artificial life and philosophy

6

Click here to load reader

Upload: alvaro-moreno

Post on 15-Jan-2017

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Artificial Life and Philosophy

Leonardo

Artificial Life and PhilosophyAuthor(s): Alvaro MorenoSource: Leonardo, Vol. 35, No. 4 (2002), pp. 401-405Published by: The MIT PressStable URL: http://www.jstor.org/stable/1577402 .

Accessed: 12/06/2014 20:55

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

The MIT Press and Leonardo are collaborating with JSTOR to digitize, preserve and extend access toLeonardo.

http://www.jstor.org

This content downloaded from 188.72.126.181 on Thu, 12 Jun 2014 20:55:43 PMAll use subject to JSTOR Terms and Conditions

Page 2: Artificial Life and Philosophy

GENERAL ARTICLE

Artificial Life and Philosophy

Alvaro Moreno

A rtificial Life (A-Life) has been defined as the study of all possible life through its artificial production. This definition raises three important issues: First is the very sense of the term "study," for A-Life is not only an epistemic or a purely theoretical process but also a technical activity, because in A-Life the objects of study are literally created through tech- nological action. Second, the actual meaning of "lifelike sys- tems" (or systems that show "lifelike" behavior) becomes much more complex (and controversial) than in traditional biology, since this concept has become understood in a much wider sense than that of empirically real biological systems. Thus, as C. Emmeche has pointed out, A-Life is founded as a modal discipline, establishing as its own objective the study of life "as it could be" and not simply "as we know it" (even if we include here any extraterrestrial forms of life that might be discovered in the future) [1]. Last, but not least, the idea of "artificiality" should also be subject to examination. In addi- tion to its generic sense of human construction, the term ar- tificial has a double-sided peculiarity in the context of A-Life. On the one hand, it has a paradoxical meaning, resulting from the idea that such humanly constructed systems should be capable-like natural living beings-of exhibiting cre- ativity. So-called emergent behavior, capabilities, morphology, etc., sought by A-Life designers must be understood precisely as a form of indirect human creation [2], what Langton has described as "getting the humans out of the loop," designing artifacts able to perform nontrivial, unpredictable activities, so that the machine itself could appear as if it were endowed with creativity. That is to say, one of the essential features of A-Life is that the artificially created system should display some type of agency, which allows us to speak (without falling into contradiction, although somewhat paradoxically) of au- tonomy in such cases. And on the other hand, artificial has the peculiar meaning (not exclusive but prevailing), as in ar- tificial intelligence (AI), of virtual "construction" as opposed to physical realization [3].

The complex combination of all these elements has pro- vided A-Life with an identity as a separate discipline, distinct not only from traditional biology but also from the whole set of traditional empirical sciences. The idea that a living entity subject to study is to be generated by a human agent-and, furthermore, in a computational universe-is truly suggestive; nevertheless, it also brings up many novel issues and chal- lenges.

Alvaro Moreno (educator), Department of Logic and Philosophy of Science, University of the Basque Country, Post Box 1249, 20080 San Sebastian, Donostia, Spain. E-mail: <[email protected]>.

Originally presented at the Seventh International Conference on Artificial Life (Alife VII), 1-6 August 2000, Portland, OR, U.S.A. First published in M.A. Bedau,J.S. McCaskill, N.H. Packard and St. Rasmussen, eds., Artificial Life VII: Proceedings of the Seventh International Conference (Cambridge, MA: MIT Press, 2000). Reprinted by permission.

ABSTRACT

Artificial Life is developing into a new type of discipline, based on computational construction as its main tool for exploring and producing a science of life "as it could be." In this area of research, the generation of complex virtual systems, in place of the traditional empirical domain, has become the actual object of theory. This entails a

Although A-Life is not necessarily to profound change in the tradi-

be understood as a computational tional relationship between ontological, epistemological and

science, its major research activity ontological, epistemological an methodological levels of analy- takes place in the computational do- is, which forces to recon- main. As mentioned above, the sider the differences apparently most prevalent meaning of the term firmly established between

artificial in A-Life research specifi- science and philosophy. Even if the frontiers between these two

cally refers to the generation of vir- kinds of knowledge do not tual systems in the computational completely disappear, new, universe. As a matter of fact, this fea- dynamic, complex, technologi- ture is not original to A-Life. For the cally mediated interactions are

past 50 years there has been a strong being developed between them.

tradition in the general study of dif- ferent types of systems that makes a radical distinction between the informational-organizational aspects of a system and its energetic-material ones. This dis- tinction already existed at the core of such disciplines as cy- bernetics, computer and systems sciences and AI. These disciplines share with A-Life a common approach based on the idea that the material or energetic aspects of an organization do not affect its logical essence. Hence, even the admitted ne- cessity of including a good deal of ancillary machinery for the actual implementation of any material system [4] would not be significant for modeling the "actual organization" of that system. That is why researchers' interest and activity is mainly focused on the study of virtual systems generated in a compu- tational environment. Even most research in the domain of physical realizations, such as the design of autonomous robots, is largely influenced by the aforementioned philosophy, since the construction of the physical "body" of the robot is still con- sidered independently of its behavior.

Fig. 1. Building up an A-Life "model." (? Alvaro Moreno)

LEONARDO, Vol. 35, No. 4, pp. 401-405, 2002 401

A-LIFE AS A COMPUTATIONAL RESEARCH PROJECT

? MIT Press

This content downloaded from 188.72.126.181 on Thu, 12 Jun 2014 20:55:43 PMAll use subject to JSTOR Terms and Conditions

Page 3: Artificial Life and Philosophy

The roots of the computational ver- sions of both AI and A-Life lie in the idea of functionalism. This position assumes and asserts that the specific materiality that sustains a certain capability (mental, biological or otherwise) is not relevant [5]. Accordingly, functionalists claim that biological phenomenology is the exclu- sive result of an organizational arrange- ment, rather than of a particular material implementation of it. In fact, the ques- tion of whether those organizational arrangements are sustained by carbon or silicon molecules or by patterns of elec- trons in a computer is considered com- pletely irrelevant. This is the reason why Langton [6] has defended the idea of universalizing classical biology by ab- stracting the materiality of biological phenomena when studying such phe- nomena: he assumes this study can take place in a purely formal organizational domain. Accordingly, the huge potential for exploration of virtual organizations in the computational realm becomes for A-Life researchers a way of "experiment- ing" with formal "lifelike" entities in for- mal environments, which are then empirically interpreted.

This new strategy is highly promising with regard to the problem of the uni- versalization of biology. However, it also poses new and intriguing questions, be- cause it is the consequence of pure ab- stractions of processes taking place in empirical environments. In particular, I want to stress three issues: The first, which is basically epistemological (al- though it has ontological implications, as well), involves the confusion between ob-

ject and model; the second one, of a methodological nature, is the problem of evaluating hypotheses (and, therefore, research programs); and the third con- cerns the interrelation between science, philosophy and technology.

THE CONFUSION BETWEEN OBJECT AND MODEL

In science, a model is essentially a sym- bolic construction whose structure is in- tended to refer to a given empirical process. There is a wide variety of mod- els, from the strictly mathematical to the purely conceptual, but in essence, they all try to capture or reproduce some generic feature of a given empirical do- main. In biology (as in all the other tra- ditional empirical sciences), models are elaborated to operatively represent em- pirical living systems. These empirical sys- tems thus constitute the objects of reference for the models. Regardless of

whether these models might allow com- putational simulations or not, the mod- els involve the pre-existence of a reference system, whose behavior is to be totally or partially reproduced.

The case of A-Life, however, is radically different. A-Life researchers attempt to create not only a symbolic model of a liv- ing system, but also a symbolic living ob-

ject. Accordingly, these computational "models" are elaborated without direct and precise reference to empirical bio- logical reality. C. Emmeche regards them as "second-order simulacra, that is, copies of the copies themselves" [7], generated not as abstractions of empirical biologi- cal systems (as in the case of concepts and theories of biology, which would be "first- order simulacra") but of the theories themselves. Their main goal is to allow a new means of "computational experi- mentation" to enable us to "discover" the universal principles of living systems.

Thus, proponents of "strong" A-Life explicitly claim that computational sim- ulations of living systems may really come to be living systems. Whereas proponents of "weak" A-Life consider models to rep- resent certain aspects of living phenom- ena, strong A-Lifers claim that the phenomenology of the computational environment is life in the strict sense. The model or simulacrum is considered a literal realization-that is to say, an ob- ject whose phenomenology would make it equivalent to any natural system of the corresponding empirical domain.

Nevertheless, among those authors who regard these systems as true realiza- tions, the use of the term "model" is quite widespread (even despite the awareness of its difference from the classical mean- ing), probably because A-Life research and experimentation generate subse- quent virtual, tentative structures for methodological reasons. As a matter of fact, then, depending on the circum- stances, a computational system is inter- preted sometimes as an object (ultimately, what is interpreted as an object is the model associated with the physical struc- ture of the machine that sustains its exe- cution [8]), at other times as a proper model or even as a tool or methodologi- cal technique mediating between the "true object" (natural biological systems) and the "true theory" (theories of em- pirical life sciences) [9].

In short, the solution to these dilem- mas involves determining the ontologi- cal status of such artificial systems. Are they real or virtual? What is the empiri- cal referent for a computational simu- lacrum? B. Smith, for instance, claims

that the computational world does not constitute a proper matter for study; it should rather be considered a complex social practice that involves the design, construction, maintenance and use of in- tentional artifacts [10]. Whatever answer may be given to these questions, it will certainly have methodological implica- tions, since all these problems are closely related to that of establishing the criteria for evaluating the hypotheses that the very design of such systems attempts to test.

THE PROBLEM OF "EMPIRICAL" EVALUATION

I shall, for simplicity, use the term "model" hereafter to designate those vir- tual systems created in computational media with the aim of enlarging our knowledge about the universal laws of life. Taking this into account, the funda- mental problem in the methodology of A-Life research programs is that their means of evaluating models are not em- pirically conclusive, for, by definition, the hypothetical empirical references of these models belong to a domain broader than the already known and even the effectively existent. Despite the fact that "virtual experimentation" pro- vides some formal rigor to the method- ology (compared to the use of theories based on ordinary language), there re- mains a problem of global empirical in- terpretation, since, as I have mentioned, there is no clear referent [11 ]. This is one of the main differences between A-Life and classical scientific methodology, in which experiments and measurements, no matter how idealized, are always per- formed or stated within the framework of models that can be given a rigorous empirical interpretation.

Typically, an A-Life model is designed by taking as a starting point some basic principles from the general theories of a certain area of empirical biology. These principles are then introduced in the de- sign of a computer program in which only low-level instructions (local rules) are made explicit, so that new patterns (which play the role of new rules at higher levels) may appear. The charac- teristics of these patterns, however, are not known in advance by the designer. In this process, the emergence in the run- ning of the program in the computer of new structures (clearly distinguishable in spatial and/or temporal terms) is fun- damental. Then, according to the afore- mentioned theories, some of these new patterns become hierarchically and func-

402 Moreno, Artificial Life and Philosophy

This content downloaded from 188.72.126.181 on Thu, 12 Jun 2014 20:55:43 PMAll use subject to JSTOR Terms and Conditions

Page 4: Artificial Life and Philosophy

tionally organized and finally become subject to empirical interpretation.

The next step in this process is the comparison of this first result with what we know of the corresponding empirical domain. As a result, the methodology consists in a continuous revision of the values of the parameters (and even of some of the actual principles according to which the original design of the model was made), depending on the degree of coherence exhibited by the results in re- lation to the phenomenology of the em- pirical domain under study (see Fig. 1).

Therefore, the manifest problem is first that in A-Life computational systems, evaluation is not discernible from the in- terpretation of "emergent" patterns or processes generated in such systems, nor from the establishment of epistemologi- cal criteria for selecting the primitives that define the models. So there arises the initial problem of how to set up the main features of the model. Here, apart from the basic criterion of simplicity, the usual way of proceeding is to search for (sub-)models that, somehow, implicitly ensure that the assumed abstraction will be "valid," i.e. that the new information will be coherent with what happens in well-known biological systems. The lim- its on simplification are set so as to avoid results that appear trivial in the selected phenomenological domain [12]. The aim is thus double-sided: simplicity is pur- sued, provided that, at the same time, the model is able to generate sufficient com- plexity in the course of the simulation.

Thus, the design of a model ought to fulfill two basic requirements. On the one hand, it is important that emergent processes or structures, which may be in- terpreted as new data in the empirical do- main under investigation, are obtained. On the other hand (and in order to avoid the problem of how to determine when the empirical interpretation of these new structures and/or processes is objectively justified), it is necessary to design the computational experiment so that at least the expected type of results can be fore- cast, and that these are specific enough within each particular domain [13]. If only the first condition is met, the prob- lem becomes how to justify its interpre- tation; if only the second one is met, the risk is that the result will be trivial. Thus, depending on the results obtained by running the program, it will be deter- mined whether the model is valid or not, although other elements will play a role in the evaluation, as in the traditional sci- ences (for example, auxiliary theories, the adequacy of the computational tools

bfubdis- m M riul tools

Fig. 2. The relation between the empirical, theoretical and computational domains in A-Life. (? Alvaro Moreno)

employed or of the set-up values for the initial parameters, etc.).

Nevertheless, the problem of the eval- uation of models goes deeper, because, depending on the epistemological status given to computational systems, the ac- tual research program may vary quite radically. Should we try to design "plau- sible" models (that is, computational A-Life models that "resemble" the be- havior of the corresponding natural sys- tems) or, on the contrary, should the main goal lie in some other direction, such as in the search for models that ful- fill certain primarily formal criteria (for example, the capacity to generate com- putational complexity)? Finally, in other cases, the model may be designed to seek, instead of prediction, to learn how a given kind of biological system works. In such a case, higher degrees of complex- ity might be introduced at the micro- scopic level of the model's design [14].

The problem with a purely formalist conception of A-Life (i.e. that the task of A-Life models is mainly to "unpack" and make explicit the logical consequences of their starting premises) is that A-Life risks becoming a discipline closer to mathematics than to any empirical sci- ence. Several authors have formulated se- rious doubts about this approach. Maynard Smith, for instance, has called A-Life a "science without facts," referring to the problem of how to assess a set of computational models whose (potential) empirical references are imprecise and generic [15]. However, this criticism is too strong, because most A-Lifers, both in the design of their models and in the evaluation of their results, usually take into account the theories as well as the

behaviors present in the corresponding empirical domains. All the same, the problem of escaping mathematization might well put the actual autonomy of A-Life as a discipline in question [16].

The attempt to answer these questions takes us well beyond the strictly method- ological realm, as it requires determin- ing the actual nature of what a model is to discern (sometimes, rather than a par- ticular phenomenon in a particular em- pirical domain, the model addresses and, one hopes, elucidates certain computa- tional issues). At the same time, one must address the role that meta-theoretical conceptual elements play in the design of models, as well as the nature of the technical objects employed.

THE INTERRELATION BETWEEN

SCIENCE, PHILOSOPHY AND TECHNOLOGY

Given the nature of the problems it ad- dresses, A-Life might appear closer to philosophy than to empirical science and, in particular, closer to the philoso- phy of biology than to biology itself. Owing to the specificity of its methodol- ogy, A-Life has to face and analyze in an entirely new way several important philo- sophical issues in biology, such as the de- bates between reductionism and vitalism, the problem of emergence, and the matter-form relationship [17]. This is a result of the deep entanglement of A-Life methodology with the meta-theoretical (i.e. philosophical) problems of biology.

This peculiarity is not unique to A-Life, but is shared by other computational or artificial sciences devoted to the study of

Moreno, Artificial Life and Philosophy 403

?

itc

This content downloaded from 188.72.126.181 on Thu, 12 Jun 2014 20:55:43 PMAll use subject to JSTOR Terms and Conditions

Page 5: Artificial Life and Philosophy

empirical phenomena, such as AI. Re- garding this question, J. Casti [18] and D. Lane [19] hold that the construction of models in these "sciences of surrogate worlds" requires theories about particu- lar empirical domains, specifically, about the kinds of objects that are found in each of them and about how these ob- jects relate to one another, or the types of processes that create or destroy them or otherwise change their nature. The relevance of computational models to their corresponding empirical domains would be mediated by these theories, which would constrain the interpretation of the models. In turn, such models aid the analysis of these theories by provid- ing more powerful tools than ordinary language.

This idea poses problems, however. First, as post-positivist philosophers of science have broadly argued, the con- cepts on which models in traditional em- pirical sciences are constructed are also mediated by theories, in a sense similar to that identified by Lane. This is why the differences between these two types of models are not stated in the terms men- tioned by Lane, although it is quite true that the use of such virtual models does indeed involve, and rather directly, the aforementioned meta-theoretical prob- lems. The question to address is why this is so.

Second, these models are, in principle, purely formal, even though the formula- tion of their basic principles is inspired in a given empirical domain. As I mentioned above, the difficulties of interpreting and testing A-Life models in terms of current biological knowledge lie in the meta- theoretical implications implicit in the epistemological status of the entities and relations that constitute such models.

Nevertheless, my main criterion for evaluating the usefulness of these mod- els is their capacity to improve our un- derstanding of the theories of the empirical biological world. Apparently, this is a vicious circle.

According to Lane, these problems could be progressively resolved. On the one hand, the models of computational sciences can improve our theories about the corresponding empirical domains; on the other hand, these theories be- come essential for testing the relevance of specific surrogate worlds as useful models. In this way, a kind of "hermeneu- tic" circularity is established between models and theories [20], since the mod- els are used for the elucidation of theo- ries about the empirical world, and these are then used to interpret the models. A

highly significant feature of this idea is that the interaction does not merely occur between a philosophical or meta- scientific level and a scientific one. A new technological level (constituted by the computational tools, with their double- sided nature as software and hardware) also plays a crucial role in this hermeneu- tic process. This is because computers make possible the exploration and visu- alization of a given set of work premises through the creation of virtual worlds of indefinite complexity [21] (see Fig. 2).

This last feature is particularly charac- teristic of the A-Life research program. As D. Dennett [22] has pointed out, A-Life can be conceived as a special sort of philosophy [23], which allows the cre- ation and testing of complex thought ex- periments, "kept honest by requirements that could never be imposed on the naked mind of a human thinker alone" [24]. That is why I consider A-Life re- search to consist basically in the creation of prosthetically controlled thought ex- periments of indefinite complexity. By in- creasing the capacity and precision of the human mind through these prostheses that are computers, we will be able to in- crease indefinitely the complexity of such experiments as well.

However, the computer does not sim- ply act as an extension of our conceptual world, as Dennett asserts; its novelty lies in the blending of concepts with com- putational technology, creating a new do- main with its own ontological claims. In A-Life, a set of systems generated as a re- sult of putting together human concepts and artifacts somehow is constituted as real objects for epistemological purposes. In other words, an ontology is founded from an epistemological standpoint. Yet this happens in such a way that concepts once embodied in the machine are trans- formed; they have acquired some onto- logical dimension and establish, in turn, a new epistemological relationship with their creators and users.

This is why A-Life, on the one hand, constitutes an engineering activity whose starting point is a set of basic biological intuitions about the empirical biological domain. A-Life then attempts to build sys- tems capable of producing emergent and functional processes by themselves. In order to do so, it proceeds by creatively integrating different technological re- sources with theoretical developments in computer science and physics. On the other hand, A-Life involves a process of interpreting the behavior of such systems with the aim of broadening the phe- nomenology of the empirical biological

domain and of developing new concepts coherent with the body of knowledge of traditional biology, which properly stud- ies that domain. Both processes- construction and interpretation-are deeply entangled, because A-Life re- search is a continuous concatenation of constructions, interpretations, modifica- tions, new interpretations and re- constructions.

CONCLUSION

If the issues are presented and under- stood as above, A-Life could help to over- come the gap between empirical science and philosophy, as each of them has tra- ditionally been conceived. In traditional philosophy of science, the interaction under analysis involves the empirical do- main, the empirical theories and the meta-theories (a relationship in which technology plays a significant role only in the interactions between the first two). In A-Life, by contrast, a new way of in- terrelating empirical theories and meta- theories arises, which involves the mediation of technological devices. On the one hand, this new connection is more limited than in the traditional phi- losophy of biology, since A-Life must be restricted to conceptual issues that arise in the context of implemented (compu- tational) systems; but on the other hand, A-Life is more rigorous and powerful, be- cause it uses quasi-experimental methods of validation (computational experi- mentation) and also because it can, in principle, make the traditional range of problems of biological meta-theories even wider. Hence, A-Life research is opening up radically new perspectives, not only by bringing about profound changes in the meaning of such concepts as experimentation, models or evalua- tion (of theory), but also because the sta- tus of the philosophy of biology is clearly being modified and becoming deeply in- terwoven with technology. All this puts in question the classical distinctions be- tween science and philosophy and de- mands a more elaborate framework for proper accounts of the ever more com- plex and dialectical bonds between them.

References and Notes

1. C. Emmeche, The Garden in the Machine (Prince- ton, NJ: Princeton Univ. Press, 1994) p. 161.

2. M. Boden, "Autonomy and Artificiality," in M. Boden, ed., The Philosophy of Artificial Life (Oxford, U.K.: Oxford Univ. Press, 1996) pp. 95-108; L. Risan, "Why Are There So Few Biologists Here?" in P. Hus- bands and I. Harvey, eds., Proceedings of theFourth Eu- ropean Conference on Artificial Life (Cambridge, MA: MIT Press, 1997) pp. 28-35.

404 Moreno, Artificial Life and Philosophy

This content downloaded from 188.72.126.181 on Thu, 12 Jun 2014 20:55:43 PMAll use subject to JSTOR Terms and Conditions

Page 6: Artificial Life and Philosophy

3. A. Moreno. and J. Fernandez, "A Vida Artificial como Projeto de Criacao de uma Nova Biologia Uni- versal," in C.N. El-Hani and A.A.P. Videira, eds., 0 que e vida? Para entender a Biologia do seculo XXI (Rio deJaneiro, Brazil: Editorial Relume Dumara, 2000) pp. 257-271.

4. A. Moreno and K. Ruiz-Mirazo, "Metabolism and the Problem of Its Universalization," BioSystems 49, No. 1, 45-61 (1999).

5. N. Block, "What Is Functionalism?" in N. Block, ed., Readings in the Philosophy of Psychology, Vol. 1 (Cambridge, MA: Harvard Univ. Press, 1980) pp. 171-184; E. Sober, "Learning from Functional- ism: Prospects for Strong Artificial Life," in C. Lang- ton, C. Taylor, D. Farmer and S. Rassmusen, eds., Artificial Life II (Redwood City, CA: Addison-Wesley, 1992) pp. 749-765.

6. C. Langton, "Artificial Life," in C. Langton, ed., Artificial Life (Redwood City, CA: Addison-Wesley, 1989) pp. 1-47.

7. Emmeche [1] p. 163.

8. E.T. Olson, for instance, has argued against the "immateriality" of those "virtual organizations," as- serting that one could actually see them as real phys- ical systems if, instead of looking at the computer program in abstract terms, one considers the pat- terns of electrons which materially support the exe- cution of that program. See E.T. Olson, "The Ontological Basis of Strong Artificial Life," Artificial Life3, No. 1, 29-39 (1997). Yet this argument misses the point, because the material structures that sup- port the operational level of computer simulations are en- tirely passive, as their intrinsic dynamics is completely constrained. See Moreno and Ruiz-Mirazo [4].

9. H.H. Pattee has criticized this confusion between computational simulations and realizations of mate- rial biological systems, arguing that the former are mere symbolic systems, whereas the latter derive their capacities precisely from the autonomously gener- ated symbolic constraint in material systems. See H.H. Pattee, "Simulations, Realizations and Theories of Life," in Langton [6] pp. 63-77.

10. B.C. Smith, On the Origin of Objects (Cambridge, MA: MIT Press, 1996) pp. 75, 359.

11.J. Casti, Would-Be Worlds: How Simulation Is Chang- ing the Frontiers of Science (New York:John Wiley, 1997) pp. 187-188. J. Casti, "The Computer as a Labora- tory," Complexity4, No. 5, pp. 12-14 (1999).

12. In fact, one of the main objectives of a model in A-Life is to show something new not only in a given phenomenological domain but also in the domain of the internal correlations of the systems under study.

13. Another problem to be avoided is that of ambi- guity, i.e. too-generic results congruent with a very large range of phenomena.

14. D. Gross and R. Strand, "Can Agent-Based Mod- els Assist Decisions on Large-Scale Practical Prob- lems? A Philosophical Analysis," Complexity 5, No. 6, 26-33 (2000).

15. Quoted inJ. Horgan, "From Complexity to Per- plexity," ScientificAmerican (June 1995) pp. 104-109.

16. Miller, for example, has claimed that A-Life can only become a fully scientific discipline through its integration into the field of theoretical biology. See G. Miller, "Artificial Life as Theoretical Biology: How to Do Real Science with Computer Simulation,"

Technical Report CSRP 378, COGS (University of Sussex, U.K., 1995).

17. At the same time, A-Life research programs al- lows the reformulation of such questions.

18. Casti, Would-Be Worlds [11].

19. D. Lane, "Models and Aphorisms," Complexity 1, No. 2, pp. 9-13 (1995).

20. G.B. Kleindorfer, L. O'Neil and R. Ganeshan, "Validation in Simulation: Various Positions in the Philosophy of Science," Management Science 44, No. 8, 1087-1099 (1998).

21. Although there are, of course, various kinds of technological tools within A-Life (such as robots or other devices), the most important and prevalent way of developing and testing models in this research field is through computers. Therefore, my discussion focuses on them.

22. D. Dennett, "Artificial Life as Philosophy," Arti- ficial Life 1, No. 3, 291-292 (1994).

23. Other authors showing enthusiasm for the future of A-Life as a philosophy include E.W. Bonabeau and G. Theraulaz, "Why Do We Need Artificial Life?" Ar-

tificial Life 1, No. 3, 303-325 (1994).

24. Dennett [22] p. 291.

Alvaro Moreno is a philosopher of science, whose specialty is the philosophy of biology and complex systems at large: he has written a book and many articles about the methodology and

epistemology of artificial life.

Moreno, Artificial Life and Philosophy 405

This content downloaded from 188.72.126.181 on Thu, 12 Jun 2014 20:55:43 PMAll use subject to JSTOR Terms and Conditions