

will be familiar to any ‘two-boxer’ about Newcomb’s paradox. Two-boxers want it to be the case that they choose one box, but do not want to bring it about that they choose one box. See Lewis, ‘Causal Decision Theory’, and Brian Skyrms, Causal Necessity, New Haven, Yale University Press, 1980, p. 128 ff.

12 Reprinted in Essays on Actions and Events. 13 Added note: however, the arguments of Chapter 13, ‘On the semantics and logic

of obligation’, can be modified (in obvious ways) to show that this inference pattern is also invalid.

14 After writing this chapter, I read David Pears, ‘How Easy is Akrasia’, Philosophia, 2 (I982):33-50. In setting up the problem he says, ‘The idea is that your thoughts ought to control your actions and akrasia occurs when they do not control them. In a case of akrasia the action fails to fit the preparatory thoughts and is therefore irrational’, p. 33, my emphasis.

15 I am indebted to David Lewis, Elizabeth Prior (Jonson), and Peter Singer for forcing changes to earlier versions of this chapter.

-196-



13 ON THE SEMANTICS AND LOGIC OF OBLIGATION Some things ought to be: for some A, it ought to be that A. Some things are better than other things: for some A, B, A is better than B. How can we spell out what ought to be in terms of what is better, or worse?

There is an approach to this question which I will refer to as the standard one. (It is standard in the sense of being akin to a simplified core of what a number of authors say, and in validating the theorems of what is sometimes called standard deontic logic; it is not standard in the sense of being received wisdom—as far as I can see, nothing closely connected with deontic logic is received wisdom.) A’s being better than B does not entail that it ought to be that A. Cheating is better than killing, nevertheless it is false that cheating ought to be. What ought to be is not what is better, but what is best. The standard semantics seeks to capture this by interpreting ‘It ought to be that A’—O(A)—as true if and only if the best possible worlds are all A-worlds (worlds at which A is true). 1 Similarly, the standard semantics interprets ‘It ought to be that A given B’—O(A/B)—as true if and only if the best B-worlds are all A-worlds. I will assume some reasonable spelling out of the betterness relation between possible worlds without going into details.

There are many ways in which this simple semantics can be modified, elaborated and built upon. For instance, the possibility that there are no best worlds, or no best B-worlds, but rather an infinite series of ever better ones, can be covered by counting O(A) as true iff some A -world is better than any A̅-world, and by counting O(A/B) as true iff some AB-world is better than any A̅B-world. The possibility that the ranking of worlds is not absolute but varies from the perspective of different worlds, persons, or times can be covered by talking, not of the best worlds, but of the

-197-



best worlds from this or that perspective. The idea that in certain contexts the way things are in some worlds can be ignored—perhaps they are too remote from the actual world, or outside any agent’s control—can be covered by counting O(A) as true in such contexts iff the best worlds accessible in that context are all A-worlds, and similarly for O(A/B). But the points we will be making only require the simplified core, though we will be remarking on some of these developments of the simplified core in passing.

If the standard semantics for obligation and conditional obligation is right, the following two principles governing O(A) are obviously valid (and remain so if O(…) is replaced by O(…/ …) uniformly).

Entailment (Ent.) If A entails B, and O(A), then O(B).

Agglomeration (Agg.) If O(A) & O(B), then O(A&B).

The converse of Agg.—Distribution over &—follows from Ent.

As the literature on deontic logic amply testifies, these initially attractive principles are controversial. One aim of this chapter is to outline an alternative semantics for obligation on which Ent. and Agg. fail, and, most importantly, which makes it intuitive why they fail. The familiar possible world semantics for subjunctive conditionals makes it intuitive why strengthening the antecedent, contraposition, and hypothetical syllogism fail for subjunctive conditionals. For instance, strengthening the antecedent:

if then , is initially attractive when stated thus in symbols. There are, however, plausible counter-examples, and the case against it is clinched by how the possible-worlds semantics for subjunctive conditionals explains these counter-examples. (P&R may take you further from actuality than P.) 2 I want to do the same kind of thing with Ent. and Agg. For instance, we will see how the kind of examples mentioned under the heading of the Good Samaritan Paradox are on our semantics exactly what they have always seemed to be, counter-examples to Ent.

A subsidiary aim is to defend the naive account of conditional obligation, roughly, that

O(A/B) is equivalent to This account has seemed to face decisive counter-examples, mainly to do with Detachment. I will be arguing that a distinction



-198-

we make in the course of developing our semantics for O(A) enables us to meet these objections.

I should emphasise that our concern, and typically that of the standard semantics, is with the semantics of what ought to be, rather than of what a person at a time ought to do. We are seeking the right kind of account of when ‘It ought to be that…’ is true, where the ellipsis may be replaced by something of the form ‘S does A at t’; but equally may not be, as in ‘It ought to be that there are no famines’. Moreover, ‘It ought to be that S does A at t’ is not just another way of saying that S ought to do A at t. The former says that the state of affairs of S doing A at t ought to obtain; the latter that S—the agent—is under an obligation to do A at t; and only the latter arguably entails that S has the ability to do A. Indeed, the former might be true in part because it ought to be that S can do A, though most unfortunately she cannot. Thus, it ought to be that Kennedy survived being shot by Oswald. Nevertheless he could not, and was under no obligation to do so—surviving being shot is not something Kennedy ought to have done, though it is something that ought to have been. 3

There is, plausibly, a connection between what ought to be and what a person ought to do at a time, and I will say something about it at the end of this chapter, but until then our concern will be with what ought to be. (Even when I use the terms ‘obligation’ or ‘conditional obligation’. They should, perhaps, have been reserved for the ‘ought to do’ notion. But they have not; witness their frequent appearance in papers where the dots in ‘O(…)’ may be replaced by any consistent sentence.)

I will start in §13.1 with an objection to the standard semantics with a view to making plausible what I will call the Subjunctive thesis. In §13.2 I argue independently for the Relativity thesis. In §13.3 I combine the two theses to obtain our semantics. In §13.4 I make a brief comparison with dyadic deontic logic. §13.5 is about the failure of Ent. and Agg.; and in §13.6 and §13.7 I discuss, respectively, conditional obligation and ‘ought to do’.

13.1 An objection to the standard semantics

There is a certain arbitrariness in the standard semantics. It seeks to capture what ought to be the case by identifying it with what is the case in the best worlds. But what is the case in the best worlds will also be the case in many far from best worlds; in fact some things that are the case in the best worlds will also be the case in



-199-

the worst worlds. Suppose A is the case in both the best and the worst worlds. A will count as something that ought to be the case on the standard semantics. But that is to treat its appearance in the best as special, for it is just as true that it appears in the worst. That seems arbitrary. A more even-handed approach would regard A’s being the case in the best as at least possibly cancelled out by its appearance in the worst. Or suppose B is the case in the best worlds and also the case in some very ordinary worlds; while C is not the case in the best, but is the case in very many very good worlds, and moreover is never the case in any worlds less than very good. Does C’s good, even record outweigh B’s decidedly patchy one? On the standard semantics it is no contest: B’s appearance in (all) the best worlds settles the matter once and for all—B ought to be, C ought not to be. Again it seems a more even-handed and less arbitrary approach is called for, on which the matter is open given what has been said so far, and settlable given more information.

What kind of further information would settle the matter? Subjunctive information: information about what would be the case. 4 Consider ‘It ought to be that the West unilaterally disarms’. ‘The West unilaterally disarms’ is true in many very good worlds, for it is true in all the worlds where the West’s unilaterally disarming leads to a rapprochement between East and West; it is also true in many very bad worlds, for it is true in all the worlds where the West’s unilaterally disarming leads to a Russian invasion. These facts in themselves do not tell us whether it ought to be that the West unilaterally disarms; what we need in addition to know (and this is the hard bit) is what would be the case if the West unilaterally disarmed.

By the Subjunctive thesis, I mean the thesis that the worlds whose values matter in determining which propositions ought to be are the worlds which would be actual. Would be actual on what conditions? In order to answer this question, we need to note a way in which obligation is relative.

13.2 The relativity of obligation

As we noted in connection with the standard semantics, what ought to be is, in some sense, what is best. Now it ought to be that Bluebeard stops murdering his wives. But evidently it would have been better had he never started. How can we acknowledge

-200-



these two facts, while retaining the link between ‘ought’ and ‘best’?

There is a sense in which Bluebeard’s stopping is best. It is best out of his stopping and his keeping on. True, it would be better had he never started; but that does not alter the fact that out of stopping and going on, stopping is best. I will say, therefore, that the sense in which it ought to be that Bluebeard stops is relative to the set of alternatives {Bluebeard stops, Bluebeard goes on}; out of that set, it ought to be that he stops. The solution to the puzzle of how we can confidently and simultaneously judge that it ought to be that Bluebeard stops, and also that it ought to be that he never started (and so is in no position to stop) is that the set of alternatives to which the ought-judgement is relative is shifting. We judge that it ought to be that he stops, because that ought to be out of {he stops, he goes on}; we judge that it ought to be that he never started, because that ought to be out of {he stops, he never started}.

My contention is that our everyday judgements of what ought to be are all relative to sets of (mutually exclusive but not necessarily jointly exhaustive) alternatives; they are judgements of what ought to be out of {…}. The fundamental locution is ‘It ought to be that A out {A, A1, A2,…}’, which is true iff A is best out of {A, A1,…}. Of course, we typically say ‘It ought to be that A’ without specifying a set of alternatives. But there always is such a set in the offing, perhaps given implicitly by context or perhaps most simply it is {A,A̅}.

If this is right, we can explain an otherwise puzzling feature of our discourse about what ought to be.

Consider

(1) It ought to be that I catch spies.

We all give (1) our assent (we may suppose). But I catch spies if and only if there are spies and I catch some of them. So (1) says nothing different from

(2) It ought to be that there are spies and I catch some.

The puzzle is that (2) is so much less appealing than (1). (2), indeed, is the statement of the prima donna rather than the concerned citizen. (Similarly ‘I want to catch spies’ and ‘I want that

-201-



there are spies and I catch some’ do not provoke equal degrees of assent). 5

The explanation is that (2) is phrased in a way which suggests that the existence of spies is being presupposed or taken for granted in the context, for it is the case that out of the obvious set of alternatives each of which presupposes that there are spies, (1) is true. It ought to be that I catch spies out of {There are spies but I don’t catch any, there are spies and I do catch some}. On the other hand, writing out ‘I catch spies’ as an explicit conjunction—‘There are spies and I catch some’—suggests as the set of alternatives to be considered {there are spies and I catch some, there are no spies, there are spies and I don’t catch any}. Out of this latter set, it is ‘There are no spies’ that ought to be. Relative to any given set of alternatives, (1) and (2) have the same truthvalue; but relative to the sets most naturally suggested by the different phrasings in (1) and (2), (1) is true and (2) is false.

Incidentally, the disparity in intuitive response to (1) and (2) cannot be explained by holding that (1) is, while (2) is not, being read as implicitly conditional on ‘There are spies’. True, ‘It ought to be that I catch spies given there are spies’ strikes us as true; but ‘It ought to be that there are spies and I catch some given there are spies’ strikes us as false, as does ‘It ought to be that there are spies given there are spies’. (The fact that some theories of conditional obligation would make the latter true is an objection to them; I return to this point later.) An apparent disparity in truth-value is not explained by a difference in what is being conditionalised on, if the disparity remains when there is explicitly no difference.

Here is a more complicated example of the same sort of thing. I say, ‘It ought to be that Lucretia used less painful poisons’. You retort, ‘Oh no, it ought to be that Lucretia used painless poisons’. I then retort, ‘Oh no, it ought to be that Lucretia used political means rather than poison to obtain her ends’. You then retort, ‘Oh no, it ought to be that Lucretia never existed at all’. I then retort, ‘Oh no, it ought to be that Lucretia existed but made people happy’. You then retort, ‘Oh no, it ought to be that everyone was already happy so that there was no question of her making them happy’. And so on and so forth. Each retort seems a fair one; how so?

What is happening is that the set of alternatives to which the ‘ought’ is relative is being implicitly changed at each stage of the conversation. I started by saying ‘It ought to be that Lucretia used

-202-



less painful poisons’. We accept this by taking it as relative to {Lucretia used less painful poisons, Lucretia used the poisons she actually did}, and out of that set it is true. But your retort strikes us as fair by directing our attention to the set {Lucretia used less painful poisons, Lucretia used painless poisons}, and out of that set it is false that it ought to be that Lucretia used less painful poisons; it ought, out of that set, to be that she used painless poisons. And so on down the series of retorts. We have a series of statements of what ought to be, such that each taken by itself seems true, but taken with its successor seems false. Our explanation for this otherwise mysterious fluctuation in apparent truth-values is that the implicitly suggested reference class of alternatives is continually changing as we go along the sequence of statements of what ought to be.

Statements of what a person wants are implicitly relative in the same kind of way as statements of what ought to be. Consider ‘I want to die painlessly’. Is it true or false? Of course it is true; no one wants to die painfully. Of course it is false; we would all rather not die at all. That is, what is true is that, out of {dying painfully, dying painlessly}, I want to die painlessly; while out of {dying painlessly, not dying}, I want not to die.

The view that ‘It ought to be that A out of…’ is the fundamental notion does not of course prevent us introducing a non-relative notion defined in terms of it, thus: it absolutely ought to be that A iff for every set of alternatives which includes A, it ought to be that A out of that set. It is perhaps this notion we have in mind when we say ‘What really ought to be is that God’s in his heaven, and all’s right with the world’. (It is perhaps this notion, or something like it, that at least some writers on deontic logic have had in mind, though their examples are not of this kind but of the kind given throughout this chapter.) It is not, though, what we have in mind when we affirm that it ought to be that people are more considerate of the feelings of others, that Bluebeard stops murdering, that spies are caught, and so on. It is, that is, not what we normally have in mind.

I intend the thesis that obligations—non-absolute ones, the ones we normally have in mind when we say ‘It ought to be that…’—are relative to sets of alternatives, to apply as much to conditional obligations as to unconditional ones. Suppose I am not sure whether spies are bad or not; then I may not be prepared to say ‘It ought to be that I catch spies’, but may be prepared to say ‘It ought to be that I catch spies given spies are bad’. I will

-203-



not, though, affirm ‘It ought to be that there are spies and I catch some, given spies are bad’. The explanation for the difference parallels that given in the unconditional case. Given spies are bad, it ought to be that I catch spies out of {there are spies but I don’t catch any, there are spies and I do catch some}; while given spies are bad, it ought to be that there are no spies out of {there are spies and I catch some, there are no spies, there are spies but I don’t catch any}. Or suppose that you and I have a shrewd suspicion that Lucretia’s victims deserved all that they got, then our earlier dialogue might have gone thus: I say, ‘Given Lucretia’s victims were innocent, it ought to be that Lucretia used less painful poisons’. You retort, ‘Oh no, given they were innocent, it ought to be that she used painless poisons’. And so on and so forth. Again, what is happening is that the set of alternatives to which the ‘ought’ is relative is being implicitly changed at each stage in the conversation, but this time the ‘ought’ is a conditional one.

Relative obligation concerns what ought to be out of a range of alternatives—what is best out of that range. Conditional obligation also concerns what is best out of some range, but what is best out of that range given the truth of some proposition or other. Unfortunately, we can use the word ‘given’ in English for both, and this may lead to confusion. Suppose Attila and Genghis are both innocent; then it ought not to be that Attila goes to jail (out of his going and not, it is best he not); likewise, it ought not to be that Genghis goes to jail. But suppose Genghis has a large dependent family, while Attila is single; then it might be that out of Attila and Genghis, it ought to be Attila who goes. And we might say this in English using ‘given’, as in ‘Given one of Attila and Genghis is to go to jail, it ought to be Attila’. But at the same time we might say ‘Given one of Attila and Genghis is to go to jail, we know (because neither ought to) that a person who ought not to go to jail is going to go to jail’. We are not contradicting ourselves. The first ‘given’ expresses relative obligation: out of Attila and Genghis, it ought to be Attila; the second ‘given’ expresses conditional obligation. No doubt given the proposition that Attila is guilty, it ought to be that he goes to jail, but merely given the disjunction that he or Genghis in fact is to go, it ought not to be that he goes. That would be a blatantly wrong derivation of ‘ought’ from ‘is’.

(A similar confusion over ‘given’ may arise with certain ‘ought-to-do’ examples sometimes raised under the heading of

-204-



‘Contrary-to-duty-imperatives’. Suppose a Republican tells me ‘You ought to vote for Reagan, but given that you are not going to, you ought to vote for Mondale, not Jackson’; and suppose I in fact vote for Mondale. Then the Republican can say both ‘Given you didn’t vote for Reagan, you voted for whom you ought to have, namely Mondale’, and ‘Given you didn’t vote for Reagan, you voted for someone you ought not to have, namely someone other than Reagan’. We unravel the Republican’s position thus: the sense in which he holds ‘Given you don’t vote for Reagan, you ought to vote for Mondale’ is true is as a statement of relative obligation: out of the obvious non-Reagan alternatives, that is, out of Mondale and Jackson, you ought to vote for Mondale; but read as a statement of conditional obligation he holds it is false (my voting for Mondale would not make him better than Reagan), and is to be contrasted with a truth like ‘Given Mondale will do a better job, you ought to vote for Mondale’.)

Notice that though out of {Attila goes, Genghis goes}, it ought to be that Attila goes; it may well be that out of {Attila goes, Genghis goes and is fully compensated, Genghis goes and is not fully compensated}, it is not the case that it ought to be that Attila goes. Out of that new set of alternatives, it ought to be that Genghis goes and is fully compensated. This makes it transparent that what ought to be out of, or relative to, a set of alternatives is not the same as what ought to be given the proposition that one or other of these alternatives is the case: for the proposition that A or B is one and the same as the proposition that A or BC or B C̅.

I have struck the following reaction to this argument. ‘But once I’ve conceded that out of {Attila goes to jail, Genghis goes and is fully compensated, Genghis goes but is not fully compensated}, it is not ‘Attila goes to jail’ that ought to be, I must revise my earlier judgement that out of {Attila goes to jail, Genghis goes}, it ought to be that Attila goes.’ But this way lies the abandonment of almost all judgements of what ought to be out of a set of alternatives. Take any two alternatives A, B, such that you judge that out of them, it ought to be that A. There will nearly always be something, C, which, so to speak, more than neutralizes B’s nasty side. Thus out of {A, BC, BC̅}, it is BC that ought to be. Out of {I do not get the sack, I do get the sack}, it ought to be that I don’t; but out of {I do not get the sack, I do get the sack but am immediately offered a better job elsewhere,…}, it is not the case that it ought to be that I do not get the sack.

The obvious objection to treating relative obligation as

-205-



fundamental is that we need more than a relative notion when we come to consider what we ought to do. A similar point has often been made about probability. If I want to know whether to bet on Phar Lap, I need to know how probable it is that he will win. How probable it is that he will win given…is in itself of no use, because for every horse in the race there is something given which it is highly probable that that horse will win. In the probability case we solve the problem by picking out a certain conditional probability as special, namely that conditional on total evidence. Similarly, when we come to discuss later the relation between what S ought to do at t and what ought to be the case, we will in effect be picking out one particular relative ‘ought to be’ as special, namely and roughly, that relative to the set of options for S at t.

13.3 Our semantics

I have argued that the values that particularly matter for what ought to be are those of what would be; and that what ought to be is what ought to be out of…; we put these two ideas together thus: it ought to be that A out of {A, A1,…} iff what would be the case were A true is better than what would be the case were Ai-true, for all i. If we assume excluded middle for subjunctive conditionals, then there is exactly one world that would be actual if A were true, and likewise for A1, A2,…; call them, respectively, the closest A-world, the closest A1-world,…. Then our account of ‘It ought to be that A, O(A), out of {A, A1,…}’ is that it is true iff the closest A-world is better than the closest Ai-world, for all i.

Notoriously, there are some A for which it seems hard, not just in practice but in principle, to say exactly which world would be actual were A true: had Bizet and Verdi been compatriots,…? There are a number of ways of responding to this problem. We might follow Robert Stalnaker’s line and say that the puzzle arises simply because similarity is vague, and that for any acceptable elimination of that vagueness, there is exactly one closest world for each A. 6 We could then frame our truth-conditions for obligation initially as conditions for truth relative to an acceptable elimination of the vagueness of similarity, and go on to count ‘O(A) out of {A, A1,…}’ as true simpliciter just if it is true relative to every acceptable elimination of vagueness. An approach more in the spirit of Lewis is to admit the vagueness

-206-



of similarity while locating the essential source of our difficulty in the possibility of ties for similarity and of ever greater similarity without end. 7 We could then count ‘O(A) out of [A, A1, …}’ as true iff there is some degree of similarity such that, within that degree, every A-world is better than every Ai-world. A third approach is to say that for each A, either there is a single world that would be actual, or there is a set of worlds each of which might be actual but none of which would be, were A the case. We would then count ‘O(A) out of {A, A1,…}’ as true iff every world that might be actual were A true is better than every world that might be actual were Ai true, for all i.

None of the points in what follows depends on which of these approaches, or (reasonable) variants of them, we take. For our purposes it will be enough to say vaguely that it ought to be that A out of a set of alternatives A, A1,…just as if what would be the case were A the case is better than what would be the case were Ai true, for each i. I will sometimes refer to what would be the case were X the case as the closest X-world, but the point being made will not depend crucially on the dubious assumption that there is always such a thing as the closest world.

The idea that in considering whether it ought to be that A, we should look particularly to what would be the case were A the case is intuitively plausible. Suppose all the best worlds, or all the best accessible worlds, or all the deontic alternatives to the actual world, are A-worlds. So what, if they would not be actual were A true. How then can their virtue reflect back on A? But notice how crucial treating obligation as relative to sets of alternatives is here. If we looked merely to what would be the case were A the case, all we could say was how many Brownie points it scored (and any two truths would score the same, namely that of the actual world); it is not until we compare its score with that of each of a set of alternatives that we get a handle on the question as to whether A ought to be, that is, on whether A ought to be out of that set.

We noted earlier that it may be that, out of {A, B}, it ought to be that A; while out of {A, BC, BC̅}, it ought to be that BC. We can now say how this can happen. Strengthening the antecedent for subjunctive conditionals is a fallacy. It may be that what would be the case were A the case is better than what would be the case were B the case; yet what would be the case were BC the case is better again. 8

-207-



13.4 Dyadic deontic logic

There is a connection between our semantics and one of the motivations for dyadic deontic logic. The standard semantics for ‘It ought to be that A’ is implausible when applied to actual assertions concerning what ought to be. When Jesse’s mother learns that he has robbed the bank and says, ‘It ought to be that he returns the money’, she is not saying that all the best worlds are ones where he returns it. The best worlds are ones where he did not take the money to start with, or perhaps ones where he never existed, or ones with angels and no money and no banks, or…. 9

One of the reasons for introducing the dyadic operator ‘It ought to be…given…’ was the recognition of this fact. 10 Consequently, it is suggested that his mother is really saying that it ought to be that Jesse returns the money given how he got it, 11 and, the suggestion goes, this counts as true iff the best worlds where he got it in the way he did, by stealing, are ones where he returns the money. But suppose that were Jesse to return the money, the sight of the money’s unexpected return would give the bank manager a heart attack. Then it is plausible that it ought not be that the money is returned. Nevertheless, it may be that all the best worlds where Jesse stole the money are all ones where he returns it; those worlds also being, of course, ones where the bank manager has a stronger heart than he actually does. It seems, that is, that what is crucial is what would be the case were Jesse to return the money.

Or perhaps we were too hasty in declaring the best possible worlds where Jesse steals to be ones where he returns the money. Perhaps the very best worlds are ones where money does not matter, and so the best worlds where Jesse steals the money are not all ones where he returns it—in some he does, in some he does not; it makes no difference one way or the other. But this possibility is academic. Someone who replied to Jesse’s mother by raising it would get short shrift. What is important is what is and would be the case, and money does and still would matter whatever Jesse did.

The idea behind the appeal to a dyadic construction in order to achieve applicability to actual assertions of what ought to be is that we should restrict the worlds we look at. I am not objecting to this. What I am suggesting is that if we follow this idea through, we end up by looking at what is and would still be

-208-



the case—just as we do in our semantics. What dyadic deontic logic initiated plausibly ends up where we are. We can make this more obvious by stating our account thus: it ought to be that A out of {A, A1,…} iff the best worlds w such that

is actual, w is actual, or…, are all A-worlds. (Not the worlds w such that A v A1 v…is true at w.)

There is, however, an important difference between our way of restricting the worlds to be looked at and the various ways suggested by those working within the general framework of the standard semantics. Our way alone makes the worlds to be looked at depend on the proposition whose obligatory status is in question. It has variously been suggested that in determining whether A is obligatory, perhaps in some ‘context’, we look at whether A holds throughout all the best worlds where some implicitly or explicitly given condition holds, or where the past is exactly the same as it is in (say) the actual world, or which are like the actual in background circumstances, or which are a possible scenario for a given agent at a given time, or which are achievable, and so on. 12 But in each case, A in itself does not play a role in settling which worlds must all be A-worlds in order for the obligation statement to be true: replacing A by B with everything else held constant leaves unaltered the ‘deontic alternatives’ throughout which the proposition must be true in order for it to be obligatory. And it is, of course, this feature of these approaches which is responsible for their being alike in validating the appropriate form of Ent., and so in being in this respect unlike our account, as we will now see.

13.5 Obligation fallacies

Before we can discuss Ent. and Agg., we need to resolve which class of alternatives the obligations in these principles are relative to. Unless explicit contrary notice is given, obligation will be construed as relative to the class consisting of the proposition in question and its negation. Thus ‘It ought to be that X’—‘O(X)’—is going from now on to be read as ‘It ought to be that X out of {X, X̅}’, which is true iff the closest X-world is better than the closest X-world. (Accordingly, we will not have ‘O(X) v O(X̅)’ as a thesis, for the closest X-world and the closest X-world might be on a par but we will have ‘O(X)⊃~O(X̅)’ as a thesis, given the asymmetry of ‘better’.)

I think that typically when we assert that it ought to be that A,

-209-



we mean that it ought to be that A out of A and its negation, that it is better that A be true than that it be false. Context, tone of voice, selection of words, explicit avowal (of course) may produce exceptions to this rule, but it is nevertheless the norm.

Now consider

Ent. If A entails B, and O(A), then O(B).

As indicated earlier, I want both to give a plausible counter-example to Ent. and to show how it is one on our semantics. Virtually any ‘Good Samaritan’ type of counter-example will do. For instance, it ought to be that Smith’s broken leg heals; ‘Smith’s broken leg heals’ entails that Smith has a broken leg; but it is false that it ought to be that Smith has a broken leg. Suppose that Smith actually has a broken leg which heals, then what makes it true that it ought to be that his leg heals is that what is actually the case is better than what would have been the case were it false that it healed; for had it been false that it healed, it still would have been broken. And this is quite consistent with the fact that what would have been the case were his leg not broken in the first place is better than what is actually the case: a broken leg which heals is better than a broken leg which does not, while being worse than no broken leg at all.

I think that there are two reasons why Ent. has been so widely accepted. One is the belief that only by retaining it can we ensure that it is always possible for a person to do everything that they ought. 13 We will see later, when discussing ‘ought to do’, how to meet this point. The other is the belief that, as Peter Schotch and Raymond Jennings put it, ‘by means of this axiom [in effect Ent.] we may persuade moral agents that they are committed to the logical consequences of their moral principles’. 14 Now it is true that a Kantian who asserts that lying is always wrong is committed to holding that lying by left-handers is always wrong. But we do not need Ent. to show that; we need Quantification Theory. 15 I do, of course, hold that ‘It ought to be that lying never happens’ does not entail ‘It ought to be that lying by left-handers never happens’. And that is plausible. If left-handers stopped, that would not make the right-handers stop (perhaps it ought to). Consequently, the left-handers would be unfairly dis-advantaged in the wheeler-dealing that (unfortunately) characterises much of life. Thus, it is consistent to hold that it ought to be that left-handers sometimes lie—for if they did not, the still

-210-



lying right-handers would take over—while holding that it ought to be that no one ever lies. Or take a real-life example. I think it is obviously true that it ought to be that no country exports uranium, but it is not obviously true that it ought to be that Australia does not export uranium. It may be true, but it is debatable in a way the first is not.

Now consider

Agg. O(A) & O(B)⊃O(A&B).

Attila and Genghis are driving their chariots towards each other. If neither swerves, there will be a collision; if both swerve, there will be a worse collision (in a different place, of course); but if one swerves and the other does not, there will be no collision. Moreover, if one swerves, the other will not because neither wants a collision. Unfortunately, it is also true to an even greater extent that neither wants to be ‘chicken’; as a result, what actually happens is that neither swerves and there is a collision. It ought to be that Attila swerves, for then there would be no collision. Genghis would drive straight on shouting ‘chicken’, though secretly mightily relieved. Equally, it ought to be that Genghis swerves. But it ought not to be that both swerve, for then we get a worse collision. Thus we have O(A) and O(B) without O(A&B).

In our terms, what is happening in this counter-example to Agg. is that we have O(Attila swerves); because it would be better had Attila swerved than not, for then Genghis still wouldn’t have, and all would have been well. Analogously, we have that O(Genghis swerves). We do not though have O(Attila swerves&Genghis swerves). For had they both swerved, there would have been a worse collision.

There is a tempting objection to this example. Why should Attila swerve rather than Genghis; or equally, why Genghis rather than Attila? The temptation is to object that it ought to be that one or the other swerve; but true of neither that it ought to be that he swerve, though true of both that it is permissible that he swerve. There is an interesting puzzle here. Focus on Attila, and it seems clear that it ought to be that he swerves; concentrate on Genghis, and it seems clear that it ought to be that he swerves; look at the two together, and it seems that what ought to be is that one or the other swerve, without it being the case that it ought to be Attila or that it ought to be Genghis. The solution is the relativity of obligation to sets of alternatives. Out of {Genghis

-211-



swerves, ~Genghis swerves}, it ought to be that Genghis swerves; similarly for Attila. But out of {Genghis swerves, Attila swerves}, it is neither the case that it ought to be that Genghis swerves, nor the case that it ought to be that Attila swerves.

13.6 Conditional obligation

We may be interested in the effect on obligation of the possible obtaining of X; perhaps A ought to be out of…, but were X the case, A ought not to be out of…. Call this: whether it ought to be that A out of {A, A1, . . .} given X, or O(A/X) out of {A, A1, …}. Instead of counting it as true if the world that would be actual were A true is better than the world that would be actual were Ai true, for each i, we count it as true if, were X true, the world that would then be actual were A true is better than the world that would then be actual were Ai true, for each i. Many have urged the

plausibility of construing O(A/X) as , and our account is squarely in this tradition. 16

Why have I given the account of conditional obligation in terms of comparing truth-

making fillings in

,…,instead of in

, …? That is, why construe ‘O(A/X) out of {A, A1…}’ in terms of

and not ‘O(AX) out of {AX, A1X, …}’ ? One reason is that I want ‘O(~A/A)’ to be sometimes true and sometimes false. (As before, O(X/Y) is to be read as ‘O(X/Y) out of {X, X}’, unless explicit contrary notice is given). I insist that ‘It ought to be that I do not have cancer given I have cancer’ is true; and equally, that ‘It ought to be that I am dead given I am alive’ is false. Now subjunctive conditionals of the form

may be determinately true or determinately false, but those of the form



are either all trivially true or fall into the ‘make your own decision’ category. However, O(~A/A) is a special case, intuitions about it are perhaps somewhat suspect, so let me make a similar point in terms of a less special case.

Consider ‘Given that Genghis drank ten doubles of whisky a day, it ought to be that he drank nothing’, and ‘Given that Genghis drank two pints of beer a day, it ought to be that he drank nothing’. On plausible assumptions, the first is true and the second is false. The reason the first is true is that were it true that Genghis drank ten doubles a day, it would be true that he had a serious drinking problem, in which case it would be best if he

-212-

drank nothing. Here we are considering which subjunctive conditionals with ‘Genghis drinks nothing’ as antecedent would be true were it true that he drinks ten doubles a day, and judging on the whole that ones with good consequents would be true. We are not considering subjunctive conditionals with the absurd antecedent ‘Genghis drinks nothing and ten doubles a day’.

Thus, sometimes when considering ‘It ought to be that A given B’, we need to consider possibilities where B is false—pace one of the intuitions lying behind the standard semantics for O(A/B). 17 Here is another example: ‘It ought to be that I play some sport given that I engage in no physical activity at all’. We judge this true by considering how I would be were I totally sedentary, and judging that in that case it would be better that I were not. But I grant that these cases are unusual. Typically, B is such that it would still be the case whether or not X or its alterntives are—consider ‘I ought to go to the party given I have promised to’; were I to have promised, I still would have whether or not I fulfil my promise—and in such cases when comparing A with its alternatives were B the case, we will in effect be comparing AB with A1B, A2B, and so on. Possibilities where B̅ will not come into it.

Our account allows O(A/A), as well as O(~A/A), to be sometimes true and sometimes

false, for is contingent if A is contingent. This seems right. ‘It ought to be that I tell the truth given I tell the truth’ seems true, while ‘It ought to be that Hitler exterminated millions of Jews given he exterminated



millions of Jews’ seems false. By contrast, the standard view makes O(A/A) always true.

The implausibility of this latter consequence undermines one way of handling the Good Samaritan Paradox often suggested by defenders of Ent. For instance, Schotch and Jennings suggest that we respond to the fact that ‘It ought to be that we feed the starving poor’ seems true, while ‘It ought to be that there are starving poor’ seems false, by holding that what is really true is that it ought to be that we feed the starving poor given or provided that there are starving poor. 18 However, defenders of Ent., including Schotch and Jennings, intend it to apply to conditional obligation as much as to unconditional obligation. Thus, from what they concede to be really true, it follows according to Ent. that it ought to be that there are starving poor given that there are starving poor. I doubt if the starving poor would agree.

The conspicuous feature of our account of conditional

-213-

obligation is that it allows unrestricted detachment (for a given set of alternatives).

This follows immediately from modus ponens for . Detachment is plausible. If it ought to be that Attila goes to jail given that he has raped and pillaged, and he has raped and pillaged, then it ought to be that Attila goes to jail. There are, however, apparent counter-examples to Detachment, and so to the naive account of conditional obligation. I will consider two.

It ought to be that Jesse returns the money given he has robbed the bank. I learn that he has robbed the bank. It is contended that it is wrong to infer that it ought to be that he returns the money. What ought to be is that he has no stolen money to return. But all (everyday) obligation statements, conditional and unconditional, are relative. When I learn of Jesse’s guilt, I do indeed learn that it ought to be that he returns the money out of {he returns it, he doesn’t}. The appeal of a remark like ‘What really ought to be is that he never stole it in the first place’ is that it shifts our attention to a different set of alternatives, namely {Jesse returns the money he stole, Jesse does not return the money he stole, Jesse never stole}. Out of this latter set, but not out of the former, it is false that it ought to be that he returns the money; what indeed ought to be is that Jesse never stole. Moreover, out of this latter set, the statement of conditional obligation is false. Given Jesse stole, out of the set including that he did not, it ought



to be that he did not. Hence, for no given set of alternatives does the example show that Detachment fails.

An instance of the second kind of example offered against Detachment is that ‘It ought to be that it is thumbscrews given it is thumbscrews or the rack’ is clearly true (we may suppose), yet we do not infer that it ought to be thumbscrews, on learning—to our horror—that it is thumbscrews or the rack. Whatever will in fact happen, it ought not to be any kind of torture—that is for sure.

The mistake in this argument is reading ‘It ought to be that it is thumbscrews given it is thumbscrews or the rack’, in the sense in which it strikes us as true, as a statement of conditional obligation. Rather, in the sense which strikes us as true, it is a statement of relative obligation. For consider ‘It ought to be that it is thumbscrews given it is thumbscrews or the rack very mildly or the rack not very mildly’. That strikes us as just as clearly false as the other statement struck us as true. But the proposition that it is thumbscrews or the rack is the very same proposition as that it is thumbscrews or the rack very mildly or the rack not very mildly.

-214-

Therefore, were the two statements being read as ones of conditional obligation, their truth-values would have to be the same, because no change has taken place in the proposition the obligation would be conditional on. What has, of course, changed is the delineation of the particular alternatives. Thus the sense in which ‘It ought to be thumbscrews given it is thumbscrews or the rack’ is judged true is as a statement of relative obligation. Out of the thumbscrews and the rack, it ought to be thumbscrews. But out of the larger number of alternatives, it ought to be the rack very mildly.

13.7 ‘Ought to do’ from ‘Ought to be’

Our first concern has been with the semantics of what ought to be. But our interest in what ought to be becomes particularly pressing when it relates to what a person ought to do at a time. Thus I should say something, if only an outline something, about how to derive a person’s obligations at a time from what ought to be. I will outline the simplest way of doing this. My concern here is with the general picture, not the detail. 19

I will take for granted the notion of a person’s options, the courses of action open to them, at a time. There are, however, two times to bear in mind: the time at which they



are options, and the time or period of time of the possible actions which constitute the options. On Monday morning, my options for Tuesday evening and night may be to go to the opera and then to bed, to go to the opera and then to an all-night party, and not to go to the opera but still to go to the party. But if by Monday lunch-time I have not bought a ticket and moreover the tickets have sold out, then on Monday afternoon my options for Tuesday evening and night no longer include going to the opera and then to the party, but still include not going to the opera and going to the party. Again, when someone on a diet freely throws the cheesecake out of the window, his options beforehand include eating the cheesecake in the near future, while his options afterwards do not.

I will use ‘at t’ for the time at which some possible action or course of action is an option, and ‘on T’ for the time of the possible action or course of action. The simplest way of deriving ‘ought to do’ from ‘ought to be’ is then: if A, A1,…are the possible actions on T which are options for S at t, then S at t ought to do A on T iff it ought to be that S does A on T out of {S does A on T, S does A1 on T,…}.

-215-

This simplest account forces us into a controversial position concerning examples like the following. 20 Dr Procrastinate is asked in January to review a book. If he accepts, the book will reach him in February; he is to write the review in March. He has the time, is the right person to do the review, and so on. What ought Procrastinate do? Say yes in January and write the review in March. That is the best option available to him. Suppose, however, he knows that he will not write the review in March. Not because something unexpected and out of his control crops up, but because he will culpably keep on putting off the task. Suppose also he knows that April, June, July,…will not do. It is March or the bush. And furthermore he knows that if he says no in January, there is a good chance that someone else will agree to do the review, while if he says yes in January but does not write the review in time, the book will miss being reviewed altogether. In short, out of what is possible for him, what is best is to say yes in January and write in March (despite the fact that this will not happen); what is next best is to say no in January; and what is worst is to say yes in January and fail to write in March. Therefore, on our view (and this is independently plausible), he ought to say yes in January and write in March. However on our view it is also true that he ought to say no in January. Out of the two relevant options for him in January, namely, saying yes and saying no, saying no is what ought to be. For what would be the case were Procrastinate to say yes would be worse than what would be the case were he to say yes: what would be the case were he to say yes in January is that the book would miss being reviewed altogether.



I am therefore committed to denying Ent., in fact a special case of it, distribution over &, for ‘ought to do’, as I did for ‘ought to be’. I must deny that ‘S ought to do A&B’ entails that S ought to do A. 21 I do not think this is an objection. The case of Procrastinate is a plausible case where distr. over & fails, and we have to hand a plausible semantics which explains how it fails. This is exactly the kind of situation which warrants rejecting a logical principle despite its initial appeal.

Moreover, the alternative position that Procrastinate ought to say yes in January has an arbitrary character of a kind we have objected to before. Saying yes in January is indeed an essential constituent in the best available course of action that starts in January and finishes in March, namely, saying yes in January and writing in March; but it is also an essential constituent in the

-216-

worst available course of action over that period, namely, saying yes in January and not writing in March. These two facts in themselves would appear to balance each other out—insisting that the first fact in itself settles the matter in favour of the view that Procrastinate ought to say yes seems arbitrary. On our view, what settles the matter are the two facts along with the fact that were Procrastinate to say yes, it would be the worst possible result and not the best that would obtain—that settles that Procrastinate ought not to say yes (and ought to say no).

Am I letting Procrastinate off too lightly? 22 I say that he ought to say no in January, and also that he ought to say-yes-in-January-and-write-in-March. He cannot do both obviously; so it may seem that I must hold that if he says no in January, he is doing about as well as he can. This is wrong. He could do everything he ought simply by saying yes in January and writing in March (as he could). The only reason he ought to say no in January is that were he to say yes, he would not write, but ‘Were he to say yes, he would not write’ would be false if he said yes and wrote. Procrastinate violates the second-order moral principle that an agent ought to make it the case that he does all that he ought.

The position I am advancing allows that it may be that a person ought to do A followed by B, yet ought to do not A. It does not, though, allow that it may be impossible for an agent to do everything they ought. Of course it is impossible both to do A followed by B, and to do not A. But these two actions can only be both what an agent ought to do if (i) it is the case that were the agent to do A, they would not do B, and (ii) it is within the agent’s power to do A followed B, so making (i) false and



making it the case that they ought to do A rather than not A. In short, though a person’s set of obligations may be inconsistent, if it is, it is within their power to change the set to a set which is consistent. Hence, it is within their power to bring it about that they do all they ought. Nothing I am saying stops us from leading blameless lives. 23

Notes 1 See, e.g., Dagfinn Føllesdal and Risto Hilpinen, ‘Deontic Logic: An Introduction’

in Føllesdal and Hilpinen, eds, Deontic Logic: Introductory and Systematic Readings, Dordrecht, Reidel, 1971. Jaakko Hintikka, ‘Some Main Problems of Deontic Logic’ in Deontic Logic: Introductory and Systematic Readings, p. 71, calls the worlds that must all be A-worlds in order for O(A) to be true at w deontic

-217-

alternatives to w, and describes them as kinds of deontically perfect worlds. Our terminology is closest to David Lewis, Counterfactuals, Oxford, Basil Blackwell, 1973, and ‘Semantic Analyses for Dyadic Deontic Logic’ in Søren Stenlund, ed., Logical Theory and Semantic Analysis, Dordrecht, Reidel, 1974.

2 See, e.g., Lewis, Counterfactuals. 3 For more examples, see I.L. Humberstone, ‘Two Sorts of “Oughts’”, Analysis, 32

(1971):8-11. 4 No ‘probably’, as it is what is sometimes called objective, not subjective,

obligation that is our concern. Added note: I argue in Chapter 14 ‘Decision-theoretic consequentialism and the nearest and dearest objection’ that the notion most central in ethics is the subjective (or better, expected value) notion of obligation, because that connects most closely with action; but here we are primarily concerned, as noted above, with what ought to be.

5 An example drawn to my attention many years ago by I.L. Humberstone; see also Richmond H. Thomason’s hands example in ‘Deontic Logic as Founded on Tense Logic’ in Risto Hilpinen, ed., New Studies in Deontic Logic, Dordrecht, Reidel, 1981, p. 170.

6 Robert Stalnaker, ‘A Defense of Conditional Excluded Middle’ in W.L. Harper et al, eds, Ifs, Dordrecht, Reidel, 1981.

7 Counterfactuals.



8 We also noted earlier that one might define ‘It absolutely ought to be that A’, O*(A), as true iff ‘It ought to be that A out of every set of A-alternatives’. It is easy to show from this definition and our semantics that if O*(A), then some A-world is better than any A̅-world. This connection between our semantics and more standard ones shows that we have the following consistency principle: if O*(A1)…& …& O*(An), then there is a possible world at which A1&…&An is true.

9 See Holly S. Goldman, ‘David Lewis’s Semantics for Deontic Logic’, Mind, 86 (1977):242-8.

10 See, e.g., the opening paragraph of Lewis, ‘Semantic Analyses for Dyadic Deontic Logic’.

11 Not ‘It ought to be that Jesse returns the money given he stole it’—his mother knew that before she knew he stole it.

12 Suggestions along one or other of these lines, but sometimes primarily focused on ‘ought to do’ rather than ‘ought to be’, are to be found in Bas Van Fraassen, ‘The Logic of Conditional Obligation’, Journal of Philosophical Logic, 1 (1972):417-38; I.L. Humberstone, ‘The Background of Circumstances’, Pacific Philosophical Quarterly, 64 (1983):19-34; P.S. Greenspan, ‘Conditional Oughts and Hypothetical Imperatives’, Journal of Philosophy, 72 (1975):259-76; Richmond H. Thomason, ‘Deontic Logic and the Role of Freedom in Moral Deliberation’ in New Studies in Deontic Logic; and Bengt Hansson, ‘An Analysis of Some Deontic Logics’ in Deontic Logic: Introductory and Systematic Readings.

13 See, e.g., Bas Van Fraassen, ‘Values and the Heart’s Command’, Journal of Philosophy, 70 (1973):5-19, p. 15. Evidence of the appeal of Ent. is that Van Fraassen in this paper wishes to retain it

-218-

while questioning both Agg. and, especially notable from our perspective, ‘O(A)⊃~O(A̅)’ .

14 Peter K. Schotch and Raymond E. Jennings, ‘Non-Kripkean Deontic Logic’ in New Studies in Deontic Logic, p. 151.

15 In particular, the principle that if every F is G, then every F which is H is G. Note that ‘X-ing is always wrong’ is not equivalent to ‘It ought to be that X-ing never happens’. Only pacifists think that waging war is always wrong; but many think that it ought to be that waging war never happens—for then the justified wars of self-defence would never be called for.

16 E.g., Peter Mott, ‘On Chisholm’s Paradox’, Journal of Philosophical Logic, 2 (1973):197-211 and Brian F. Chellas, ‘Conditional Obligation’ in Logical Theory and Semantic Analysis.

17 See, e.g., Van Fraassen, ‘Values and the Heart’s Command’, p. 421.



18 See, e.g., Schotch and Jennings, ‘Non-Kripkean Deontic Logic’, p. 152. 19 For more details, see Holly S. Goldman, ‘Dated Rightness and Moral

Imperfection’, Philosophical Review, 75 (1976):449-87. But see Note 21 immediately following.

20 Added note: I have edited the next three paragraphs from the original publication to avoid confusing ways of saying things. The example and surrounding issues are discussed in detail in Frank Jackson and Robert Pargetter, ‘Oughts, Options, and Actualism’, Philosophical Review, 95 (1986):233-55.

21 Incidentally, (i) the fact that B follows A in time in our example against Distr. over & is not essential to the example, though it makes it easier to grasp, and (ii) our denial insulates us from an objection P.S. Greenspan (‘Oughts and Determinism’, Philosophical Review, 77 (1978):77-83) makes to Goldman, and which was partly responsible for Goldman herself having second thoughts—see her ‘Doing The Best One Can’ in A.I. Goldman and J. Kim, eds, Values and Morals, Dordrecht, Reidel, 1978.

22 This is, I think, what Thomason would say; see ‘Deontic Logic and the Role of Freedom in Moral Deliberation’.

23 I.L. Humberstone and David Lewis commented on an early version of this chapter. I am heavily indebted to them but they should not be held responsible.

-219-

14 DECISION-THEORETIC CONSEQUENTIALISM AND THE NEAREST AND DEAREST OBJECTION Our lives are given shape, meaning and value by what we hold dear, by those persons and life projects to which we are especially committed. This implies that when we act we must give a special place to those persons (typically, our family and friends) and those projects. But according to consequentialism classically conceived, the rightness and wrongness of an action is determined by the action’s consequences considered impartially, without reference to the agent whose actions they are consequences of. It is the nature of any particular consequence that matters, not the identity of the agent responsible for the consequence. It seems then that consequentialism is in conflict with what makes life worth living.



I take this to be one part of Bernard Williams’s well-known attack on consequentialism. 1

One way to reply to it would be to break the implicit connection between acting morally and living a life worth living. Doing what is morally right or morally required is one thing; doing what makes life worth living is another. Hence, runs the reply, it is no refutation of a moral theory that doing as it enjoins would rob life of its shape and meaning.

This is a chilling reply and I will say no more about it. My reply will be that consequentialism—properly understood—is perfectly compatible with the right actions for a person being in many cases actions directed towards achieving good consequences for those persons and projects that the agent holds dear. Consequentialism, I will argue, can make plausible sense of the moral agent having and giving expression in action to a special place for

-220-

family, friends, colleagues, chosen projects, and so on and so forth.

I will start by explaining how, in my view, consequentialism should be understood. This explanation will help us sort out certain potential confusions as well as providing the springboard for our reply to the nearest and dearest objection.

I should emphasise that I claim no great originality for my account of consequentialism. It is, I think, a natural extension of what, for instance, J.J.C. Smart had in mind all along (shorn of the commitment to a utilitarian construal of consequences), though he did not say it quite the way I will. 2 I hope that my way of putting things will make certain matters clearer.

14.1 Understanding consequentialism

Consequentialism approaches the question of whether an action is right or wrong in terms of a comparison of the possible outcomes of the action with the possible outcomes of each available alternative to that action. The notion of a possible outcome of an action is interpreted so as to include the action itself, and the comparison of the various outcomes is carried out in terms of a consequentialist value function. The interesting question of exactly what makes a value function warrant being described as consequentialist can here be left to one side. The details



of the value function will not particularly concern us, and any reasonable ranking of outcomes of the usual agent-neutral kind will serve our purposes in what follows. Similarly, exactly how the available alternatives to the action in question are specified can be left vague. What will, however, concern us is how the values assigned to the outcomes feed into the determination of what ought to be done. We will be presupposing that the matter is approached in the usual maximising way—classical consequentialism is our subject, not satisficing varieties thereof—but that in itself leaves a major issue open, an issue which will turn out to be crucial for the argument of the chapter. This major issue can be most easily approached via a simple example.

The drug example, mark I

Jill is a physician who has to decide on the correct treatment for her patient, John, who has a minor but not trivial skin complaint. She has three drugs to chose from: drug A, drug B, and drug C.

-221-

Careful consideration of the literature has led her to the following opinions. Drug A is very likely to relieve the condition but will not completely cure it. One of drugs B and C will completely cure the skin condition; the other though will kill the patient, and there is no way she can tell (beforehand) which of the two is the perfect cure and which the killer drug. What should Jill do?

The possible outcomes we need to consider are: a complete cure for John, a partial cure, and death. It is clear how to rank them: a complete cure is best, followed by a partial cure, and worst is John’s death. That is how Jill does, and also how she ought to, rank them. But how do we move from that ranking to a resolution concerning what Jill ought to do? The obvious answer is to take a leaf out of decision theory’s book and take the results of multiplying the value of each possible outcome of each contemplated action by Jill’s subjective probability of that outcome given that the action is performed, summing these for each action, and then designating the action with the greatest sum as what ought to be done. In our example, the issue will boil down to a comparison between:

Pr(partial cure/drug A taken).V(partial cure)+Pr(no change/ drug A taken).V(no change)



Pr(complete cure/drug B taken).V(complete cure)+Pr(death/ drug B taken).V(death) and

Pr(complete cure/drug C taken).V(complete cure)+Pr(death/ drug C taken).V(death).

Obviously, in the situation as described, the first will take the highest value, and so we get the answer that Jill should prescribe drug A. The obvious answer all along. The difference between a complete cure and a partial cure in the case of a minor skin complaint does not compensate for a significant risk of death, as we might say it in normal English.

Generalising, the proposal is to recover what an agent ought to do at a time according to consequentialism from consequentialism’s value function—an assignment of value that goes by total consequent happiness, average consequent preference satisfaction, or whatever it may be in some particular version of consequentialism—together with the agent’s subjective probability function

-222-

at the time in question in the way familiar in decision theory, with the difference that the agent’s preference function that figures in decision theory is replaced by the value function of consequentialism. That is to say, the rule of action is to maximise Σi Pr(Oi / Aj ).V(Oi ), where Pr is the agent’s probability function at the time, V is consequentialism’s value function, Oi are the possible outcomes, and AJ are the possible actions. We can express the idea in words by saying that whereas decision theory enjoins the maximisation of expected utility, consequentialism enjoins the maximisation of expected moral utility. 3

How else might one seek to recover what, according to consequentialism, a person ought to do from consequentialism’s value function? Two alternatives to the decision-theoretic approach call for discussion.

We can think of consequentialism’s value function as telling us what, according to consequentialism, we ought to desire. For a person’s desires can be represented—with, of course, a fair degree of idealisation—by a preference function which ranks state of affairs in terms of how much the person would like the state of affairs to obtain, and we can think of consequentialism as saying that the desires a person ought to have are those which would be represented by a preference function which



coincided with consequentialism’s value function. The other ingredient in the decision-theoretic account of what consequentialism says a person ought to do, the agent’s subjective probability function, is an idealisation of the agent’s beliefs. Hence, the decision-theoretic account is one in terms of what the person ought to desire and in fact believes. But, in addition to distinguishing what a person in fact desires from what he or she ought to desire, we also distinguish what a person in fact believes from what a person ought to believe. And in a sense of ‘ought’ which has a moral dimension—there is, for instance, such a thing as culpable ignorance. Hence, it might well be suggested that we should recover the consequentialist answer to what a person ought to do from the value function via what that person ought to believe rather than from what he or she in fact believes. 4

However, the clearest cases of culpable ignorance can be handled in terms of what a person in fact believes. The decision problem which faces a doctor considering whether to prescribe a certain drug is not simply the choice between prescribing the drug and not doing so—though we may pretend that it is that simple in order to make some point that is independent of the complexities

-223-

—it is more accurately described as the choice between: deciding now to prescribe the drug, deciding now against prescribing the drug, and postponing the decision until more information has been obtained and, on the basis of what one then knows, deciding between prescribing the drug and not prescribing the drug. Now, in the same way that prescribing the drug and not prescribing the drug have an expected moral utility, so does obtaining more information and doing what then has greatest expected moral utility; and we can investigate the conditions under which getting more information and doing what then has greatest expected moral utility has itself greater expected moral utility than either prescribing the drug or not prescribing the drug. It is easy to prove the following. Getting more information and then doing what has greatest moral utility has itself greatest moral utility provided the possible change in utility consequent on the new information when weighed by the probability of getting that new information is great enough to compensate for the effort and cost of getting the new information. 5 Thus, working solely with a person’s subjective probability function, with what he or she actually believes, we can distinguish plausibly between cases where more information ought to be obtained and where we may legitimately rest content with what we have. Hence, it seems to me at least arguable that our approach to what a person ought to do according to consequentialism in terms of what he or she ought to desire and does in fact believe, does not need to have the reference to what is believed replaced by a reference to



what ought to be believed. 6 However the bulk of what I have to say about the nearest and dearest objection to consequentialism is independent of this issue.

The other possible account of how to recover what a person ought to do from consequentialism’s value function that we need to consider holds that a person’s beliefs, rational or not, do not come into the picture. What is crucial is simply which action in fact has, or would have, the best consequences. 7 Many consequentialists write as if this was their view. In a well-known passage, Sidgwick says ‘that Universal Happiness is the ultimate standard must not be understood to imply that Universal Benevolence is the only right…motive of action…it is not necessary that the end which gives the criterion of rightness should always be the end at which we consciously aim’. 8 Here it seems clear that he is assuming that what makes an act right—the criterion of rightness, as he puts it—is the extent to which it

-224-

in fact achieves a certain end. 9 Similarly, Peter Railton distinguishes subjective consequentialism, the doctrine that ‘whenever one faces a choice of actions, one should attempt to determine which act of those available would most promote the good, and should then try to act accordingly’, from objective consequentialism, ‘the view that the criterion of the rightness of an act or course of action is whether it in fact would most promote the good’, and goes on to argue in support of objective consequentialism. 10

There are two problems with this proposal. First, it gives the intuitively wrong answer in the drugs case. In the drugs case, either it is prescribing drug B or it is prescribing drug C which is the course of action which would in fact have the best consequences—and Jill knows this, although she does not know which of the two it is—but neither prescribing drug B nor prescribing drug C is the right course of action for Jill. As we observed earlier, it is prescribing drug A which is the intuitively correct course of action for Jill despite the fact that she knows that it will not have the best consequences. We would be horrified if she prescribed drug B, and horrified if she prescribed drug C.

The second problem arises from the fact that we are dealing with an ethical theory when we deal with consequentialism, a theory about action, about what to do. In consequence, we have to see consequentialism as containing as a constituitive part prescriptions for action. Now, the fact that an action has in fact the best consequences may be a matter which is obscure to an agent. (Similarly, it may be



obscure to the agent what the objective chances are.) In the drugs example, Jill has some idea but not enough of an idea about which course of action would have the best results. In other examples, the agents have very little idea which course of action would have the best results. This was the case until recently in the treatment of AIDS. Hence, the fact that a course of action would have the best results is not in itself a guide to action, for a guide to action must in some appropriate sense be present to the agent’s mind. We need, if you like, a story from the inside of an agent to be part of any theory which is properly a theory in ethics, and having the best consequences is a story from the outside. It is fine for a theory in physics to tell us about its central notions in a way which leaves it obscure how to move from those notions to action, for that passage can be left to something which is not physics; but the passage to action is the very business of ethics.

Railton is well aware of the need to give an account of the

-225-

passage to action, for he remarks that a ‘further objection [to objective consequentialism] is that the lack of any direct link between objective consequentialism and a particular mode of decision making leaves the view too vague to provide adequate guidance in practice’. His reply is that: ‘On the contrary, objective consequentialism sets a definite and distinctive criterion of right action, and it becomes an empirical question…which modes of decision making should be employed and when’. 11 In short, Railton’s proposal is, I take it, that the moral decision problem should be approached by setting oneself the goal of doing what is objectively right—the action that has in fact the best consequences—and then performing the action which the empirical evidence suggests is most likely to have this property. 12 However, this approach to the decision problem gives the wrong answers. I will illustrate the point with a modification of the drug example.

The drug example, mark II

As before, Jill is the doctor and John is the patient with the skin problem. But this time Jill has only two drugs, drug X and drug Y, at her disposal which have any chance of effecting a cure. Drug X has a 90 per cent chance of curing the patient but also has a 10 per cent chance of killing him; drug Y has a 50 per cent chance of curing the patient, but has no bad side effects. Jill’s choice is between prescribing X or prescribing Y. It is clear that she should prescribe Y, and yet that course of action is not the course of action most likely to have the best results, for it is not the course of action most likely to be objectively right. It has only a 50 per cent chance of being



objectively right, whereas prescribing drug X has a 90 per cent chance of being objectively right.

This example is one among many. Consider, for instance, the question of whether or not to place a bet on a horse race. Clearly, it is often the right thing to do not to place a bet. There may be no horse about which the bookies are offering good enough odds. And yet in declining to place a bet you know that you are pursuing the one course of action guaranteed not to have the best outcome. Your problem, of course, is that although you know that there is a course of action with a better outcome, you do not know which one it is.

In general, it seems to me potentially misleading to speak of consequentialism as giving the moral agent the aim of doing what

-226-

has the best consequences. If it means that the agent ought to do what has the highest expected value or moral utility, where value is determined consequentially, then of course there is no problem. But it is easy to slide into thinking that consequentialism holds that people should aim at the best consequences in the sense of trying to select the option with the best consequences, whereas in fact most of the time we should select an option which we know for sure does not have the best consequences. Most of the time we are in the position of the person who declines to bet. The right option to select is a ‘play safe’ one chosen in the knowledge that it does not have the best consequences and in ignorance of which option does have the best consequences.

I argued that consequentialism must, as I put it, tell a story from the inside about how to recover what an agent ought to do from consequentialism’s value function, a story in terms of what is in the agent’s mind at the time of action. I thus am agreeing with Thomas Nagel’s claim that ‘morality requires of us not only certain forms of conduct but also the motives required to produce that conduct’. 13 And the proposal I borrowed from decision theory meets this constraint, because it is in terms of the agent’s probability function, that is, in terms of the agent’s belief state at the time of action. Indeed, the proposal I borrowed from decision theory can be viewed as an account of what an agent ought to do that yields an account of what an agent’s motives ought to be, of how an agent’s mind ought to be as far as the springs of action go. For, as we remarked earlier, we can view consequentialism’s value function as an account of how an agent’s preference function, the agent’s desires, ought to be—the preference function ought to assign the same values to the various



states of affairs as does the value function. Hence, when the proposal recovers what an agent ought to do from the agent’s probability function combined with consequentialism’s value function, this can be described as recovering what an agent ought to do from what the agent believes combined with what the agent ought to desire, and thereby as yielding a theory of right motivation. Although consequentialism of the decision-theoretic kind described here (henceforth, consequentialism) has built into its very account of right action, a doctrine about right motivation, it is not committed to any particular view about the mental processes an agent ought to go through in deciding what to do; indeed, it is compatible with consequentialism that at least sometimes one ought not to go through anything that might naturally

-227-

be described as a process of thought at all. It is, for instance, compatible with consequentialism that an agent ought to go through a distinctively deontological cast of thought before acting. The agent may, for instance, have a terrible track record in calculating likely consequences and have found out by bitter experience that he does better following simple rules; or perhaps he knows that the world is under the control of a demon who rewards people for thinking in the Kantian style. 14 But then the agent’s probability function will give the probability that an act has good consequences given that it was reached by rule following (or given that it is the Kantian act) a high value, and feeding this into the expected value equation will give the rule-satisfying act (or the Kantian act) the highest expected moral utility.

Similarly, we can acknowledge the often made point that sometimes consequentialist considerations support not going through consequentialist deliberations. 15 It may be that acting spontaneously in the situation to hand is known to have the best results—ducking, swerving, smiling, playing a drop shot and the like—are commonly best done straight off as the spirit moves one and without further ado. But, in such cases, the conditional probability of good results given that one acts without further ado will be high, or at any rate higher than the probability of good results given one acts after deliberation, and so, consequentialism will give the right result that one should in such cases act spontaneously. In such cases, the consequentialist should hold that one ought to be consequentially motivated, although one should not consciously reason consequentially.

It might be objected that this is an impossible position for the consequentialist to take up, given our account of consequentialism. How can the probability function give a high value to good consequences conditional on spontaneous action? For probability



here means the agent’s subjective probability function, that is, the agent’s beliefs in quantitative guise; and the whole point of these examples of spontaneous action is that the agent acts ‘without thinking’. Accordingly, it might be suggested, he or she will not have any beliefs of the needed kind; the agent will not, for instance, believe that ducking the blow will have good results. There will not be enough time for that thought, only time to duck instinctively. Consider, however, the familiar example in the philosophy of perception where you drive past an advertising hoarding without consciously registering what is on the hoarding. You are later asked what the hoarding was advertising and to your

-228-

surprise you are able to answer. This shows that you have seen the hoarding and what was on it, although you were not conscious of the fact. In the same way, one who ducks believes that not ducking will have results that are unpleasant despite the fact that the thought to that effect is not consciously entertained. Spontaneous action is not action without belief; it is action without the conscious reviewing of belief.

What is central here is the distinction between the immediate springs of action and the processes that lead up to the obtaining of those springs. Otherwise consequentialism can seem to face a dilemma. Suppose that consequentialism says nothing about the mind of the agent at all. It says merely that right action is action with property Φ, for some consequentialist treatment of Φ which pertains solely to what in fact would happen and not at all to what the agent thinks. In that case, consequentialism, as Williams puts it, ‘has to vanish from making any distinctive mark in the world’, by which, I take it, he is, at least in part, making the point we made earlier that consequentialism must say something about right decision. 16 On the other hand, suppose that consequentialism is expressed as a doctrine about how to go about making the morally right decision, as a variety of subjective consequentialism in Railton’s terms, and suppose in particular that it says to think along Ψ lines. What, then, if thinking along Ψ lines is discovered to have bad consequences in certain situations? 17 Our decision-theoretic account of consequentialism disarms the second horn of the dilemma by answering that, in such situations, the agent ought not to think along Ψ lines, for the agent’s beliefs will then include that thinking along Ψ lines in such situations has a low expected moral utility.

It is important to note that on our account consequentialism is not committed to the view that maximising expected moral utility is the right motive for action. A number of writers have made the point that doing something because you consider you ought to



do it rather than doing it because you want to is, generally speaking, not the mark of the kind of person it is comfortable to be around. Being nice to someone solely because it is your duty to be nice to them is the kind of niceness we can all do without. 18 Michael Stocker sees this as a problem for what he calls ‘the standard view’. As he puts it:

The standard view has it that a morally good intention is an essential constituent of a morally good act. This seems

-229-

correct enough. On that view, further, a morally good intention is an intention to do an act for the sake of its goodness or rightness. 19

Perhaps this is true of some ethical views properly referred to as ‘standard’, but it is not true of consequentialism as characterised here. The right motive for action on the consequentialist view is the agent’s beliefs combined with desires that conform to the consequentialist value function, and the consequentialist value function assigns no value as such to maximising expected utility. Take the drugs, mark I, case. Prescribing drug A is the right thing to do because Σi Pr(Oi /Aj ).V(Oi ) takes a maximum value when Aj is prescribing drug A. But this fact that it takes a maximum value does not then confer additional value on prescribing drug A. That would be double-counting. What ought to move a person to action, according to consequentialism, are desires which may be represented as ranking states of affairs in the consequentialist way, but maximising expected utility is not a factor in this ranking.

Before we turn to how our account of consequentialism helps with the nearest and dearest objection, I need to note an annoying complication. I have been arguing for an interpretation of consequentialism which makes what an agent ought to do the act which has the greatest expected moral utility, and so is a function of the consequentialist value function and the agent’s probability function at the time. But an agent’s probability function at the time of action may differ from her function at other times, and from the probability function of other persons at the same or other times. What happens if we substitute one of these other functions in place of the agent’s probability function at the time of action? The answer is that we get an annoying profusion of ‘ought’s. Consider the drug case, mark I. I said that the intuitively correct answer to what Jill ought to do is prescribe drug A, and so it is. But suppose Jill later conducts a piece of definitive research which establishes that with patients of John’s blood type there is absolutely no chance that drug B will cause death, and in fact with



such patients drug B is certain to effect a complete cure without any bad side-effects. What will she then say about her past treatment of John: was it the right treatment or the wrong treatment? The natural thing to say would be something like ‘By the light of what I now know, I ought to have prescribed drug B, but it would have been quite wrong to do so at the time’. But if it

-230-

would have been the wrong thing to do at the time, how can it be what she ought to have done?

I think that we have no alternative but to recognise a whole range of ‘ought’s—what she ought to do by the light of her beliefs at the time of action, what she ought to do by the lights of what she later establishes (a retrospective ‘ought’, as it is sometimes put), what she ought to do by the lights of one or another onlooker who has different information on the subject, and, what is more, what she ought to do by God’s lights, that is, by the lights of one who knows what will and would happen for each and every course of action. 20 The last will be a species of objective ‘ought’ of the kind that features in Railton’s (and Brink’s) account of objective consequentialism. I hereby stipulate that what I mean from here on by ‘ought’, and what I meant, and hope and expect you implicitly took me to mean when we were discussing the examples, was the ‘ought’ most immediately relevant to action, the ‘ought’ which I urged it to be the primary business of an ethical theory to deliver. When we act we must perforce use what is available to us at the time, not what may be available to us in the future or what is available to someone else, and, least of all, not what is available to a God-like being who knows everything about what would, will and did happen.

It might be tempting to conclude that my acknowledgment that there are a variety of ‘ought’s means that I am not really disagreeing with one who urges that what a person ought to do according to consequentialism is what in fact has the best consequences; we are rather talking past each other. However, the substantive issue remains of the need for a moral theory to elucidate the ‘ought’ most immediately relevant to action, and of how this should be done, quite independently of whether or not the target notion is unambiguously captured by ‘ought’ in English.

14.2 A reply to the nearest and dearest objection

The decision-theoretic way of understanding consequentialism gives a major role to



the agent’s subjective probability function. This fact is the key to our reply to the nearest and dearest objection. I think that the reply can be most easily grasped by leading up to it via two examples: the drug example, mark III, and the crowd control example.

-231-

The drug example, mark III

In mark I, Jill had three drugs, A, B, and C, and one patient. This time Jill has three patients, A, B, and C, and one drug, and only enough of that drug to administer to a single patient. Her choice in mark I was among drugs; her choice this time is among patients, but it is, all the same, a similar style of choice situation that faces her. For we are given that she knows that patient A will derive considerable benefit from the drug without being completely cured, and also that one or other of patients B and C would be completely cured by the drug. However, she also knows that one or other of patients B and C would be killed by the drug. She has no way of telling which of B and C would be the one completely cured and which would be the one killed. What ought Jill to do?

The answer obviously is to administer the drug to patient A, and this is of course the answer our decision-theoretic approach delivers. 21 The expected moral value of administering the drug to A is higher than that of administering to B, and higher than that of administering to C, because the possibility of a rather better result in those two cases goes along with a significant chance of a very much worse result. Of course, Jill knows that there is a better course of action open to her in the sense of a course of action which would have better consequences than administering the drug to A, but her problem is that she does not know whether it is administering the drug to B or administering it to C which is that better course.

What do we learn from this example to help us with the nearest and dearest objection? Well, it would clearly be a mistake to accuse Jill of an illegitimate bias towards patient A when she gave the drug to him rather than the others. Jill is biased towards patient A in the sense that her actions are directed towards securing his good, but the explanation for this fact is not that her preference function gives a greater weight to a benefit for A rather than one for B or for C. The explanation lies in her probability function. Consequentialism demands of us an impartial preference function, for its value function gives equal weight to the happiness, or preference satisfaction, or pleasure, or share of the ideal good, or…of each individual, but what the example tells us is that the fact, in and of itself, that our behaviour is directed



towards securing the happiness, or preference satisfaction or…of a small group—our family, friends, and so on—does not

-232-

in itself show that we have an illegitimately biased preference function by the standards of consequentialism. The explanation of the directed nature of our behaviour may lie in our probability functions.

The question, then, for consequentialists is the following. Can the special regard we have for a relatively very small group of people, to the extent that it is morally justified, be explained probabilistically, in terms of our special epistemological status with regard to our nearest and dearest, rather than in terms of an agent-relative preference function? The drug example mark III does not show that the answer to this question is yes. What it shows is that this is the key question that we need to ask.

I do not have a decisive argument that the answer to this key question is yes. What I do have are two considerations that suggest that it may well be yes. The first I will introduce with the crowd control example.

The crowd control example

Imagine that you are a police inspector who has been assigned the task of controlling a large crowd at a forthcoming soccer match. You have to choose between two plans: the scatter plan, and the sector plan. The scatter plan is put to you in the following terms. ‘Each person in the crowd is of equal value. Any plan which told a member of the police squad to focus his or her attention on any particular person or group of persons would be immoral. Therefore, each member of the squad must roam through the crowd doing good wherever he or she can among as widely distributed a group of spectators as possible.’ The sector plan is put to you in the following terms. ‘Each member of the squad should be assigned their own sector of the crowd to be their special responsibility. This way members of the squad will not get in each other’s way, and will build up a knowledge of what is happening in their sector and of potential trouble makers in it, which will help them decide on the best course of action should there be trouble. Also, we will avoid a major problem for the scatter plan, namely the possibility that at some particular time there will be part of the crowd which no one is covering. Of course, the sector plan should be administered flexibly. Although, as a general rule, each squad member should confine his or her attention to their assigned sector, if things are going particularly badly in another



-233-

sector and it is clear that an extra helping hand would make a big difference, then a transfer of attention may well be justified.’ The plan that we follow in day to day life is of course the sector plan. We focus on a particular group, our family, friends and immediate circle, while allowing that we may properly neglect them if the opportunity arises to make a very big difference for the better elsewhere. As the point is sometimes put, though it would be quite wrong to neglect family and friends in order to achieve a small increase in welfare elsewhere, it would be quite proper to neglect them in order to achieve peace in the Middle East. And we can approach the question of whether probabilistic considerations can provide a justification from the consequentialist’s point of view for our focusing on family and friends by asking when the sector plan would be the right plan for a consequentialistically minded inspector to adopt. The sector plan would be strongly indicated in the following circumstances.

(a) When getting to know certain individuals is important for achieving good results. The scatter plan distributes any given squad member’s attention very widely, making any detailed knowledge of the psychology of particular individuals difficult. If good results depend on such knowledge, then squad members should restrict themselves to a smaller group, as in the sector plan. (b) When achieving good results involves co-ordinating a series of actions. Sometimes an isolated action has little effect in itself. What is needed is an extended plan of action, with later actions chosen on the basis of positive and negative feedback from the results of earlier actions. Think of the contrast between a ‘one shot’ drug treatment and an extended course of treatment with later drugs and dosages being chosen in the light of the effects of earlier treatments. (c) When achieving good results depends on setting up mutual trust and respect and understanding between individuals. The traditional ‘bobby on the beat’ is a special kind of sector arrangement, and rests on exactly this kind of point. (d) When there is a significant chance of different squad member’s actions nullifying each other if directed towards the same people. When we are in a situation where ‘too many cooks spoil the broth’, the sector plan is clearly superior to the scatter plan.

-234-



(e) When there is an obvious way to assign police to separate sectors which coincides with their natural inclinations and enthusiasms, particularly when this fact is common knowledge. This reduces, and is known to reduce, the setting up costs of the sector plan by avoiding costly debate over who takes responsibility for which sector. It also increases the penalty consequent upon a squad member not policing their ‘natural’ sector, by increasing the chance that that sector will remain unattended through other police wrongly assuming that it is attended.

Clearly, there is a great deal more to be said here, much of it to do with straight empirical facts. 22 But I hope that I have said enough to make it plausible that the sector plan is indicated in the kind of circumstances that apply in our day to day interactions with the world around us and the people in it. It is hard to know what actions will have good effects, and our opinions on the matter are much better founded in the case of people we know well, precisely because we know them well. Achieving good results is very often a matter of co-ordinating a series of actions rather than scattering largesse around. Mutual trust and affection is important for good results. Too many cooks can spoil the broth when it comes to interacting in a beneficial way with one’s fellow human beings. There is very obviously a group of people whose welfare we are naturally inclined to concern ourselves with, namely, those nearest and dearest to us. And, most importantly for our decision-theoretic approach to consequentialism, facts such those just adumbrated are, I take it, pretty much common knowledge.

My suggestion, then, is that the consequentialist can reply to the nearest and dearest objection by arguing that the kind of direction of attention towards those we hold dear which is so characteristic of a worthwhile life can be explained without attributing a biased value function. It is instead a reflection of the nature of our probability functions, in particular of the kinds of facts about the epistemology of achieving good consequences that we have been rehearsing. The suggestion is not, of course, that the kind of direction of attention we typically manifest in fact towards those we hold dear can be explained without attributing a biased value function. It is no objection to consequentialism that, according to it, we ought to do more than we in fact do for people we hardly know. We ought to do more for people we hardly

-235-



know. We are too tribal. The suggestion is that a considerable degree of focus on our family and friends, enough to meet the demand that our lives have a meaningful focus, is plausibly consistent with living morally defensible lives according to consequentialism.

14.3 On three objections

(i) Williams has argued that

it [consequentialism] essentially involves the notion of negative responsibility: that if I am ever responsible for anything, then I must be just as much responsible for things that I allow or fail to prevent, as I am for things that I myself…bring about. Those things must also must enter my deliberations, as a responsible moral agent, on the same footing…what matters [according to consequentialism] with respect to a given action is what comes about if it is done, and what conies about if it is not done, and those are questions not intrinsically affected by the nature of the causal linkage, in particular by whether the outcome is partly produced by other agents. 23

If Williams is right, we are in trouble. The key idea behind our reply to the nearest and dearest objection was that the reflections we grouped under the heading of the sector plan made it plausible that a consequentialist ought to take special responsibility for what is within his or her ken, and that will obviously involve making who does something a very important matter in many cases—and that runs directly counter to Williams’s claim that who does something is irrelevant for consequentialists. However, it is crucial here to bear in mind the distinction between value and expected value. Williams is right that consequentialism’s value function gives no weight per se to who does something (and that no doubt was what he had in mind), but nevertheless, who does something can be enormously important to the expected value of a course of action, and it is that which is crucial according to our account of consequentialism, and that is how who does something can ‘enter my deliberations’. In particular, Smith may hold that it would, all in all, be better were A to obtain than were B to obtain, but it does not follow that Smith qua consequentialist should seek

-236-



to bring about A rather than B. For Smith may hold, in addition, that someone else, Jones, knows more about the matter than he himself does and that Jones has good values. In such a case, the decision facing Smith is between he himself bringing about A rather than B, or instead leaving the decision as to whether or not to bring about A rather than B to Jones, and it is easy to show that the latter may have the greater expected moral utility for Smith. The crucial point is that, though for Smith the probability of good consequences given he does B is low, the probability of good consequences given Jones does B may be high, because of Smith’s opinion that Jones is best placed to make the decision. In general, in cases where we judge it best to leave a decision between A and B to the experts, as we say, although we may have ourselves a view as to which of doing A and doing B has the greatest expected value, leaving the matter to the experts may have the greatest expected value of all. Thus, who does something can be crucial according to consequentialism.

It might be replied on Williams’s behalf that he did not have expected value in mind, and that the point we have just made only holds for expected value. However, this would make nonsense of Williams’s (correct) insistence that consequentialism be an ethical decision theory, and of the talk in the above quotation of how things ‘must enter my deliberations’ (my emphasis).

(ii) Railton raises (as a preliminary to a reply) the nearest and dearest difficulty for consequentialism with the following example.

Juan and Linda…have a commuting marriage. They normally get together only every other week, but one week she seems a bit depressed and harried, and so he decides to take an extra trip in order to be with her. If he did not travel, he would save a fairly large sum that he could send Oxfam to dig a well in a drought-stricken village. Even reckoning in Linda’s uninterrupted malaise, Juan’s guilt, and any ill effects on their relationship, it may be that for Juan to contribute the fare to Oxfam would produce better consequences overall than the unscheduled trip. 24

It might be objected that what I have said so far in no way meets the objection posed by this example. But from the decision-theoretic

-237-



point of view, what is crucial is not that ‘it may be that for Juan to contribute the fare to Oxfam would produce better consequences’, it is how likely it is to do so. And of course the effects of isolated acts of charity on the Third World are a matter of considerable debate, whereas Juan can be pretty certain of at least some of the effects of his making the unscheduled trip. It is important here to remember that the relevant consequence of sending, say, $500, should not be approached by asking what $500 will buy in the Third World, but by addressing the likely difference between what would be achieved by the sum Oxfam would have without Juan’s $500, and what would be achieved by the sum with Juan’s $500. 25

What plausibly is obvious is that many of us in advanced Western societies could achieve a great deal more good if we devoted our energies to a systematic, informed programme of transferring any excess wealth towards the Third World. I do not mean isolated donations of airfares, but nor do I mean just sending a lot more money until it really hurts. I mean becoming actively involved in, and knowledgeable about, what is going on in the Third World; learning how aid agencies work, which ones do good, which, knowingly or unknowingly, do harm; finding out exactly how villages use money sent to them; the effects outside money and services typically have on the local social and economic structures; and so on and so forth. But how can this observation possibly constitute a nearest and dearest objection to consequentialism? A person who behaved in the way that I have just described would be directing her attention to those nearest and dearest to her. For she would be paying special attention to a relatively small section of the world’s population, and she would be giving a special place to her own projects, one of which would precisely be helping certain people in the Third World. For her, the people whose welfare she was particularly concerned with would be those people she had studied, and so got to know and understand, living in various villages in the Third World, as opposed to those living in the same house or the same neighbourhood as she does.

(iii) It might well be objected that we can distinguish two nearest and dearest objections, and that I have replied to only one of them. One objection is, ‘How can consequentialists make sense of the fact that there is a relatively small group of people whose welfare plays a special role in our lives, given the agent-neutral

-238-



nature of consequentialism’s value function?’ Our reply was that consequentialism should be viewed decision-theoretically. The way right value translates into right action is through an agent’s beliefs, and that when this is appreciated, empirical facts about our cognitive powers and situation make it plausible that our actions should be highly focused much of the time. The other objection is, ‘How can consequentialists make sense of its being the particular small group of people that it mostly is?’ Perhaps consequentialism can make sense of there being a small group, but why the small group of family, friends, fellow citizens, and the like that it so often is?

One possible reply is that consequentialism cannot make sense of this, but this is no objection to it. We are outrageously tribal in our everyday morality, and so much the better for consequentialism that it makes this clear. I cannot believe this. I grant that we are unduly tribal, but not that we are outrageously so. I think that we can give a consequentialist explanation of why, for most of us, the special group is our family and friends, in terms of empirical facts about human character and psychology.

One way you might draw on empirical facts about human nature is to argue that some particular action giving preference to family and friends in a way which goes against consequentialist principles is wrong but excusable in some sense because it is the exercising of a character which is good in consequentialist terms. This is William Godwin’s claim about his famous example of your having to choose between rescuing Fenelon, a famous author and archbishop, and a valet who happens to be your own father, from a burning house: rescuing your father is the wrong action but, at the same time, the action which springs from the right character. 26 The idea is that although there is perhaps in theory a better character which would lead to the best action as judged consequentially, in practice such a character is not available to us, or at least not to most of us.

I think that there is an element of ‘ducking the question’ about this reply. If the action which favours family and friends is right and consequentialism says that it is wrong, then consequentialism is false, and there’s an end on it. If, on the other hand, the claim is that it is the action which favours family and friends which is wrong, then it is that which needs to be established, not facts about good character. Or perhaps what we are being offered is a variant on consequentialism, according to which an action is to be judged not directly but via the status, judged consequentially,

-239-



of the character which gives rise to it; but then we appear to be landed with a dubious compromise reminiscent of rule utilitarianism. 27 If consequences are the key in one place, why not across the board?

I am not here denying the correct and important point that some particular action may be wrong in consequentialist terms, and yet spring from a character which is right in consequentialist terms. 28 I am denying that the point helps with the essentials of the nearest and dearest objection. For the consequences of having a character which gives a special place in one’s affections and concerns to those persons who are closest to one are, in the main, consequences of the manifestations of such a character, that is, of the actions which are especially directed to the needs of those closest to one. Hence, a consequentialist justification of such a character presupposes a consequentialist justification of those actions—which returns us to the very question raised by the nearest and dearest objection. Of course, it is not the case for every character trait that the consequences of possessing it are, in the main, the consequences of manifesting it. A major consequence of possessing a certain character trait may be that people know that you possess it, which knowledge may in turn have a major effect on their behaviour without it ever being necessary to manifest that disposition. Being disposed to react with pointless violence on being attacked is an example; a nation’s disposition to make a nuclear response to a major nuclear attack is another possible example familiar in the literature on nuclear deterrence. Our point, though, is that the character trait of being especially concerned with the welfare of those closest to us is, like most dispositions, known by and large through its manifestations, and has its effects principally via those manifestations. I know that you are especially concerned about the welfare of your family because your actions display that concern. Hence, a consequentialist justification of that character trait awaits a consequentialist justification of those actions. 29

Be all this as it may, I think that points about character and human nature can be put to more direct work here. One’s character can be a major factor in settling what consequences are likely, and so can be a major factor in settling what acts are right from the (act) consequentialist’s point of view.

Some actions are such that they only have good results if they are followed up in the right way. Taking the first capsule in a course of antibiotics will only have good results if the remainder

-240-



are taken at the right times; agreeing to write a book review will only have good results if you write the review in good time; going on a beach holiday will only have good results if you avoid getting badly sunburnt, and so on and so forth. In all these cases, it is better not to start if you are not going to follow up in an appropriate way. From the consequentialist perspective, whether action A ought to be done depends in part on what the agent would in fact do subsequent to doing A. 30

This means that in deciding what to do here and now an agent must take account of what he or she will do in the future, and that involves taking very seriously questions of character. Do I have the persistence that will be called for, will I remain sufficiently enthusiastic about the project to put in the time required, will I be able to retain a sufficiently impartial outlook, will I be able to avoid the various temptations that will arise, and so on and so forth? For some of us in some situations, these kinds of considerations count against attempting to secure benefits for our friends and family. We do better sometimes with people we are not so close to. Some men should most definitely not play doubles in tennis with their wives as partners. But, as a rule, we do better for reasons of character (that no doubt have an evolutionary explanation) with projects that involve family and friends rather than strangers. This is simply because we are much less likely to lose the enthusiasm required to see the project through to a successful conclusion when the project benefits people we have a particular affection for. Perhaps a mundane example will make the point clearer. Jones may be able, in principle, to do an equally good job of organising the seminars in History or the seminars in Philosophy for the forthcoming year. She has the required knowledge and contacts in both areas. She is, however, much more excited by Philosophy than by History. In such a case, even prescinding from her own enjoyment, it may well be that she ought to take on the task of organising the Philosophy seminars. For, although she knows that she could do equally well at either, she knows that she would most likely do better if she takes on the Philosophy programme, and her knowledge of this fact should influence her to agree to take on the Philosophy programme rather than the History programme.

We have seen that the good consequentialist should focus her attentions on securing the well-being of a relatively small number of people, herself included, not because she rates their welfare more highly than the welfare of others, but because she is in a

-241-



better position to secure their welfare. Typically, this will involve her in settling on a relatively extended programme of action which will take some resolution and strength of character to carry forward successfully. If she is at all like most of us, before she starts she knows that the chances of success are much greater if she makes the relatively small group in question those who are her family and friends, rather than people she hardly knows. There are exceptions to this generalisation about human psychology, perhaps Mother Teresa was one, perhaps Ralph Nader is another; from reports it seems that they have had the ability to carry through a demanding programme of action which benefits a group of people which, though tiny by comparison with the population of the world, is large by comparison with the circle of family, friends and associates that provide the principal focus of action for most of us. They do not seem to be dependent on the kind of close personal relationships that are essential to keep most of us from being outrageously selfish.

14.4 Conclusion

Consequentialism tackles the question of what an agent ought to do in terms of the values of outcomes, and assigns those values in an agent independent way, and yet the lives we consider worth living give a quite central place to certain of our fellow human beings. We have an agent-relative moral outlook. My argument has been that the consequentialist can plausibly explain agent relativity in terms of the role probability plays in the recovery of what an agent ought to do from the consequentialist’s value function. The injunction to maximise expected moral utility, when combined with the facts we listed under the heading of the sector plan, means that the consequentialist can accommodate our conviction that a morally good life gives a special place to responsibilities towards a smallish group. Which smallish group is another question, and here I argued that for most of us the group should be chosen tribally. Because of empirical facts about our natures, that choice decreases the chance that we will backslide.

One objection to consequentialism is that it conflicts with firmly held moral convictions, in particular concerning our obligations towards our nearest and dearest. It may be urged that my reply to this objection is seriously incomplete. For we can reasonably easily describe a possible case where the factors I mentioned as providing a justification in consequentialist terms for favouring

-242-



one’s nearest and dearest do not apply, and yet, according to commonsense morality, one should favour, or at the least it is permissible to favour, one’s nearest and dearest. 31 My concern, though, has been to reply to the objection that consequentialism would given the way things more or less are, render the morally good life not worth living. I take this to be the really disturbing aspect of the nearest and dearest objection to consequentialism. Consequentialists can perhaps live with the conflict with commonsense morality, drawing, for instance, on the notorious difficulties attending giving a rationale for its central features. 32 But it seems to me that they cannot live with the conflict with a life worth living, given the way things more or less are. That would be to invite the challenge that their conception of what ought to be done had lost touch with human morality. 33

Notes 1 In, e.g., ‘A Critique of Utilitarianism’ in J.J.C. Smart and Bernard Williams,

Utilitarianism: For and Against, Cambridge, Cambridge University Press, 1973. 2 J.J.C. Smart, ‘An Outline of a System of Utilitarian Ethics’ in Utilitarianism: For

and Against. It is arguable that something like the account can also be found in some classical presentations, for instance, Jeremy Bentham, An Introduction to the Principles of Morals and Legislation, London, Athlone, 1970, and Henry Sidgwick, The Methods of Ethics, Chicago, University of Chicago Press, 7th edition, 1907, though other interpretations are very possible, and, as we will see, a much quoted passage from Sidgwick points in a quite different direction.

3 Decision theory comes in a number of varieties. For example, in some ‘Pr(Oi /Aj )’ is replaced by . The points I wish to

make here are independent of the particular variety. (Though I in fact favour the latter, which is indeed the most obvious way of capturing a consequentialist approach to matters provided the is read appropriately.)

For a recent discussion of the varieties see Ellery Eells, Rational Decision and Causality, Cambridge, Cambridge University Press, 1982. Incidentally, the preference function in decision theory is often referred to as a value function, but I will reserve the latter term for what an agent ought to prefer in the moral sense.

4 J.J.C. Smart, ‘An Outline of a System of Utilitarian Ethics’, appears to favour a proposal of this kind. See also Philip Pettit and Geoffrey Brennan, ‘Restrictive Consequentialism’, Australasian Journal of Philosophy, 64 (1986):438-55.

5 For a clear presentation of the proof, see Paul Horwich, Probability and Evidence, Cambridge, Cambridge University Press, 1982, pp. 125ff.

6 For a more detailed development of this argument, see Frank Jackson,



-243-

‘A Probabilistic Approach to Moral Responsibility’ in R. Barcan Marcus, et al., eds, Proceedings of the 7th International Congress of Logic, Methodology, and Philosophy of Science, Amsterdam, Elsevier Science Publishers, 1986, pp. 351-66.

7 Another approach in which the agent’s beliefs do not come into the picture is one in which it is objective, one-place chances, rather than probabilities construed epistemically, of the various possible outcomes which matter, but I take the essentials of the critical discussion that follows to apply equally against this approach.

8 Henry Sidgwick, The Methods of Ethics, p. 413. 9 And this is certainly how this passage is typically read, see, e.g., David O. Brink,

Moral Realism and the Foundations of Ethics, Cambridge, Cambridge University Press, 1989, p. 257. And Brink is explicit in endorsing the idea that what makes an act right according to consequentialism should be recovered from what in fact does or would happen. See also Fred Feldman, Doing the Best We Can, Dordrecht, Reidel, 1986.

10 Peter Railton, ‘Alienation, Consequentialism, and the Demands of Morality’, Philosophy and Public Affairs, 13 (1984), reprinted in S. Scheffler, ed., Consequentialism and its Critics, New York, Oxford University Press, 1988, p. 113 of the reprinted paper (my emphasis). Railton, p. 113, n. 24, mentions the decision-theoretic approach in passing, and it is unclear going on the printed word how much substantive disagreement there is between us on the question of what, according to consequentialism, a person ought to do. However, very helpful discussion with Railton has made it clear to me that we are in substantive disagreement. What is, in any case, clear from the printed word alone is that we are in substantive disagreement over how to answer the nearest and dearest objection, for in the footnote he remarks that his arguments go through independently of whether or not the decision-theoretic approach is adopted; whereas, as will be become very clear, our treatment of the nearest and dearest objection rests quite crucially on the adoption of the decision-theoretic approach. In Bart Gruzalski, ‘The Defeat of Utilitarian Generalization’, Ethics, 93 (1982):22-38, a decision-theoretic approach is given a crucial role in assessing the relative merits of act and rule or generalised versions of utilitarianism.

11 Both passages are from ‘Alienation, Consequentialism, and the Demands of Morality’, p. 117, my emphasis.



12 See also his earlier remarks on p. 116 about objective consequentialism ‘not blurring the distinction between the truth-conditions of an ethical theory and its acceptance-conditions. Feldman, Doing the Best We Can, explicitly takes this approach to the moral decision problem.

13 Thomas Nagel, The View from Nowhere, Oxford, Oxford University Press, 1986, p. 191. See also Bernard Williams, ‘A Critique of Utilitarianism’, p. 128, though the focus there is on utilitarianism rather than ethics in general.

14 An example of Railton’s, ‘Alienation, Consequentialism, and the Demands of Morality’, p. 116.

-244-

15 See, for three examples among many, Railton, ‘Alienation, Consequentialism, and the Demands of Morality’; Smart, ‘An Outline of a System of Utilitarian Ethics’, p. 43; and the helpfully detailed account of a number of different examples in Pettit and Brennan, ‘Restrictive Consequentialism’.

16 ‘A Critique of Utilitarianism’, p. 135. Also, his otherwise rather dark remarks on pp. 134-5 about utilitarianism retiring or being left with nothing more than ‘total assessment from the transcendental standpoint’ may be a way of expressing the idea that utilitarianism must at some level be a species of decision theory. I am indebted here to a discussion with Thomas Scanlon.

17 A similar point can arise for deontological theories of course. ‘Keep your promises’ is not in itself a rule of decision, though ‘Keep what you take to be your promises’ is. But what if you know that you are very bad at remembering what it is that you promised to do?

18 For some convincingly detailed examples showing this, see Michael Stocker, ‘The Schizophrenia of Modern Ethical Theories’, Journal of Philosophy, 73 (1976):455-66, and Railton, ‘Alienation, Consequentialism, and the Demands of Morality’.

19 Stocker, ‘The Schizophrenia of Modern Ethical Theories’, p. 462. Incidentally, on our view, Consequentialism does not imply that a morally good intention is essential to a morally good act, at least if morally good act here means what an agent ought to do. It is possible to do the right thing for the wrong reason. For an act which maximises expected moral utility might also maximise expected utility according to some immoral value function that just happens to agree, near enough, with the moral value function for the states of affairs that matter in the case at hand, and the agent’s own value function might be the immoral one. (This is rephrased from the original publication.) What is true is that doing an act for the right reason is sufficient but not necessary for it being what ought to be done in the sense we are insisting is central in ethics.



20 There are also the various non-moral ‘ought’s—prudential etc., but that is another, and here irrelevant, dimension of variation.

21 Assuming, of course, that getting more information and then acting is not a viable option.

22 See, e.g., the discussion of the allocation of responsibilities in Philip Pettit and Robert Goodin, ‘The Possibility of Special Duties’, Canadian Journal of Philosophy, 16 (1986):651-76.

23 Williams, ‘Consequentialism and Integrity’ in Sheffler, Consequentialism and Its Critics, p. 31. Williams is of course supposing that effects due to the identity of the agent have been incorporated into the consequences.

24 ‘Alienation. Consequentialism, and the Demands of Morality’, p. 120. 25 You may not like this way of approaching the consequences to be assigned to

giving $500 to Oxfam, perhaps influenced by the examples in Derek Parfit, Reasons and Persons, Oxford, Oxford University Press, 1984, ch. 3, but that is a different objection—and, in my view, a tempting but mistaken one; see Frank Jackson,

-245-

‘Group Morality’ in Philip Pettit and Richard Sylvan, eds, Metaphysics and Morality, Oxford, Basil Blackwell, 1987, pp. 91-110.

26 William Godwin, Enquiry Concerning Political Justice, Oxford, Oxford University Press, 1971. The example is given on p. 71; the point about character was something of an afterthought prompted by the reception that greeted his answer that you ought to abandon your father, see p. 325. Railton, ‘Alienation, Consequentialism, and the Demands of Morality’, takes a similar position on his example quoted earlier about a commuting marriage.

27 Railton, ‘Alienation, Consequentialism, and the Demands of Morality’, makes it clear that he is not offering such a variant. I am not so sure about Pettit and Brennan, ‘Restrictive Consequentialism’.

28 See, e.g., Railton, ‘Alienation, Consequentialism, and the Demands of Morality’, Sidgwick, The Methods of Ethics, and Derek Parfit, Reasons and Persons, § 14.

29 Nor am I denying the relevance of the distinction between evaluation of action and evaluation of character in explaining the very mixed feelings we would have towards a father who saved a stranger in preference to his daughter on the ground that he happened to know that the stranger was slightly more worthy of saving from an agent-neutral point of view. The distinction enables the consequentialist to explain the mixed feelings as a response to witnessing simultaneously the manifesting of a wrong character in a right action. I think that this is part of the purpose to which Godwin wished to put the distinction.



30 In Frank Jackson and Robert Pargetter, ‘Oughts, Options, and Actualism’, Philosophical Review, 95 (1986):233-55, we argue that this is true in general, not just for Consequentialism, provided only that consequences sometimes play a central role in determining what ought to be done. This view is controversial; see the extensive literature referred to therein. In Frank Jackson, ‘Understanding the Logic of Obligation’, Proceedings of the Aristotelian Society, supp. vol. 62 (1988):255-70, I argue that the best way to understand what is going on is in terms of early actions of the agent being actions of early temporal parts. (In both papers the argument is conducted, for reasons of expository convenience, mainly in terms of what objectively ought to be done, rather than the decision theoretic ‘ought’ that we have put in centre-stage here. I now think that this was an unfortunate choice of argumentative strategy: it obscured the point we have highlighted here—that ethics pertains most particularly to acting.)

31 I am indebted to David Lewis and Kim Sterelny for forcibly reminding me of this fact.

32 See, e.g., Shelly Kagan, The Limits of Morality, Oxford, Clarendon Press, 1989. 33 I am indebted to discussions with a number of audiences and to comments from

Michael Smith, Peter Singer, Philip Pettit and a referee (to whom I owe the title).

-246-

Part V INDUCTION

-247-


-248-



15 GRUE This chapter is concerned with an aspect of the problem of describing or specifying those inductive practices we take to be rational.

At the level of description, there is no doubt that one common inductive practice we take to be rational is to project common properties from samples to populations, to argue from certain Fs being G to certain other Fs being G. There are many ways we can try to spell out this practice in semi-formal terms: by saying ‘Fa & Ga’ confirms ‘(x)(Fx⊃Gx)’, or ‘All examined As are B’ supports ‘All unexamined As are B’, or ‘Fa1 &…& Fan’ gives a good reason for ‘Fan+1’, and so on. The precise way chosen will not particularly concern us, and I will simply refer to the kind of inductive argument pattern reflected in the various formalisations as the straight rule (SR). The discussion will be restricted to the simplest case where everything in a sample, not merely a percentage, has the property we are concerned with.

To say that the SR is one common inductive argument pattern we all acknowledge as rational is not to say that it is the most fundamental inductive argument pattern, or the most important in science, or the pattern that must be justified if induction is to be justified; it is simply to say what is undeniable—that we all use it on occasion and take it as rational to do so. This chapter is not concerned with how important or fundamental the SR is—for example, vis-à-vis hypothetico-deduction—it is concerned with the description of those applications of the SR which we regard as rational.

Since Nelson Goodman’s 1946 paper and the development of it in Fact, Fiction, and Forecast, 1 it has been very widely supposed that the rough description of the SR given above—as certain Fs

-249-



being G supporting certain other Fs being G—requires the insertion of a substantial proviso to the effect that the properties or predicates (or, in an alternative terminology, the hypotheses) involved be projectible. 2 The notion is that, though there are certain values of ‘F’ and ‘G’ for which it is manifestly true that the SR applies, there are other values for which it is manifestly false that the SR applies.

This gives rise to a new problem (Goodman’s new riddle) in inductive logic—that of demarcating the projectible predicates from the nonprojectible. The extensional aspect of this problem has not been as controversial as the intentional. There has been reasonable agreement about which predicates go into which class: ‘green’, ‘blue’, ‘round’, etc. into the projectible; ‘grue’, ‘bleen’, ‘sampled’, into the nonprojectible. But there has been enormous controversy over the rationale for this division: over what makes, for example, ‘grue’ nonprojectible and ‘green’ projectible. It has, to say the least, proved difficult to give a plausible, nonarbitrary account of the projectible/nonprojectible distinction other than the circular, useless one that a predicate is projectible just if the SR applies with respect to it.

I believe we can resolve the apparently interminable conflict over what it is about nonprojectible predicates that makes them so by challenging its very foundation. I will argue in this chapter that there is no ‘new riddle of induction’, by arguing that all (consistent) predicates are projectible and that there is no paradox resulting from ‘grue’ and like predicates.

The almost universal view that we need a distinction between projectible and nonprojectible predicates and hypotheses has had, I believe, three sources: one, a tendency to conflate three different ways of defining ‘grue’; two, a lack of precision about just how, in detail, the ‘grue’ paradox or new riddle of induction is supposed to arise; and, three, a failure to note a counterfactual condition that governs the vast majority of our applications of the SR. I will consider these matters in turn.

15.2 The three ways of defining ‘grue’

In this section I will consider the three common kinds of ways of defining ‘grue’ by considering typical instances of each. I will argue that the first two ways do not pose even a prima facie problem for the SR, leaving us with the third way to consider in later sections.

-250-



A typical example of the first way is:

D1 x is grue iff x is green before T and blue thereafter

where T is a chosen time in the future. 3

On D1 , ‘grue’ is atemporal—an object is grue or not once and for all; it cannot be grue at one time and not grue at another—and in this respect differs from ‘green’.

There seems no case for regarding ‘grue’ as nonprojectible if it is defined in this way. An emerald is grue1 4 just if it is green up to T and blue thereafter, and if we discovered that all the emeralds so far examined had this property, then, other things being equal, we would probably accept that all emeralds, both examined and unexamined, have this property of being green to a certain time and then turning blue; or, at least, would regard this hypothesis as supported.

We would in this case be regarding emeralds as like tomatoes and oranges, one of those things which change colour dramatically during their life cycles. No doubt we would seek an explanation for the fact that the change in emeralds occurs at a fixed time, T; but there would in principle be no impossibility about finding a satisfactory explanation. For example, we might discover that emeralds contain a radioactive element the radiation of which makes them green instead of blue, and that the level of this radiation is due to drop below a crucial figure at T.

(A puzzling feature of the discussions of the new riddle of induction by those who employ a D1 -type definition is that they take it as not in dispute that all emeralds observed to date are grue, as well as green. For example, Stephen Barker simply asserts as if it were an evident truth that ‘all the numerous emeralds that we have observed have been grue’ 5 —but what is an evident truth is that these emeralds were green at the time of observation; what we all believe is that they are always green; and what none of us believe is that they are grue1 , for none of us believe they will change to blue in the year 2000 (Barker’s choice for T).) 6

A typical example of the second way of defining ‘grue’ is:

D2 x is grue at t iff (x is green at t & t < T)

or (x is blue at t & t ≥ T). 7

-251-



‘Grue’ on this definition is like ‘green’ in being temporal: an object may be grue2 at one time and not at another.

It sometimes seems to be thought that D2 really amounts to D2 .1 : ‘x is grue’ means ‘x is green’ before T and ‘x is blue’ after T, which is an explicit case of ambiguity; and consequently that it raises no problem for the SR. 8 For when we read the SR as licensing the projection of a common predicate, it is understood that the predicate has the same meaning throughout.

The two definitions are not, however, equivalent. The appearance of equivalence arises from a failure to be explicit about time in D2 .1 , and if we write in a time variable to give: ‘x is grue at t’ means ‘x is green at t’ before T, and ‘x is blue at t’ after T, the disparity becomes obvious. Consider a time t1 before T, and whether a green emerald is grue at t1. According to D2 , the answer is an unequivocal yes; but, according to D2 .1 , the answer depends on the time at which the question is being asked. If the question is asked before T, the answer is yes; because before T, ‘x is grue2 .1 at t’ means ‘x is green at t’, and the emerald is green at t1: if asked after T, the answer is no; because after T, ‘x is grue2.1 at t’ means ‘x is blue at t’, and the emerald is not blue at t1. In short, D2 and D2.1 are not equivalent because the time at which we consider the question of an object’s grueness is relevant on D2.1 and not relevant on D2 —on D2, the time at which the object is green or blue is relevant, but not the time at which we consider the matter.

There is, I believe, no getting away from the fact that D2 is a perfectly proper, intelligible definition. Nevertheless, D2 does not give rise to a paradox or ‘new riddle’ when conjoined with the SR, and so does not give grounds for supposing that there are non-projectible predicates of which ‘grue2 ’ is the best-known example.

The contrary view has arisen from confusion over whether we are considering the SR in conjunction with ‘grue2 ’ as applied to objects that endure through time, that is, four-dimensional objects, or as applied to three-dimensional objects at times, that is, time-slices of the four-dimensional objects. 9

If we are considering the SR as applied to enduring objects like tables and emeralds, if we take the members of the samples and populations we discuss to endure through time, then we must read a temporal factor into the predicates with which the SR is concerned. Enduring objects aren’t red, or green, or square, simpli-

-252-



citer: they are red at t1, green at t2, and so on. A tomato isn’t both red and green; it is green early in its life history and red later.

From this it follows that when we read the SR (applied to enduring objects) as licensing the projection of common predicates from samples to populations, we must incorporate a temporal factor into these predicates. What we project must be understood as at a time; not just being green but being green at t. Only when this is overlooked does the appearance of paradox arise from applying SR with D2 , because the apparently paradoxical result only comes about with projections across T. To illustrate with the usual emerald case, suppose we have a sample of emeralds that are green at t1, where t1 is before T, then they will also be grue2 at t1; and, hence, the SR will equally lead to ‘All emeralds are green at t1’ and ‘All emeralds are grue2 at t1’. And these two universals are in no way incompatible. Whereas for time t2 after T, it is impossible that a sample of emeralds be both green at t2 and grue2 at t2, and so we cannot be led by the SR to hold together the incompatible universals: ‘All emeralds are green at t2 and ‘All emeralds are grue2 at t2’.

It is only if we slide illegitimately from t1 to t2 that an appearance of paradox arises. Only if we start from the fact that the sampled emeralds are both green at t1 and grue at t1 , and then, by conflating being green (grue) at t1 with being green (grue) at t2, wrongly take the SR to provide support equally for the incompatible ‘All emeralds are green at t2 and ‘All emeralds are grue at t2 , do we obtain an apparent paradox.

It may be objected that my insistence on the distinction between the predicates ‘x is grue (green) at t1’ and ‘x is grue (green) at t2 forces an unwelcome restriction on the role of the SR: sometimes we use the SR to argue from certain examined emeralds being green now to others being green now; sometimes from certain emeralds being green at one time, now, say, to certain others being green at a different time, in the future, say; and it may be thought that the second kind of use—when we go from the present to the future—requires ignoring the distinction between being F at t1 and being F at t2.

But this is to overlook the application of the SR to time-slices of objects as distinct from enduring objects. When we argue from the greenness of present emeralds to the greenness of future emeralds, we do best to view this as an application of the SR to temporal parts of emeralds, and so as an application involving, not being green at t true of an enduring emerald, but rather being

-253-



green simpliciter true of the temporal part at t of an emerald. When we wish to explicate our intuitive feeling that emeralds being green now supports their being green in the future by reference to the SR, by reference to the projection of common properties, we ought not to fudge the clear distinction between being green now and being green in the future; rather we should regard the projected property, being green, as a tenseless characteristic of present emerald temporal parts which is being projected to future temporal parts in accord with the SR.

(A question that might well be asked now is what happens to D2 if we recast it as a predicate on temporal parts instead of enduring objects. What happens, as can easily be seen, is that D2 becomes like D3 , below, in all respects essential to whether there is a ‘grue’ paradox; and hence does not call for separate treatment.)

Although D1 and D2 figure prominently in the ‘grue’ literature, they are not the kind of predicate with which Goodman launched it. 10 Goodman’s predicates are of the kind, ‘(x is green & Φx) v (x is blue & ~Φx)’, where ‘Φx’ is chosen so that its extension includes all the sampled (observed, examined, etc.) emeralds, that is, the emeralds from which we are imagined to be projecting, and so that the extension of ‘~Φx’ includes the other emeralds, those to which we are projecting. A simple way of doing this is to introduce a temporal factor into ‘Φx’, which is Goodman’s usual but not invariable practice; in particular, the following definition is close to that he uses in Fact, Fiction, and Forecast:

D3 x is grue at t iff (x is examined by T and x is green at t) or (x is not examined by T and x is blue at t).

As indicated by the ‘at t’ in D3 , this definition is for enduring objects. To avoid tedious repetition of the ‘at t’, we will conduct our discussion in the editorial present. Likewise, we will commonly drop the ‘at T’ by taking T to be a moment in the near future such that ‘examined by T’ just amounts to ‘examined (to date)’, and ‘not examined by T’ amounts to ‘unexamined (to date)’. Both these procedures are implicitly adopted by Goodman, so that being grue3 can be simply characterised as being green and examined, or being blue and unexamined. The paradox D3 appears to lead to, as we will see, is not essentially time-linked. It is not essential that we consider the sampled emeralds at one time, the remaining at another, to get an apparently paradoxical

-254-



result; so that, with D3 , by contrast with D2 , there is no objection to making things simpler by fudging a bit with respect to time.

D3— the correct definition in the sense that it gives rise to more trouble than D1 and D2 , as well as being Goodman’ s—will be the only definition we will be concerned with in the following sections, and when I refer to ‘grue’ and the alleged associated paradox or new riddle, I will mean ‘grue’ as defined in D3 .

15.2 The argument to paradox examined

Just what is the ‘grue’ paradox supposed to be; just what objectionable result is obtainable? In outline, the picture is clear enough. The idea is that, by suitable choice of predicates, the SR can be deployed to reach two incompatible conclusions starting from the same evidence. In particular, it is argued that a certain fact about emeralds when expressed in terms of ‘green’ leads to one projection about other emeralds when we apply the SR, and the same fact expressed in terms of ‘grue’ leads to another, incompatible projection when we apply the SR.

Though the picture is clear enough in outline, it starts to get murky as soon as we try to fill in the details. If, to fix our discussion, we consider a series of emeralds, a1,…, an, an+1, such that a1,…, an are known to be green and examined, while an+1 is known to be unexamined and is the emerald whose colour we are concerned to predict; precisely how does the SR lead to incompatible projections about an+1 from equivalent evidential bases?

Well, if we use ‘Grx’ for ‘x is green’, ‘Ex’ for ‘x is examined’, ‘Bx’ for ‘x is blue’, and ‘Gux’ for ‘x is grue’=‘(Grx & Ex) v (Bx & ~Ex), we are given

(1) Gra1, &…& Gran

and

(2) Gua1 &…& Guan.

But, first, (1) and (2) are not equivalent (neither entails the other), so there is no objection to the SR leading to different predictions (‘Gran+1’ and ‘Guan+1’, respectively) regarding an+1; second, the predictions are not inconsistent (neither entails the



-255-

denial of the other); and, finally, neither (1) nor (2) embodies our total evidence. 11

Our total evidence (or near enough for present purposes) is, rather, expressed by

(3) Gra1, & Ea1 &…& Gran & Ean

which is, of course, equivalent to

(4) Gua1 & Ea1 &…& Guan & Ean

But what (3) and (4) support by the SR is

(5) Gran+1 & Ean+1

and

(6) Guan+1 & Ean +1

respectively; which, far from being incompatible, are equivalent. Perhaps it will be argued that (5) entails (as it does)

(7) ~Ean+1⊃ Gran+1

and that (6) entails (as it does)

(8) ~Ean+1 ⊃Guan+1

which is equivalent to

(9) ~Ean+1⊃Ban+1

And that (7) expresses the prediction that if an+1 is not examined, it is green, whereas (9) expresses the incompatible prediction that if an+1 is not examined, it is blue. So we have derived opposite, incompatible predictions from equivalent bases, (3) and (4). 12



But this is like arguing that our observations of black ravens support white ravens being black, as follows: our observations support Joey, an as yet unobserved raven, being black. But ‘Joey is black’ entails ‘Joey is white ⊃ Joey is black’ so that our

-256-

observations support the prediction that if Joey is white, then he is black.

The fallacy here is obvious. We do have support for ‘Joey is white ⊃ Joey is black’, but only because we have support for the falsity of the antecedent. Likewise, we do have support on the basis of (3) and (4) for (7) and (9), but only because we have support for the falsity of their antecedents. It may be replied here that we do not have support for the falsity of their antecedents, that is, for an+1 being examined, because being examined, or ‘Ex’, is not projectible. But this is to assume that there are nonprojectible properties, in the course of an argument designed to show that there are; moreover, we will be giving reason later for allowing that being examined is projectible (in §15.4).

15.3 The counterfactual condition

So far I have not used the fact that we are given that an+1 is not examined. And in Goodman’s view our knowledge that there are unexamined emeralds is essential to deriving a paradox. He says, for instance,

If the hypothesis that all emeralds are green is also projected [i.e., in addition to ‘All emeralds are grue’], then the two projections disagree for unexamined emeralds. In saying these projections thus conflict, we are indeed assuming that there is some unexamined emerald to which only one of the two consequent predicates applies, but it is upon just this assumption that the problem arises at all. 13

But just how can we use the fact that an+1 is not examined? It sometimes seems to be thought that it is proper to add in this additional information in a more or less mechanical fashion, somewhat as follows:

Our evidence supports that an+1 is green and examined. We know independently that an+1 is not examined, hence our overall evidence supports that an+1 is green and not examined. Equally, as far as the SR goes, our evidence supports that an+I is grue and examined, and so, via



the same line of argument, we arrive at our overall evidence

-257-

supporting that an+1 is grue and not examined, which entails that an+1 is not green.

There is no question that we have here genuinely incompatible predictions about the colour of an+1 for we have categoricals, not material implications. But we also have a pattern of argument that is quite certainly fallacious.

The pattern is: if a proposition, p, which we know to be true, supports a conjunction, q & r, one conjunct, r, of which we know independently to be false; we have, overall, support for q & ~r, and so, for anything it entails.

Once this pattern is explicitly set out, I doubt if anyone would assent to it; for it leads easily to an inconsistency, as follows: p supports (q & r) if and only if p supports [(q & r) v (~q & ~r)] & r, for the latter is truth-functionally equivalent to (q & r). Hence, by the argument pattern just displayed, when I know r to be false on independent grounds, I have, over-all, support equally for (q & ~r) and for [(q & r) v (~q & ~r)] & ~r, which are truth-functionally inconsistent—the latter is equivalent to (~q & ~r).

It is equally clear from actual examples that this argument pattern is fallacious. Suppose a reliable friend tells me that Hyperion won the cup by five lengths, then I have support for the conjunction, ‘Hyperion won the cup and Hyperion won by five lengths’. Further suppose I have quite decisive, independent evidence that the winning margin in the cup was only three lengths, but that this evidence is neutral as to who won by that margin. Do I have, over-all, evidence for Hyperion winning, though not by five lengths? If the argument pattern in question were valid, the answer would be an invariable yes; but in fact the answer is that it all depends on the circumstances. In some it will be most rational for me to take the error as to winning margin as indicating that my normally reliable friend is having one of her few off days and so is not to be trusted concerning the identity of the winner either; in other circumstances it will be most rational for me to take it that my friend regarded the winning margin as a relatively unimportant detail compared to the identity of the winner, and was her usual reliable self concerning the latter.

What we have here is, of course, just an aspect of the universally acknowledged fact that inductive support is defeasible; and it is strange how often this defeasibility is overlooked in the context of discussions of ‘grue’. For example, it is common to find it



suggested that by means of a ‘grue’-type manoeuvre it is

-258-

easy to show that an unrestricted SR leads to the unacceptable consequence that any n objects support some (n+1)st object being G, for any ‘G’, 14 as follows: For any (n+1) objects, there will be an ‘Fx’ such that it is true of the first n, but not the (n+ 1)st. But if ‘Fx’ is true of the first n, so is ‘Fx v Gx’ for any ‘Gx’; therefore, runs the argument, the (unrestricted) SR supports the (n + 1)st object being F or G. But it is given as not being F; so it is concluded that we are led to the absurdity that we have support for the (n+1)st object being G, for any ‘G’. Now if something like: ‘If p supports q, then (p & r) supports (q & r)’ were valid, all would be well with this argument; but we all know that nothing like this is valid, and so, that the information that the (n+1)st object is not F cannot be incorporated in so simple a fashion.

We must, therefore, proceed very carefully when attempting to incorporate the additional information that an+1 is unexamined, and in particular we must, I think, see the matter in context.

The general context is this: we have a sample, a1…, an, each of which has a property, being examined, in addition to the particular properties we are interested in (being green and being grue) and which is given as not being possessed by an+1. This kind of situation arises virtually whenever we use the straight rule. When we use the SR to project common properties from a sample to members of the population from which the sample comes, there are nearly always features common to every member of the sample which we know are not features of all (or any) members of the population outside the sample. Some of these common sample features are normally disregarded as being unimportant, indeed trivial, like being sampled, being one of a1, a2, …, and being examined (before…); while others clearly cannot be disregarded, as, for instance, in the following cases. Every diamond I have observed has glinted in the light. Does this support the contention that the next diamond I observe will glint in the light? Clearly, yes. But suppose we add a detail to the story, namely, that the next diamond that I observe is unpolished. Now all the diamonds I have observed so far have been polished, and, moreover, I know that they glint because they have been polished—that is, if the diamonds had not been polished, then they would not have glinted. It is clear that once we add this detail, it is no longer reasonable for me to regard it as likely that the next diamond I observe will glint in the light. The fact that all the diamonds I have observed glint in the light supports the next diamond I observe

-259-



will glint; but the fact that all the polished diamonds I have observed glint, when taken in conjunction with my knowledge that they would not have glinted if unpolished, does not support an unpolished one glinting.

A similar example is afforded by lobsters. Every lobster I have observed has been red. This supports that the next lobster I observe will be red, and no doubt it will be. But every lobster I have observed has been cooked, and I know that it is the cooking that makes them red—that is, that the lobsters I have observed would not have been red if they had not been cooked. Hence I do not regard myself as having good evidence that the next uncooked lobster I observe will be red.

We have here two cases where certain Fs being G supports, by the SR, other Fs being G, but certain Fs which are H being G does not support other Fs which are not H being G; in each case the reason being that it is known that the Fs that form the evidence class would not have been G if they had not been H. The condition: that certain Fs which are H being G does not support other Fs which are not H being G if it is known that the Fs in the evidence class would not have been G if they had not been H, will be referred to as the counterfactual condition. I cannot think of any way of proving it, as opposed to illustrating it as I just have, but also I cannot think that anyone will seriously deny it. (For ease of reading, I have expressed the condition in a conditional form. Strictly, it should be expressed as that the conjunction of certain Fs which are H being G with these Fs being such that if they had not been H, they would not have been G, does not support other non-H Fs being G.)

We are now in a position to discuss the incorporation of the additional information that an+1 is unexamined. When we argue from examined emeralds a1…, an, being green to the unexamined an+1 being green, we are arguing from certain Fs which are H being G to an F which is not H being G, in the special case got by replacing ‘F’ by ‘emerald’, ‘G’ by ’green’, and ‘H’ by ‘examined’. Hence the counterfactual condition is that the emeralds a1,…, an, would still have been green even if they had not been examined; and, in the world as we know it, this condition is satisfied. The emeralds we have examined are green not because they have been examined but because of their chemical composition and crystalline structure, and so, like most objects in our world, they would have had the colour they do have whether or not they had been examined.

-260-



Precisely the opposite is the case with ‘grue’. We know that an emerald that is grue and examined would not have been grue if it had not been examined; for if it is grue and examined, it is green and examined, and, as noted already, if it had not been examined would still have been green; but then it would have been green and unexamined, and so, not grue. In other words, a green, examined emerald would have been a green, unexamined emerald if it had not been examined, and so a1,…, an would not have been grue if they had not been examined. Therefore, to use the SR to yield the prediction that an+1 is grue (and unexamined) is to violate the counterfactual condition.

In sum, the position is this. If we use the SR with the evidence that a1,…, an are green and examined, and grue and examined, ignoring the fact that an+1 is unexamined, we get support for ‘an+1 is green and examined’ and for ‘an+1 is grue and examined’; which, far from being inconsistent, are equivalent. If we bring in the fact that an+1 is unexamined, we no longer are dealing with a case of certain Fs being G supporting other Fs being G, but of certain Fs which are H being G supporting certain other Fs which are not H being G, and hence must take note of the counterfactual condition. But if we take note of this condition, we do not get an inconsistency because—although a1,…, an would still have been green if they had not been examined—they would not have been grue if they had not been examined. Moreover, not only don’t we get an inconsistency, we cannot get one, because it cannot be the case both that if X had not been H, it would not have been G, and if X had not been H, it would have been G—at least, on standard views about the logic of counterfactuals.

Our discussion of the SR has been couched in terms of constants, ‘a1’,…, ‘an+1’, taken to designate emeralds. It is common to discuss the SR in terms of universals. The counterfactual condition shows, I think, that it can be misleading to characterise the SR as ‘All examined As are B’ supports ‘All unexamined As are B’.

There are cases where it is absurd to take ‘All examined As are B’ as supporting ‘All unexamined As are B’. Some properties of the elementary particles of physics are known to be affected by examination of the particles. It would be absurd to argue, for such a property, that, since all examined particles have it, so do all unexamined particles, just because we know that if the particles

-261-



in question had not been examined, they would not have had the property.

Moreover, we do not have to turn to recondite entities like sub-microscopic particles for examples of properties such that something would not have them if they had not been examined. Examined emeralds have a property of just this kind, namely, being grue. Take an emerald that is green and examined, and so, grue. If it had not been examined, it would still have been green, because examining emeralds (and indeed examining most things) doesn’t alter their colour; therefore, if the emerald had not been examined, it would have been green and unexamined, and so, not grue. Hence, it is a mistake to argue from ‘All examined emeralds are grue’ to ‘All unexamined emeralds are grue’, not because ‘grue’ is intrinsically nonprojectible, but simply because the counterfactual condition is violated.

Parallel remarks apply to functor expressions of the SR such as: ‘All examined As are B’ supports ‘The first unexamined A is a B’. We get an apparently simple and decisive development of the ‘grue’ paradox by noting that: ‘All examined emeralds are green’ and ‘All examined emeralds are grue’ are equivalent, and that: ‘The first unexamined emerald is green’ and ‘The first unexamined emerald is grue’ are inconsistent. 15 But, evidently, it is reasonable to use this kind of version of the SR only when being B is appropriately independent of being examined, and this is not the case when being B is being grue.

It is perhaps unfortunate that being examined (observed, sampled, etc.) appears so frequently in statements of the SR. The SR is intended as an essentially relational principle of inductive support concerning whether p supports q, quite independently of whether p is known. Examining, observing, sampling, and so on are how we human beings come to know that certain As are B; but our knowing this is separate from these As being B supporting certain other As being B. What we come to know does the supporting (if any), not our coming to know it.

Though it is a fact about our world that the emeralds we have examined would still have been green if they had not been examined, it might not have been a fact. We might have lived in a world in which they would not have been green if they had not been examined. For example, we might have lived in a world in which all examined emeralds were green and in which investigation of the crystalline structure of these emeralds reveals that they are naturally blue; this structure being affected by the light

-262-



necessarily involved in examining them in such a way that emeralds turn green instantaneously on being examined. 16 In this world, all emeralds we have direct observational evidence concerning are green and examined and grue. What ought we believe about those not examined? Obviously, that they are blue, and hence that all emeralds are grue. Our counterfactual condition explains this. In this world, examined emeralds are both green and grue, as in our world, but, as not in our world, if they had not been examined, they would have been grue, not green.

15.4 The projectibility of being sampled

Our counterfactual condition also bears on the question of the projectibility of such properties as being sampled, being examined, and being one of a1,…, an. Richard Jeffrey holds that—whereas it may just be doubted that ‘grue’ is nonprojectible—it is beyond doubt that such properties as these are nonprojectible. 17

Why is he so certain? No doubt the kind of case he has in mind is where I am drawing marbles from a barrel and noting that each marble is red. Normally we suppose this to support that the remaining marbles are red. But, equally, each marble drawn will have the property of being sampled, and we do not normally regard the proposition that the remaining marbles are sampled as being supported by this.

But it would be too hasty to infer from this point that being sampled is not projectible. Suppose my reason for thinking that all the marbles drawn out have been sampled is that they each have Jones’s finger prints on them, and so must have been sampled (in the past, by Jones). Then it is clear that I will be entitled to increase my degree of belief that the remaining marbles have been sampled.

What is the explanation for the dramatic change whereby it is evidently absurd to increase one’s expectation that the remaining marbles are sampled in the first case, and evidently not absurd in the second? I think it would be a mistake to explain this change in terms of the projectible/nonprojectible distinction by saying that being sampled by me now is nonprojectible, whereas being sampled by Jones in the past is projectible. For suppose that in the first case I am Jones and that after drawing the red marbles I go out for a cup of coffee; on my return I am confronted by a group of marbles all of which have the property of being sampled by Jones in the past. Do I now increase my expectation that the

-263-



remaining marbles have this, allegedly projectible, property? Quite obviously, no. The projectible/nonprojectible property distinction cannot explain the divergence in our inductive behaviour in the two cases—and, surely, this is just the kind of case that the distinction, if it is worth making, ought to help us with.

What does explain the divergence is our counterfactual condition. In the first case, we have certain marbles, all of which are sampled and all of which have just been drawn from the barrel, and are concerned with whether we have support for certain other marbles, not drawn from the barrel, being sampled. We do not, because we know how it is that the sampled marbles came to be sampled, namely, by being drawn out. Hence, if the marbles had not been drawn out, they would not have been sampled; and our counterfactual condition is violated. On the other hand, in the second case (where I discover that the marbles have been sampled by observing Jones’s fingerprints on them), the marbles drawn out would still have been sampled (by Jones, in the past) even if they had not been drawn out by me. The counterfactual condition is not violated, and we therefore have in the second case support for the marbles not drawn out being sampled.

There is nothing intrinsically nonprojectible about being sampled. In some cases, it is perfectly reasonable to project it, and in those cases where it is not, the explanation does not have to do with the nature of the property or the meaning of the corresponding predicate, that is, does not relate to a feature of being sampled that calls for a label such as ‘nonprojectible’, but is rather that the counterfactual condition is violated. Exactly similar remarks apply to being examined and to being one of a1,…, an. I will look briefly at the latter.

Despite the frequency and confidence with which it is said that properties of the being one of a1 ,…kind are not projectible, it is easy to describe the counter cases. Suppose I am a policeman investigating a series of cat burglaries, and I discover that in each case the person responsible is one of Tom, Dick and Harry; then I will be entitled to regard ‘The person responsible for the next cat burglary will be Tom, Dick or Harry’ as supported. Again, if I am drawing marbles from a barrel and find each one stamped with a name, and the name is always one of ‘a1’, ‘a2’,…,‘an’, I will have increasing support for the next marble being one of a1, …, an. (Of course, after I have drawn out all of a1,…, an marbles and if I am drawing without replacement, I won’t expect the next marble to be one of a1, …, an; but this is because I am acquainted

-264-



with the necessary truth that n things cannot be identical with n+1 things, and shows, not nonprojectibility, but the role of additional negative evidence). 18

By way of contrast, if I don’t find the names already stamped on the marbles, but give the names to the marbles as they are drawn out, I won’t expect marbles not drawn out to be identical with one of a1,…, an. This is because the counterfactual condition is violated; I know that the marbles drawn out would not have the names they do if they had not been drawn out.

Whenever we apply the SR, we know, as it were, too much. I am drawing marbles from the ubiquitous barrel, and in consequence the drawn marbles are in my hand, recently exposed to light, and observed. These are all things I know about the marbles which I would not dream of projecting to the marbles remaining in the barrel: not because these properties are intrinsically non-projectible—there are obviously many cases where we would project, for instance, being recently exposed to light—but because I know how the drawn marbles came to be recently exposed to light (to single this property out for discussion), namely, as a result of being sampled. Therefore, if they had not been drawn out, they would not have been recently exposed to light, and so, the argument from the drawn marbles being recently exposed to light to the undrawn marbles being so exposed violates the counterfactual condition.

I expect that two objections will be generated by the prominent role of the counterfactual condition in the above discussion. The first is the general objection that counterfactuals raise some of the most difficult problems in philosophy. This is true, but the fact remains that we do, on occasion, know with certainty that certain counterfactuals are true, despite the difficulties in analysing just what it is that we know on such occasions and how we know it. Perhaps one day we will have a good theory of counterfactuals, or a way of eliminating the need for them; until then we must put up with them.

The second objection is the more particular one that, by appealing to the counterfactuals that I appeal to, I am introducing a kind of circularity. Take for example my reason for saying that the SR favours unexamined emeralds being green rather than grue: that the emeralds we have in fact examined would have been green, not grue, if they had not been examined. There is no disputing the fact that we do know this—it is as certain as is any knowledge of the form: if a had not been X, it would have been Y—but it might

-265-



be objected that we know this only because we know unexamined emeralds are green. Hence, on pain of circularity, we cannot appeal to this fact to explain why the SR leads to the prediction that unexamined emeralds are green.

However, our knowledge that the examined emeralds would still have been green if they had not been examined is knowledge about the examined emeralds, not about the unexamined ones. It is knowledge we might have had even if there were no unexamined emeralds to be green or not green. If it turned out that there were very many fewer emeralds than we at first thought, and that in fact every emerald has been discovered and examined, this would not alter the fact that if the examined emeralds had not been examined they would have been green. Moreover, this fact is quite consistent with the unexamined emeralds turning out to be, to our great surprise, red; the result, say, of the emeralds so far examined all coming from regions in the world where certain minerals that make things green abound, and those not so far examined coming from regions containing minerals that make things red. This surprising discovery would not undermine our belief that the so-far-examined emeralds would have been green even if not examined.

It follows that our knowledge that the examined emeralds would be green even if not examined does not tacitly rest on our knowledge that unexamined emeralds are green. It is knowledge we might have had even if unexamined emeralds were not green or, indeed, were nonexistent, and so is knowledge we may appeal to without circularity in describing our application of the straight rule in a way that makes clear why we have support for unexamined emeralds being green rather than grue.

The point is more obvious in the marble-barrel case. I may know that each of the three red marbles that I drew out of the barrel would still have been red even if it had not been drawn out, without knowing the colour of the remaining marbles—indeed, I commonly will know the former without knowing the latter. To know the former is to know something about the lack of connection between the colour of an object and whether or not it is examined in the given case, whatever that colour may be, and is not dependent on knowledge of the particular colour of a particular object or objects, be they drawn out or not. Similar remarks apply in the converse case where I discover that the red marbles are painted with a special paint that turns red immediately on contact with a human hand (due, say, to the warmth). In this case

-266-



the marbles would not have been red if not drawn out (by hand), and we would not increase our expectation that the remaining marbles are red for just this reason. It is quite obvious that I may know the relevant facts about the paint without knowing or having any idea of the colour of the remaining marbles. There is thus no circularity. (Likewise, Goodman’s appeal to the entrenchment of predicates isn’t circular, though it also involves appeal to inductively gained knowledge. The objection to entrenchment is rather that it is excessively anthropocentric.)

15.5 Summary

The over-all position is this. The SR: certain Fs being G supporting other Fs being G does not lead to incompatible predictions when combined with ‘grue’ and like predicates.

When we apply the SR in practice, we commonly argue on the modified pattern: certain Fs which are H being G supports Fs which are not H being G (‘H’ often being ‘examined’, ‘sampled’, etc.). When we argue on this modified pattern, we take it that it is not the case that the Fs which are H would not have been G if they had not been H. This guarantees that we can never be led from the same evidence to opposite predictions concerning whether the non-H Fs are G. For, though Fs are H and G if and only if they are H and G*, where ‘G*x’=‘(Hx & Gx) v (~Hx & ~Gx)’, we cannot be led both to the non-H Fs being G and to their being G*, and so—as a non-H F is G* just if ~G—to opposite predictions. This is because we know from the logic of counterfactuals that it cannot both be the case that the Fs which are H and G would have been G if they had not been H, and that they would have been G* if they had

not been H; for this amounts to and

being true together, since a non-H is G* only if ~G. 19

To arrive at counterfactuals of the required form, we must, of course, draw on our knowledge of the world. Just which knowledge is as controversial as counterfactuals—that is, very. But it is clear that the knowledge required is not the knowledge at issue in the particular application of the SR in question, and so it is not circular to appeal to it. And of course it is not controversial that applying the straight rule in a particular case requires reference, inter alia, to knowledge gained inductively from other applications of the SR. Even knowing that certain Fs are G requires trusting



one’s senses, memory, the reliability of reference books,

-267-

and so on. This may well raise fundamental problems at the level of justification, in the context of the ‘old problem of induction’, but this has not been our concern here. Our concern here at the level of description has been to urge that the SR can be specified without invoking a partition of predicates, properties, or hypotheses into the projectible and the nonprojectible. 20

Added note: discussions subsequent to the original publication of this chapter have convinced me that I should emphasise more strongly that my claim is that grue and the like are projectible, that is, able to be projected. My objection is to the division of properties into those which can and those which cannot properly be projected. Nothing I say in the chapter counts against the view that, against the background assumptions most of us share, we would need more evidence to project grue than we would to project green. My hope was to direct attention away from the misguided (as I see it) attempt to discern what makes some properties unprojectible, and towards the enterprise of describing the general features of our epistemic take on the world that underlie our readiness to project one or another property. (If I am right in what I argue in the chapter, one important feature is our acceptance of certain counterfactuals.) It may be that, in carrying out this enterprise, we find that the distribution of prior probabilities is crucial, and further, that in this distribution, not all properties are equal. It may be that our distribution of priors discriminates against the ‘gruesome’; indeed, this seems to me to be the plausible view to take. So nothing I say in the chapter goes against the view that there is an important distinction between the more or less gruesome. My target is the common view that the gruesome is, by its very nature, unfit for projection.

Notes 1 Nelson Goodman, ‘A Query on Confirmation’, Journal of Philosophy, 43

(1946):383-5, and Fact, Fiction, and Forecast, Cambridge, Mass., Harvard University Press, 1955, ch. 3.

2 I will talk primarily in terms of the projectibility or otherwise of properties, predicates, or open sentences (the differences among these three not being relevant to the arguments that follow), rather than hypotheses; that is, I will follow Goodman’s usage in ‘A Query’ rather than in Fact, Fiction, and Forecast.

3 D1 -type definitions appear in the discussions of projectibility by H. Kyburg, Probability and Inductive Logic, Toronto, Macmillan, 1970;



-268-

I. Hacking, The Logic of Statistical Inference, New York, Cambridge University Press, 1965; S. Barker, Induction and Hypothesis, Ithaca, N.Y., Cornell University Press, 1957.

4 When it is not clear from the context, ‘gruei ’ is used for: ‘grue’ defined according to Di

5 Induction and Hypothesis, p. 189. 6 I have kept my discussion of D1 brief, since similar points have been well made

by Simon Blackburn, ‘Goodman’s Paradox’, American Philosophical Quarterly, monograph no. 3, 1969, and M. Kelley, ‘Predicates and Projectibility’, Canadian Journal of Philosophy, 1 (1971):189-206.

7 This kind of definition appears in W. Salmon, ‘On Vindicating Induction’ in H. Kyburg and E. Nagel, eds, Induction: Some Current Issues, Middletown, Conn., Wesleyan, 1963; P. Achinstein and S. Barker, ‘On the New Riddle of Induction’, Philosophical Review, 69 (1960):511-22; B. Skyrms, Choice and Chance, Belmont, Calif., Dickenson, 1966. The tendency (e.g., by Barker and Kyburg) to slip between D1 and D2 may be due to the fact that ‘x is grue1 ≡ (t)(x is grue2 at t)’ is true.

8 See, e.g., Kelley, ‘Predicates and Projectibility’, §III. 9 Added note: the phrasing here and immediately following might well suggest that

the point being made assumes four-dimensionalism, but in fact it is independent of this controversial (though plausible, I think) view. The point simply turns on the need, in discussions of SR, to distinguish projecting time-indexed predicates and properties from projecting properties and predicates of things identified at times.

10 As Goodman points out in Problems and Projects, Indianapolis, Bobbs-Merrill, 1972, see p. 359.

11 As, in effect, Rudolf Carnap points out in ‘On the Application of Inductive Logic,’ Philosophy and Phenomenological Research, 8 (1947):133-47, see §3.

12 I take this to be essentially the argument in H. Leblanc, ‘That Positive Instances Are No Help’, Journal of Philosophy, 60 (1963): 452-62.

13 Fact, Fiction, and Forecast, p. 94 of the second edition, Indianapolis, Bobbs-Merrill, 1965. My emphasis.

14 See, e.g., Skyrms, Choice and Chance, pp. 61-2. 15 As in W.V. Quine, ‘Natural Kinds’ in Ontological Relativity and Other Essays,

New York, Columbia University Press, 1969, pp. 114-38, see p. 115. 16 And we could bring in the time factor by, for instance, supposing the method of

examining changes at T. 17 Richard C. Jeffrey, ‘Goodman’s Query’, Journal of Philosophy, 63 (1966):281-8,

see p. 288. He actually has ‘bleen’ for ‘grue’ in the relevant passage.



18 Cf.Kelley, ‘Predicates and Projectibility’, p. 196. 19 Added note: Strictly, the key point is that it cannot be rational to accept

‘divergent’ counterfactuals. 20 This chapter has benefited considerably from discussions with colleagues,

particularly with Robert Pargetter.

-269-

BIBLIOGRAPHY a

Achinstein, P. and S. Barker, ‘On the New Riddle of Induction’, Philosophical Review, 69 (1960):511-22. Adams, Ernest W., The Logic of Conditionals, Dordrecht, Reidel, 1975. ——‘Subjunctive and Indicative Conditionals’, Foundations of Language, 6 (1970):89-94. ——‘Probability and the Logic of Conditionals,’ in J. Hintikka and P. Suppes, eds, Aspects of Inductive Logic, Amsterdam, North-Holland, 1966, pp. 265-316. ——‘The Logic of Conditionals’, Inquiry, 8, 2 (1965):166-97. Anderson, A.R., and N.D. Belnap, Entailment, Princeton, N.J., Princeton University Press, 1975. Armstrong, D.M., A Combinatorial Theory of Possibility, Cambridge, Cambridge University Press, 1989. ——Universals: An Opinionated Introduction, Boulder, Westview Press, 1989. ——What is a Law of Nature?, Cambridge, Cambridge University Press, 1983. ——‘Identity Through Time’ in Peter Van Inwagen, ed., Time and Cause, Dordrecht, Reidel, 1980. ——Belief, Truth and Knowledge, Cambridge, Cambridge University Press , 1973. ——A Materialist Theory of the Mind, London, Routledge & Kegan Paul, 1968. Ayer, A.J., Probability and Evidence, London, Macmillan, 1972.

b Barker, S., Induction and Hypothesis, Ithaca, N.Y., Cornell University Press, 1957. Bennett, Jonathan, ‘Classifying Conditionals: The Traditional Way is Right’, Mind, 104 (1995):331-54. ——‘Farewell to the Phlogiston Theory of Conditionals’, Mind, 97 (1988):509-27. ——Linguistic Behaviour, Cambridge, Cambridge University Press, 1976.

-270-



Bentham, Jeremy, An Introduction to the Principles of Morals and Legislation, London, Athlone, 1970. Bigelow, John and Robert Pargetter, Science and Necessity, Cambridge University Press, Cambridge, 1990. Blackburn, Simon, ‘Goodman’s Paradox’, American Philosophical Quarterly, monograph no. 3, 1969. ——Spreading the Word, Oxford, Oxford University Press, 1984. Block, Ned, ‘Can the Mind Change the World?’ in George Boolos, ed., Meaning and Method: Essays in Honor of Hilary Putnam, Cambridge, Cambridge University Press, 1990, pp. 137-70. Braddon-Mitchell, David and Frank Jackson, ‘The Teleological Theory of Content’, Australasian Journal of Philosophy, 75 (1997):474-89. ——Philosophy of Mind and Cognition, Oxford, Basil Blackwell, 1996. Brink, David O., Moral Realism and the Foundations of Ethics, Cambridge, Cambridge University Press, 1989. Burge, Tyler, ‘Mind-Body Causation and Explanatory Practice’ in John Heil and Al Mele, eds, Mental Causation, Oxford, Clarendon Press, 1993, pp. 97-120. ——‘Individualism and Psychology’, Philosophical Review, 95 (1986): 3-45. ——‘Individualism and the Mental’, Midwest Studies in Philosophy, 5 (1979):73-122.

c Campbell, Keith, Metaphysics, Encino, Calif., Dickenson, 1976. ——Body and Mind, London, Macmillan, 1971. Carnap, Rudolf, ‘On the Application of Inductive Logic,’ Philosophy and Phenomenological Research, 8 (1947):133-47. Chellas, Brian F., ‘Conditional Obligation’ in Søren Stenlund, ed., Logical Theory and Semantic Analysis, Dordrecht, Reidel, 1974. Child, William, Causality, Interpretation, and the Mind, Oxford, Clarendon Press, 1994. Chisholm, R.M., Perceiving, Ithaca, Cornell University Press, 1957. Churchland, Paul, ‘Folk Psychology and the Explanation of Human Behaviour’, Proceedings of the Aristotelian Society, supp. vol. 62 (1988):209-21. ——‘Reduction, Qualia, and the Direct Introspection of Brain States’, Journal of Philosophy, 83 (1985):8-28. Clark, Michael, ‘Ifs and Hooks: A Rejoinder,’ Analysis, 34 (1974):77-83. ——‘Ifs and Hooks’, Analysis, 32 (1971):33-9. Cohen, L. Jonathan, ‘Some Remarks on Grice’s Views about the Logical Particles of Natural Language’ in Y. Bar-Hillel, ed., Pragmatics of Natural Languages, Dordrecht, Reidel, 1971.



Cornman, J.W., Materialism and Sensations, New Haven, Yale University Press, 1971.

-271-

Crane, Tim, ‘Mental Causation’, Proceedings of the Aristotelian Society, supp. vol. 69 (1995):211-36.

d Davidson, Donald, ‘Thinking Causes’ in John Heil and Al Mele, eds, Mental Causation, Oxford, Clarendon Press, 1993, pp. 3-17. ——‘Knowing One’s Own Mind’, Proceedings of the American Philosophical Society, 60 (1986):441-58. ——‘Actions, Reasons and Causes’, reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980, pp. 3-19. ——‘Agency’, reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980, pp. 43-61. ——‘Causal Relations’, reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980, pp. 149-62. ——‘How is Weakness of Will Possible?’ reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980, pp. 21-42. ——‘Psychology as Philosophy’, reprinted in Essays on Actions and Events, Oxford, Clarendon Press, 1980, pp. 229-39. Davies, M.K. and I.L. Humberstone, ‘Two Notions of Necessity’, Philosophical Studies, 38 (1980):1-30. Dennett, Daniel, ‘Reflections: Real Patterns, Deeper Facts, and Empty Questions’, reprinted in The Intentional Stance, Cambridge, Mass., MIT Press, 1989, pp. 37-42. ——‘True Believers’, reprinted in The Intentional Stance, Cambridge, Mass., MIT Press, 1989, pp. 13-35. ——‘Current Issues in the Philosophy of Mind’, American Philosophical Quarterly, 15 (1978):249-61. Devitt, Michael, ‘A Narrow Representational Theory of Mind’ in W.G. Lycan, ed., Mind and Cognition, Oxford, Basil Blackwell, 1990, pp. 371-98. ——Realism and Truth, Oxford, Basil Blackwell, 1984. Dretske, Fred I., ‘Mental Events as Structuring Causes of Behaviour’ in John Heil and Al Mele, eds, Mental Causation, Oxford, Clarendon Press, 1993, pp. 121-36. ——‘Laws of Nature’, Philosophy of Science, 44 (1977):65-72. Dudman, V.H., ‘Jackson Classifying Conditionals’, Analysis, 51 (1991): 131-6. Dummett, Michael, Frege, London, Duckworth, 1973.

e



Edgington, Dorothy, ‘Do Conditionals have Truth Conditions?’, Critica, 98 (1986):3-30, reprinted in Frank Jackson, ed., Conditionals, Oxford, Oxford University Press, 1991. Eells, Ellery, Rational Decision and Causality, Cambridge, Cambridge University Press, 1982. Ellis, Brian, ‘A Unified Theory of Conditionals’, Journal Of Philosophical Logic, 7 (1978):107-24.

——‘An Epistemological Concept of Truth,’ in R. Brown and C.D.

-272-

Rollins, eds, Contemporary Philosophy in Australia, London, Allen and Unwin, 1969.

f Feldman, Fred, Doing the Best We Can, Dordrecht, Reidel, 1986. Field, Hartry, ‘Mental Representation’ in Ned Block, ed., Readings in Philosophical Psychology, vol. II, Cambridge, Mass., Harvard University Press, 1981, pp. 78-114. Fodor, Jerry, ‘Methodological Solipsism Considered as a Research Strategy in Cognitive Psychology’, Behavioral and Brain Sciences, 3 (1980):63-109. ——‘Special Sciences, or the Disunity of Science as a Working Hypothesis’, Synthese, 28 (1974):97-115. ——Psychosemantics, Cambridge, Mass., MIT Press, 1987. Føllesdal, Dagfinn and Risto Hilpinen, ‘Deontic Logic: An Introduction’ in Føllesdal and Hilpinen, eds, Deontic Logic: Introductory and Systematic Readings, Dordrecht, Reidel, 1971. Forrest, Peter, ‘Universals and Universalizability’, Australasian Journal of Philosophy, 70 (1992):93-8. ——‘Ways Worlds Could Be’, Australasian Journal of Philosophy, 64 (1986):15-24.

g Geach, P.T., ‘Some Problems about Time’ in Logic Matters, Oxford, Basil Blackwell, 1972, pp. 302-18. Gibbard, Allan, ‘Two Recent Theories of Conditionals’ in W.L. Harper et al, eds, Ifs, Dordrecht, Reidel, 1981, pp. 211-47.



Gibbard, Allan and William L. Harper, ‘Counterfactuals and Two Kinds of Expected Utility’, reprinted in W.L. Harper et al., eds, Ifs, Dordrecht, Reidel, 1981, pp. 153-90. Godwin, William, Enquiry Concerning Political Justice, Oxford, Oxford University Press, 1971. Goldman, Alvin, A Theory of Human Action, Englewood Cliffs, N.J., Prentice Hall, 1970. Goldman, Holly S., ‘Doing The Best One Can’ in A.I. Goldman and J. Kim, eds, Values and Morals, Dordrecht, Reidel, 1978. ——‘David Lewis’s Semantics for Deontic Logic’, Mind, 86 (1977): 242-8. ——‘Dated Rightness and Moral Imperfection’, Philosophical Review, 75 (1976):449-87. Goodman, Nelson, Problems and Projects, Indianapolis, Bobbs-Merrill, 1972. ——Fact, Fiction, and Forecast, Cambridge, Mass., Harvard University Press, 1955. ——‘A Query on Confirmation’, Journal of Philosophy, 43 (1946): 383-5. Greenspan, P.S., ‘Oughts and Determinism’, Philosophical Review, 77 (1978):77-83.

-273-

——‘Conditional Oughts and Hypothetical Imperatives’, Journal of Philosophy, 72 (1975):259-76. Grice, H.P., Studies in the Ways of Words Cambridge, Mass., Harvard University Press, 1989. ——‘Further Notes on Logic and Conversation,’ in Peter Cole, ed., Syntax and Semantics, New York, Academic Press, vol. 9, 1978. ——‘Logic and Conversation’ in Peter Cole and Jerry L. Morgan, eds, Syntax and Semantics, New York, Academic Press, vol. 3, 1975. Gruzalski, Bart, ‘The Defeat of Utilitarian Generalization’, Ethics, 93 (1982):22-38.

h Hacking, Ian, The Logic of Statistical Inference, New York, Cambridge University Press, 1965. Haldane, John, ‘Folk Psychology and the Explanation of Human Behaviour’, Proceedings of the Aristotelian Society, supp. vol. 62 (1988): 223-54. Hansson, Bengt, ‘An Analysis of Some Deontic Logics’ in Dagfinn Føllesdal and Risto Hilpinen, eds, Deontic Logic: Introductory and Systematic Readings, Dordrecht, Reidel, 1971.



Haslanger, Sally, ‘Persistence, Change, and Explanation’, Philosophical Studies, 56 (1989):1-28. Hintikka, Jaakko, ‘Some Main Problems of Deontic Logic’ in Dagfinn Føllesdal and Risto Hilpinen, eds, Deontic Logic: Introductory and Systematic Readings, Dordrecht, Reidel, 1971. Honderich, Ted, ‘The Argument for Anomalous Monism’, Analysis, 42 (1982):59-64. Horgan, Terence, ‘Actions, Reasons, and the Explanatory Role of Content’ in Brian McLaughlin, ed., Dretske and His Critics, Oxford, Basil Blackwell, 1991, pp. 73-101. ——‘Jackson on Physical Information and Qualia’, Philosophical Quarterly, 34 (1984):147-52. ——‘Supervenience and Microphysics’, Pacific Philosophical Quarterly, 63 (1982):29-43. Hornsby, Jennifer, ‘Which Physical Events are Mental Events?’, Proceedings of the Aristotelian Society, 81 (1980):73-92. ——Actions, London, Routledge & Kegan Paul, 1980. Horwich, Paul, Probability and Evidence, Cambridge, Cambridge University Press, 1982. Humberstone, I.L., ‘Two Sorts of “Oughts”’, Analysis, 32 (1971):8-11. ——‘The Background of Circumstances’, Pacific Philosophical Quarterly, 64 (1983):19-34. Hume, David, A Treatise of Human Nature, Oxford, Clarendon Press, 1888.

j Jackson, Frank, ‘Reference and Description Revisited’, Philosophical Perspectives, vol. 12, forthcoming.

-274-



——From Metaphysics to Ethics: A Defence of Conceptual Analysis, Oxford, Oxford University Press, 1998. ——‘II—Naturalism and the Fate of the M-Worlds’, Proceedings of the Aristotelian Society, supp. vol. 71 (1997):269-82. ——‘Mental Properties, Essentialism and Causation’, Proceedings of the Aristotelian Society, 95 (1995):253-68. ——critical notice of Susan Hurley, Natural Reasons, Australasian Journal of Philosophy, 70 (1992):475-87. ——‘Understanding the Logic of Obligation’, Proceedings of the Aristotelian Society, supp. vol. 62 (1988):255-70. ——‘Group Morality’ in Philip Pettit et al., eds, Metaphysics and Morality, Oxford, Basil Blackwell, 1987, pp. 91-110. ——Conditionals, Oxford, Basil Blackwell, 1987. ——‘A Probabilistic Approach to Moral Responsibility’ in R. Barcan Marcus, et al., eds, Proceedings of the 7th International Congress of Logic, Methodology, and Philosophy of Science, Amsterdam, Elsevier Science Publishers, 1986, pp. 351-66. ——‘Two Theories of Indicative Conditionals: Reply to Brian Ellis’, Australasian Journal of Philosophy, 62 (1984):67-76. ——‘Epiphenomenal Qualia’, Philosophical Quarterly, 32 (1982):127-36. ——‘Conditionals and Possibilia’, Proceedings of the Aristotelian Society, 81 (1980/81):126-37. ——‘A Causal Theory of Counterfactuals’, Australasian Journal of Philosophy, 55 (1977):3-21. ——review of Keith Campbell, Body and Mind, Australasian Journal of Philosophy, 50 (1972):77-80. Jackson, Frank and Robert Pargetter ‘A Question about Rest and Motion’, Philosophical Studies, 52 (1987):21-6. ——‘Oughts, Options, and Actualism’, Philosophical Review, 95 (1986):233-55. ——‘Causal Origin and Evidence’, Theoria, 51(1985):65-76. Jackson, Frank and Philip Pettit ‘Moral Functionalism and Moral Motivation’, Philosophical Quarterly, 45 (1995):20-40. ——‘Folk Belief and Commonplace Belief’, Mind and Language, 8 (1993):298-305. ——‘Some Content is Narrow’ in John Heil and Al Mele, eds, Mental Causation, Oxford, Clarendon Press, 1993, pp. 259-82. ——‘Causation in the Philosophy of Mind’, Philosophy and Phenomenological Research, 50 (supplement, Fall 1990):195-214. ——‘Functionalism and Broad Content’, Mind, 97 (1988):381-00. Jackson, Frank, Robert Pargetter and Elizabeth Prior, ‘Functionalism and Type-Type Identity Theories’, Philosophical Studies, 42 (1982): 209-25. Jeffrey, Richard C., Formal Logic, New York, McGraw-Hill, 1967.



-275-

——‘Goodman’s Query’, Journal of Philosophy, 63 (1966):281-8. ——The Logic of Decision, New York, McGraw-Hill, 1965. ——‘If’, Journal of Philosophy, 61 (1964):702-3. Johnson, W.E., Logic, part I, New York, Dover, 1964. Johnston, Mark, ‘Is There a Problem about Persistence?’, Proceedings of the Aristotelian Society, supp. vol. 61 (1987):107-35.

k Kagan, Shelly, The Limits of Morality, Oxford, Clarendon Press, 1989. Kelley, M., ‘Predicates and Projectibility’, Canadian Journal of Philosophy, 1 (1971):189-206. Keynes, J.M., Treatise on Probability, London, Macmillan, 1921. Kim, Jaegwon, ‘Epiphenomenal and Supervenient Causation’, reprinted in Supervenience and Mind, Cambridge, Cambridge University Press, 1993. Kirk, Robert, ‘From Physical Explicability to Full-Blooded Materialism’, Philosophical Quarterly, 29 (1979):229-37. ——‘Sentience and Behaviour’, Mind, 83 (1974):43-60. Kripke, Saul, Naming and Necessity, Oxford, Basil Blackwell, 1980. Kyburg, Henry, Probability and Inductive Logic, Toronto, Macmillan, 1970.

l Land, Edwin H., ‘Experiments in Color Vision’, Scientific American, 200 (5 May 1959):84-99. Leblanc, H., ‘That Positive Instances Are No Help’, Journal of Philosophy, 60 (1963):452-62. Leiter, Brian and Alex Miller, ‘Mind Doesn’t Matter Yet’, Australasian Journal of Philosophy, 72 (1994):220-32. Lewis, David, ‘Reduction of Mind’ in Samuel Guttenplan, ed., A Companion to Philosophy of Mind, Oxford, Basil Blackwell, 1994, pp. 412-31. ——‘Rearrangement of Particles: A Reply to Lowe’, Analysis, 48 (1988):65-72. ——On the Plurality of Worlds, Oxford, Basil Blackwell, 1986. ——‘How to Define Theoretical Terms’ in Philosophical Papers, vol. I, New York, Oxford University Press, 1983, pp. 78-95. ——‘New Work for a Theory of Universals’, Australasian Journal of Philosophy , 61 (1983):343-77. ——‘Postscript to “Mad Pain and Martian Pain”’, Philosophical Papers, vol. I, New York, Oxford, 1983. ——postscript B to ‘Survival and Identity’ in Philosophical Papers, vol. I, New York, Oxford University Press, 1983.



——‘Causal Decision Theory’, Australasian Journal of Philosophy, 59 (1981):5-30. Lewis, David, ‘Attitudes De Dicto and De Se’ Philosophical Review, 88 (1979):513-43. ——‘Probabilities of Conditionals and Conditional Probabilities’, Philosophical Review, 75 (1976):297-315.

-276-

——‘Semantic Analyses for Dyadic Deontic Logic’ in Søren Stenlund, ed., Logical Theory and Semantic Analysis, Dordrecht, Reidel, 1974. ——Counterfactuals, Oxford, Basil Blackwell, 1973. ——Convention, Cambridge, Mass., Harvard University Press, 1969. Locke, Don, ‘Zombies, Schizophrenics and Purely Physical Objects’, Mind, 85 (1976):97-9. ——‘Reasons, Wants and Causes’, American Philosophical Quarterly, 2 (1974):169-79. Lockwood, Michael, Mind, Brain and the Quantum, Oxford, Basil Blackwell, 1989. Lowe, E.J., ‘Jackson on Classifying Conditionals’, Analysis, 51 (1991): 126-30. Lycan, W.G., ‘A New Lilliputian Argument against Machine Functionalism’, Philosophical Studies, 35 (1979):279-87.

m MacDonald, Graham, ‘The Nature of Naturalism’, Proceedings of the Aristotelian Society, supp. vol. 66 (1992):225-44. Mackie, J.L., Truth, Probability and Paradox, Oxford, Clarendon Press, 1973. McDowell, John, ‘Singular Thought and the Extent of Inner Space’ in Philip Pettit and John McDowell, eds, Subject, Thought, and Context, Oxford, Clarendon Press, 1986, pp. 137-68. ——‘Functionalism and Anomalous Monism’ in Ernest LePore and Brian McLaughlin, eds, Actions and Events, Oxford, Basil Blackwell, 1985, pp. 387-98. ——‘Criteria, Defeasibility and Knowledge’, Proceedings of the British Academy, 68 (1982):455-79. McGinn, Colin, Mental Content, Oxford, Basil Blackwell, 1989. ——‘The Structure of Content’ in Andrew Woodfield, ed., Thought and Object, Oxford, Clarendon Press, 1982. McLaughlin, Brian, ‘On Davidson’s Response to the Charge of Epiphenomenalism’ in John Heil and Al Mele, eds, Mental Causation, Oxford, Clarendon Press, 1993.



Meehl, Paul E., ‘The Compleat Autocerebroscopist’ in Paul Feyerabend and Grover Maxwell, eds, Mind, Matter, and Method, Minneapolis, University of Minnesota Press, 1966. Mellor, D.H., The Facts of Causation, London, Routledge, 1991. ——Real Time, Cambridge, Cambridge University Press, 1981. ——‘Materialism and Phenomenal Qualities’, Proceedings of the Aristotelian Society, supp. vol. 47 (1973):107-19. Menzies, Peter, ‘Against Causal Reductionism’, Mind, 97 (1988): 551-74. Mott, Peter, ‘On Chisholm’s Paradox’, Journal of Philosophical Logic, 2 (1973):197-211.

n Nagel, Thomas, The View from Nowhere, Oxford, Oxford University Press, 1986.

-277-

——‘What is it Like to be a Bat?’, Philosophical Review, 83 (1974): 436-50. ——The Possibility of Altruism, Oxford, Oxford University Press, 1970. Nemirow, Lawrence, review of Thomas Nagel, Mortal Questions, Philosophical Review, 89 (1980):473-7.

o Oliver, Alex, ‘The Metaphysics of Properties’, Mind, 105 (1996):1-80.

p Parfit, Derek, Reasons and Persons, New York,Oxford University Press, 1984. Pargetter, Robert, ‘Laws and Modal Realism’, Philosophical Studies, 46 (1984):335-47. Peacocke, Christopher, ‘Externalism and Explanation’, Proceedings of the Aristotelian Society, 93 (1993):203-30. Pears, David, ‘How Easy is Akrasia’, Philosophia, 2 (1982):33-50. Pendelbury, Michael, ‘The Projection Strategy and the Truth Conditions of Conditional Statements’, Mind, 98 (1989):179-205. Pessin, Andrew, ed., The Twin Earth Chronicles, London, Paragon House, 1995. Pettit, Philip and Geoffrey Brennan, ‘Restrictive Consequentialism’, Australasian Journal of Philosophy, 64 (1986):438-55. Pettit, Philip and Robert Goodin, ‘The Possibility of Special Duties’, Canadian Journal of Philosophy, 16 (1986):651-76. Piaget, Jean, ‘The Child’s Conception of Physical Causality’, reprinted in Howard E. Gruber and J. Jacques Vonedue, eds, The Essential Piaget, London, Routledge & Kegan Paul, 1977. Pollock, John, Subjunctive Reasoning, Dordrecht, Reidel, 1976.



Putnam, Hilary, ‘The Meaning of “Meaning”’ in Mind, Language and Reality, Cambridge, Cambridge University Press, 1975, pp. 215-71.

q Quine, W.V., ‘Natural Kinds’ in Ontological Relativity and Other Essays, New York, Columbia University Press, 1969, pp. 114-38. ——‘Identity, Ostension and Hypostasis’ in From a Logical Point of View, New York, Harper and Row, 1963. ——Word and Object, Cambridge, Mass., MIT Press, 1960.

r Railton, Peter, ‘Alienation, Consequentialism, and the Demands of Morality’, Philosophy and Public Affairs, 13 (1984), reprinted in S. Scheffler, ed., Consequentialism and its Critics, New York, Oxford University Press, 1988, pp. 93-133. Ramsey, F.P., Foundations of Mathematics, London, Routledge & Kegan Paul, 1931. Ramsey, William, Stephen Stich and Joseph Garon, ‘Connectionism, Eliminativism and the Future of Folk Psychology’, Philosophical Perspectives, 4 (1990):499-533. Robinson, Howard, Matter and Sense, New York, Cambridge University Press, 1982. Ryle, Gilbert, The Concept of Mind, London, Hutchinson, 1949.

s Salmon, W.C., ‘On Vindicating Induction’ in H. Kyburg and E. Nagel,

-278-

eds, Induction: Some Current Issues, Middletown, Conn., Wesleyan, 1963. Scheffler, S., ed., Consequentialism and its Critics, New York, Oxford University Press, 1988. Schiffer, Stephen, Remnants of Meaning, Cambridge, Mass., MIT Press, 1987. Schotch, Peter K. and Raymond E. Jennings, ‘Non-Kripkean Deontic Logic’ in Risto Hilpinen, ed., New Studies in Deontic Logic, Dordrecht, Reidel, 1981. Segal, Gabriel and Elliott Sober, ‘The Causal Efficacy of Content’, Philosophical Studies, 63 (1991):1-30. Shoemaker, Sydney, ‘Functionalism and Qualia’, Philosophical Studies, 27 (1975):291-315. Sidgwick, Henry, The Methods of Ethics, Chicago,University of Chicago Press, 7th edition, 1907. Skyrms, Brian, Causal Necessity, New Haven,Yale University Press, 1980.



——‘Physical Laws and Philosophical Reduction’ in G. Maxwell and R.M. Anderson Jr, eds, Induction, Probability, and Confirmation, Minnesota Studies in Philosophy of Science, vol. 6, Minneapolis, University of Minnesota Press, 1975. ——Choice and Chance, Belmont, Calif., Dickenson, 1966. Smart, J.J.C., ‘An Outline of a System of Utilitarian Ethics’ in J.J.C. Smart and Bernard Williams, Utilitarianism: For and Against, Cambridge, Cambridge University Press, 1973. ——Philosophy and Scientific Realism, London, Routledge & Kegan Paul , 1963. Stalnaker, Robert, Inquiry, Cambridge, Mass., MIT Press, 1984. ——‘A Defense of Conditional Excluded Middle’ in W.L. Harper et al, eds, Ifs, Dordrecht, Reidel, 1981, pp. 87-104. ——‘Assertion’ in P. Cole, ed., Syntax and Semantics, vol. 9, New York, Academic Press, 1978, pp. 315-32. ——‘Indicative Conditionals’ in A. Kasher, ed., Language in Focus, Dordrecht, Reidel, 1976. ——‘Possible Worlds’, Nous, 10 (1976):65-75. ——‘Probability and Conditionals’, Philosophy of Science, 37 (1970): 64-80. ——‘A Theory of Conditionals’ in N. Rescher, ed., Studies in Logical Theory, Oxford, Basil Blackwell, 1968. Sterelny, Kim, The Representational Theory of Mind, Oxford, Basil Blackwell, 1990. Stevenson, Charles, L., ‘If-iculties,’ Philosophy of Science, 37 (1970): 27-9. Stich, Stephen, The Fragmentation of Reason, Cambridge, Mass., MIT. Press, 1990.

-279-

Stacker, Michael, ‘The Schizophrenia of Modern Ethical Theories’, Journal of Philosophy, 73 (1976):455-66.

t Thomason, Richmond H., ‘Deontic Logic and the Role of Freedom in Moral Deliberation’ in Risto Hilpinen, ed., New Studies in Deontic Logic, Dordrecht, Reidel, 1981. ——‘Deontic Logic as Founded on Tense Logic’ in Risto Hilpinen, ed., New Studies in Deontic Logic, Dordrecht, Reidel, 1981. Tich´y, Pavel, ‘Kripke on Necessity A Posteriori’, Philosophical Studies, 43, 1983:225-41. Tooley, Michael, ‘The Nature of Laws’, Canadian Journal of Philosophy, 7 (1977):667-98.



Tye, Michael, Ten Problems of Consciousness, Cambridge, Mass., MIT Press, 1995. ——The Imagery Debate, Cambridge, Mass., MIT Press, 1992.

v Van Fraassen, Bas, ‘Values and the Heart’s Command’, Journal of Philosophy, 70 (1973):5-19. ——‘The Logic of Conditional Obligation’, Journal of Philosophical Logic, 1 (1972):417-38. Von Wright, G.H., ‘On Conditionals’ in Logical Studies, London, Routledge & Kegan Paul, 1957.

w Watson, Gary, ‘Skepticism about Weakness of Will’, Philosophical Review, 86 (1977):316-39. Wells, H.G., The Country of the Blind and Other Stories, London, n.d. Wiggins, David, Sameness and Substance, Oxford, Basil Blackwell, 1980. Williams, Bernard, ‘A Critique of Utilitarianism’ in J.J.C. Smart and Bernard Williams, Utilitarianism: For and Against, Cambridge, Cambridge University Press, 1973. Williams, D.C., Principles of Empirical Realism, Springfield, Ill., Thomas, 1966.

y Yablo, Stephen, ‘Mental Causation’, Philosophical Review, 101 (1992): 245-80.

-280-

INDEX a

Achinstein, Peter 269 Adam’s thesis 15 ff, 51 ff Adams, Ernest W. 15 , 24 , 25 , 41 , 42 agglomeration 198 ff, 211 ff Anderson, A.R. 24 Armstrong, D.M. 87 , 108 , 110 , 127 , 134 , 140 , 150 , 151 , 153 , 154 , 173 , 174 assertability 3 ff, 51 ff Ayer, A.J. 23

b



Barker, Stephen 251 , 269 behaviour 93 ff belief-desire psychology 81 Belnap, N.D. 24 Bennett, Jonathan 29 , 41 , 42 , 54 Bentham, Jeremy 243 Bigelow, John 152 , 153 Blackburn, Simon 174 , 269 Block, Ned 110 , 124 , 128 Braddon-Mitchell, David 111 , 128 , 129 Brennan, Geoffrey 243 , 245 , 246 Brink, David O. 231 , 244 Burge, Tyler 108 , 111

c Campbell, Keith 68 , 69 Carnap, Rudolf 269 Chellas, Brian F. 219 Child, William 107 , 108 , 129 Chisholm, R.M. 151 Churchland, Paul M. 70 , 72 ff, 107 Clark, Michael 23 , 25 Cohen, L. Jonathan 23 , 26 conceptual analysis 134 ff, 145 ff, ch. 11 conditional non-contradiction ch. 2, ch. 3 conditional obligation 198 , § 13.6 conditionalisation 6 , 17 contraposition 16 , 17 , 198 conventional implicature 10 Cornman, J.W. 67 Crane, Tim 89 , 108 , 110 culpable ignorance 223-4

d Darwin, Charles 64 , 65 Davidson, Donald 83 ff, 91 , 95 , 107 , 108 , 109 , 110 , 179 , 184 , 187 , 189 , 196 Davies, Martin K. 175 decision theory 184 ff, 196 , 222 ff Dennett, Daniel 68 , 94 , 109



Devitt, Michael 111 , 112 , 173 dispositions and causation 98 ff, 120 ff Dretske, Fred 110 , 111 , 134 , 150 Dudman, V.H. 49 , 50 Dummett, Michael 24 dyadic deontic logic 199 , § 13.4 dynamics of value § 12.3

e Edgington, Dorothy 41 Eells, Ellery 243 Ellis, Brian 25 , 28 , 41 embedded conditionals 26 entailment § 11.5 , 172 entry by entailment 155 , 157 ff, § 11.4

-281-

epiphenomenalism 63 ff, 98-9 , 124 equivalence thesis 3 ff, 22-3 , 26 see also supplemented equivalence thesis expected moral utility 223 ff, 242 externalism and causation 104 ff

f Feldman, Fred 244 Feymann, Richard 100 , 101 , 121 , 122 Field, Hartry 128 Fodor, Jerry 104 , 107 , 110 , 111 , 112 , 128 folk theory 139-40 , 144 ff, 150 , 169-71 Føllesdal, Dagfinn 217 Forrest, Peter 150 , 174 free action 145 , 146 , 147 , 164 Frege 10 functionalism 96 ff, 133

g Garron, Joseph 126 , 129 Geach, P.T. 138 , 139 , 151



Gibbard, Allan 29 , 31 , 41 , 42 , 43 Godwin, William 246 Goldman, Alvin 109 Goldman, Holly S. 218 , 219 Goodin, Robert 245 Goodman, Nelson 249 , 250 , 254 , 257 , 267 , 268 , 269 Greenspan, P.S. 218 , 219 Grice, H.P. 23 grue, various definitions of § 15.2 Gruzalski, Bart 244

h Haldane, John 107 Harper, William L. 42 Haslanger, Sally 152 Hilpinen, Risto 217 Hintikka, Jaakko 217 Holton, Richard 173 Honderich, Ted 107 Hopkins, James 84 Horgan, Terence 75 , 111 , 173 Hornsby, Jennifer 109 Horwich, Paul 243 Humberstone, I.L. 26 , 42 , 153 , 174 , 175 , 218 , 219 Hume, David 64 , 97 , 98 , 110 , 134 , 183 hypothetical syllogism 16 , 17 , 198

i identity through time 135 , 137 ff indicative conditionals ch. 1, ch. 2, ch. 3, 51 ff individualism about belief 113-14 , 126 intuitions about possible cases 133 ff, 164 ff

j Jackson, Frank 24 , 41 , 54 , 79 , 102 , 106 , 107 , 109 , 110 , 111 , 128 , 129 , 150 , 152 , 173 , 219 , 243 , 245 , 246 Jeffrey, Richard C. 23 , 24 , 25 , 196 , 263 , 269 Jennings, Raymond E. 219 Johnson, W.E. 24 Johnston, Mark 151

k



Kagan, Shelly 246 Kelley, Michael 269 Keynes, J.M. 24 Kim, Jaegwon 111 Kirk, Robert 68 knowledge argument 58 ff, ch. 6, 77 ff Kripke, Saul 167 , 174 Kyburg, Henry 268

l Land, Edwin 68 Langton, Rae 174 language of thought 113 , 117 lawhood, various accounts of 133-5 , 146 ff, 150 , 152 Leblanc, H. 269 Leiter, Brian 110 Lewis, David 8 , 15 , 23 , 24 , 25 , 27 , 41 , 42 , 54 , 68 , 73 , 75 , 88 , 100 , 108 , 111 , 116 , 121 , 127 , 129 , 135 , 140 , 145 , 151 , 153 , 173 , 174 , 175 , 196 , 206 , 218 , 219 , 246 Locke, Don 68 , 196 Locke, John 52 Lockwood, Michael 108 Lowe, E.J. 42 , 45 , 46 , 47 , 48 , 50 Lycan, W.G. 68

m MacDonald, Graham 109 , 128

-282-



McDowell, John 107 , 108 McGinn, Colin 92 , 98 , 108 , 110 , 111 Mackie, J.L. 23 , 25 material conditional 4 ff, 23-4 , 39 Meehl, Paul E. 69 Mellor, D.H. 67 , 89 , 108 , 151 mental causation ch. 8, ch. 9 Menzies, Peter 42 , 121 , 128 Michael, Michaelis 174 Miller, Alex 110 , 153 mind-brain identity views 90 ff, 124 ff modal argument (for qualia) 61 ff modus ponens 13 modus tollens 14 Moorean facts 135 , 147 ff motion 76 ff, 134-5 , 144 , 148-9 Mott, Peter 219 Mulligan, Kevin 153

n Nagel, Thomas 62 , 63 , 75 , 196 , 227 , 244 narrow content 104 , 106 naturalism 154 ff Neander, Karen 129 necessary a posteriori 87 , 166 ff, § 11.6 Nemirow, Laurence 68 , 73 , 75

o O’Leary Hawthorne, John 174 Oliver, Alex 108 ought to be v. ought to do 199 , 204 , § 13.7

p paradoxes of material implication 4 ff, 16 Parfitt, Derek 245 , 246 Pargetter, Robert 26 , 69 , 75 , 79 , 109 , 152 , 219 , 246 , 269 Peacocke, Christopher 109 Pears, David 196 Pendelbury, Michael 41 , 42 persisting, perduring, enduring 137 ff



personal identity 76 , 134-5 , 144 , 145 , 149 Pettit, Philip 102 , 106 , 107 , 109 , 110 , 111 , 128 , 152 , 243 , 245 , 246 physical properties 57 , 70 , 81 ,158ff physicalism ch. 5, ch. 6, 158 ff, § 11.4 Piaget, Jean 69 placement problem 155 ff Pollock, John 25 possible world semantics for conditionals 26 , 27 , 30-31 , 198 Prior, Elizabeth 109 , 196 problem of temporary intrinsics 139 , 141-3 program explanation 102 , 103 projectible-nonprojectible distinction 250 ff Putnam, Hilary 111 , 167 , 174

q qualia ch. 5, ch. 6, ch. 7 Quine, W.V. 129 , 135 , 140 , 144 , 151 , 167 , 269

r radical holism about belief 116 ff, 122 ff, 126-7 Railton, Peter 174 , 225-6 , 231 , 237 , 244 , 245 , 246 Ramsey, F.P. 24 , 114 Ramsey, William 126 , 129 relativity of obligation 200 ff representation: sentences, maps, holograms 114 ff Rey, Georges 129 Robinson, Denis 153 Robinson, Howard 75 robustness 6 ff, § 1.3 , 24 Ryle, Gilbert 110

s Salmon, W. 269 Scanlon, Thomas 245 scepticism about weakness of will 183 Schiffer, Stephen 107 , 108 , 152 Schotch, Peter K. 219 Segal, Gabriel 111 serious metaphysics 155 ff, 168 Shoemaker, Sydney 68 Sidgwick, Henry 224 , 243 , 244 , 246



Singer, Peter 196 , 246 singular thought 161-3 Skyrms, Brian 24 , 196 , 269

-283-

Smart, J.J.C. 135 , 140 , 151 , 221 , 243 , 245 Smith, Michael 246 Sober, Elliot 111 Stalnaker, Robert C. 26 , 27 , 42 , 127 , 128 , 173 , 175 , 206 , 218 Sterelny, Kim 155 , 166 , 173 , 246 Stevenson, Charles L. 23 Stich, Stephen 126 , 128 , 129 Stocker, Michael 229 , 245 straight rule 249 ff strengthening the antecedent 16 , 198 , 207 strong autonomy (of psychology) 82 ff subjunctive conditionals 17 ff, 27 ff, 37 ff, 198 supervenience of the psychological 159 ff supplemented equivalence thesis 13-14 , § 1.4 , § 1.5 , 23 systematicity and productivity 115

t theory of evolution 64 ff Thomason, Richmond H. 218 , 219 three dimensionalism v. four dimensionalism 136-46 Tich y, Pavel 175 Tooley, Michael 134 , 150 truth conditions 51 ff Twin Earth 165 two-dimensional modal logic 170 , 172 Tye, Michael 109 , 111

u utilitarianism 186 , 221 , 240

v Van Fraassen, Bas 218 , 219 Von Thun, Manfred 24



Von Wright, G.H. 23 w

Watson, Gary 196 Wells, H.G. 68 West, Caroline 153 Wiggins, David 139 , 144 , 151 Williams, Bernard 220 , 236-7 , 243 , 244 , 245 Williams, B.C. 173

y Yablo, Stephen 174

-284-


-285-


-286-


int1 jackson - mind method and conditionals

Documents

conditionals logic

classifying conditionals

indicative conditionals

v file

publication data file

new york

british library cataloguing

selected essaysfrank