Implicit Decision-Making: RPD, Experts, Thin-Slicing,

Spontaneity, and Future Directions

William StadermanInformation Processing Technology

OfficeDARPA

Agenda

Decision making Recognition-primed decision model Mental simulation Experts Thin-slicing Spontaneity Future directions

Flavors of decision-making

Deductive logical thinking classical approaches what people are taught (more often than not)

Naturalistic decision making Some say this is what experts really do Certain settings provoke it:

• time pressure, high-stakes, experienced decision makers, inadequate information, ill-defined goals, poorly-defined procedures, cue learning, context, dynamic conditions

Classical Approaches

Rational choice strategy (P. Soelberg, 1967):1. Identify the set of options2. Identify ways of evaluating the options3. Weight each evaluation dimension4. Perform the rating5. Pick the option with the highest score

Cool idea, but he found people didn’talways use it when he expected them to.

Making Better RCS decisions(I.L. Janis and M.L Mann)

Thoroughly canvas a wide range of options Survey a full range of objectives Carefully weight the costs, risk and benefits of

each option Intensively search for new information in

evaluating options Assimilate all new information Reexamine the positive and negative

consequences of each option Carefully plan to include contingencies if various

risks occur

Stage Models (of Problem Solving)(Raanan Lipshitz and Orna Bar-Ilan)

Two stages: idea getting and idea evaluation Three stages: problem finding, design of alternatives,

and choice Four stages: understanding the problem, devising a plan,

carrying out the plan, and evaluating it Five stages: identifying the problem, defining it,

evaluating possible solutions, acting, and evaluating success

Six stages…Interesting, but these are hard with ill-defined goals

Recognition-Primed Decision Model(Klein 1998)

The ‘R’ in RPD

Recognition allows for the identification of analogues and goals (and the linkage)

Analogues generate expectancies and provide a structure for predictions

Expectancies help us perceive: Relevant cues Fine distinctions, i.e. important and unimportant

differences (zone of indifference) Leverage points/choke points Absence, anomaly, and counter-factuals

Mental Simulation

Used in three places in RPD:

1. Forming situation awareness

2. Generating expectancies

3. Evaluating course of action

Problems with Mental Simulation

Sometimes they lead you astray Discard contrary evidence – de minimus

explanations May cause overconfidence, especially if you

become wedded to a particular one A simulation is not situational awareness, but

can be confused as such Humans are not very good at it when

things become complicated

How People Choose Between These Models

Expert DM Perception is different

Fine discrimination, big picture, opportunity/cue/pattern (and anomaly/absence) detection skills

See options, not decisions Comparative vs. singular evaluation Herbert Simon’s term “satisficing”

• Not optimization, that is too hard/takes too much time• How do you know you’ve got a good option? Mental simulation

Handle time pressure better (perhaps because of RPD) Don’t “fly behind the plane” Know their own limitation and when they are losing the big picture Can improvise Problems: stereotypes and assumptions

How Do Experts Learn?(Klein)

Deliberate practice (with goals and evaluation criteria)

Compile extensive experience bank Obtain feedback that is accurate,

diagnostic, and reasonably timely Review prior experiences to derive new

insights and lessons from mistakes

Stories Ingredients (Klein):

Agents Predicament Intentions Actions Objects Causality Context Surprises

Good ways of encapsulating complex/ill-formed models not just shared stories, but shared interpretations

Some stories are more useful than others PBL is largely stories

Thin-Slicing

Gottman (1999) predicts whether married couples are married 15y later based on coding 15min discussion with 90% accuracy

SPAFF (Specific AFFect) coding and Morse code fists• Distinctive signatures• Content is irrelevant• Focuses on key cues (SPAFF: disgust, whining, contempt, defensiveness,

neutral, stonewalling, etc.) Medical malpractice suits

Doctor/Patient relationship; doctors not sued spent more time talking with patients and more engaged in conversation (Levinson, 1997)

Content-filtered recordings; 40 total seconds; cues: dominance, hostility, warmth, anxiousness; raters accurately predicted who was sued (Ambady, 2002)

Perceptual cuts – expressions, sweat, tone, gestures, posture…

Van Riper: Creating Structure from Spontaneity

JFCOM Millennium Challenge ’02 (wargame) – Red Team was commanded by rogue commander (Van Riper) in middle-east

Spontaneity isn’t random – successful rapid cognition is a function of training and rules and rehearsal

Create conditions for successful spontaneity Command, not control

Van Riper’s spontaneity and rapid actions circumvented and upset more formal and more powerful Blue Team

Limited traditional communication Verbal Overshadowing – words and descriptions interfere with

DM

Future directions Perception in DM

People clearly gather more information from subtle perceptual cues than acknowledged.

“micro” decisions based on this info can influence and accelerate “macro” decisions – priming? Pre-judgment?

Neural-net analogy Inductive Decision Making

“micro” decisions can accumulate to motivate actions and prime responses

Rules for spontaneity (improv) Hawkins (2005), On Intelligence – intelligence is a matter of

memory-prediction Affordances - perceivable possibilities for action

Future directions (cont) Affordances (Gibson, 1979) 

Perceive in order to operate on the environment; perception is designed for action; perceivable possibilities for action = affordances.

Perceive affordance properties of the environment in a direct and immediate way.

• Visualization In short, Gibson claims that we perceive possibilities for action. i.e.

surfaces for walking, handles for pulling, space for navigation, tools for manipulating, etc. In general, our whole evolution has been geared toward perceiving useful possibilities for action.

Affordance example (Norman, 1988) : You are approaching a door through which you eventually want to pass. The door, and the manner in which it is secured to the wall, permits opening by pushing it from its 'closed' position. We say that the door affords (or allows, or is for) opening by pushing. On approaching that door you observe a flat plate fixed to it at waist height on the 'non-hinge' side, and possibly some sticky finger marks on its otherwise polished surface. You deduce that the door is meant to be pushed open: you therefore push on the plate, whereupon the door opens and you pass through. Here, there is a perceived affordance, triggered by the sight of the plate and the finger marks, that is identical with the actual affordance. Note that the affordance we discuss is neither the door nor the plate: it is a property of the door (the door affords opening by pushing ).

References and Acknowledgements

David N. Blank-Edelman; Northeastern University; LISA 2001 Invited Talk

Gary Klein, Sources of Power: How People Make Decisions, MIT Press

Malcolm Gladwell, 2005, Blink, Little, Brown, and Company.

Gibson, J.J. (1979). The Ecological Approach to Visual Perception , Houghton Mifflin, Boston. (Currently published by Lawrence Eribaum, Hillsdale, NJ.)

Norman, D. (1988). The Psychology of Everyday Things , New York, Basic Books, pp. 87-92.