Simple heuristics in a social world - ETH Z · – Sequential processing – Noncompensatory – Limited information (a “pruned” tree) – Provide a good descriptive account of

Simple heuristics in a social world

Ralph Hertwig

Presenter

Presentation Notes

The aim of the conference is to explore the potential of game theory for sociological theory and its application to sociological research broadly considered. For this purpose the conference brings together scholars with different disciplinary backgrounds to focus on topics of game theory relevant to sociology and society at large.

Herbert A. Simon1916 - 2001

Simon‘s question

How do human beings reason when the conditions for rationality postulated by the model of neoclassical economics are not met? (Simon, 1989, p. 377)

Bounded rationality

Presenter

Presentation Notes

Standard knowledge and rationality assumptions of game theory. Common knowledge assumption: The specification of the game and the players‘ preferences among the outcomes together with everything that can be logically deduced about the game, are common knowledge among the players. The players are rational in the sense that they always seek to maximize their own expected utility, and this is common knowledge amoung the players.

Bounded rationality: Three visions• Optimization under constraints (as-if rationality)

– “Boundedly rational procedures are in fact fully optimal procedures when one takes account of the cost of computation in addition to the benefits and costs inherent in the problem as originally posed.” (Arrow, 2004)

• Cognitive illusions – Our research attempted to obtain a map of bounded

rationality, by exploring the systematic biases that separate the beliefs that people have and the choices they make from the optimal beliefs and choices assumed in rational-agent models (Kahneman, 2003)

• Homo heuristicus (“Darwinian model of rationality”)– “Relatively simple choice mechanisms could enable an

organism, searching through its life’s maze, to survive in an uncertain environment. It depicted a procedural rationality for organisms that was squarely based on satisficing rather than optimizing” (Simon, 1996) 3

Presenter

Presentation Notes

Carry coals to Newcastle

44

Fast and frugal heuristics

• Take-the-best: Gigerenzer & Goldstein (1996). Psychological Review

• Fast & frugal trees: Martignon, Katsikopoulos, & Woike (2008). Journal of Mathematical Psychology

• Priority heuristic: Brandstätter, Gigerenzer, & Hertwig (2006, 2008). Psychological Review

• Recognition heuristic: Goldstein & Gigerenzer (2002). Psychological Review

• Fluency heuristic: Schooler & Hertwig (2005). Psychological Review; Hertwig, Herzog, Schooler, & Reimer (2008). Journal of Experimental Psychology: LMC

• Equity heuristic: Hertwig, Davis, & Sulloway (2002). Psychological Bulletin

• QuickEst Heuristic, Categorization-by-elimination, default heuristic, social-circle heuristic ...

5

Take-the-best heuristic

Example: Which city has more residents: Basel or Bern? [airport, state capital, industrial center etc.]

• Search rule: Look up the cue with highest validity vi (= correct inferences / all inferences)

• Stopping rule: If cue values discriminate (+/-; +/?), stop search. Otherwise go back to search rule.

• Decision rule: Predict that the alternative with the positive cue value has the higher criterion value

66

Which city has the higher population?Cues: soccer team, university, state capital, intercity train line, exposition site etc

Pred

ictiv

e ac

cura

cy

Sample size

Gigerenzer & Brighton 2009 topiCS

Gigerenzer & Brighton (2009).Topics in Cognitive Science, 1.

So far, so good but ...

7

8

Cognitive simplicity falters in the face of environmental complexity

• “Thus I doubt that rational behavior can be found in ‘fast and frugal’ heuristics. I think it is no accident that the examples of such heuristics in action ignore interactions with other intelligent agents, especially competitive agents. For it is precisely in such situations that simple rules of thumb will go wrong…. Catching a ball is one problem; catching a liar is another.”(Sterelny, 2003, p. 53)

• How Sterelny (and other proponents of this thesis) got it wrong!

9

Gaze heuristic

Gaze heuristic: Fixate your gaze on the ball, start running, and adjust your running speed so that the angle of gaze (i.e., the angle between your eye and the ball, relative to the ground) remains constant.

Presenter

Presentation Notes

Maintenance of the optical angle appears to be used not only in the interception of inanimate objects (games against nature) but also in the pursuit of prey (social games). Bats, birds, and dragonflies, for instance, have been found to maintain a constant optical angle between themselves and their prey (see Shafer et al., 2004). Prey such as moths appear to thwart predators’ use of the gaze heuristic by tumbling and looping unpredictably when hit by bat ultrasound (signaling the approach of a predator)—an instance of protean behavior (Miller, 1997). explosion of hot gases and debris killed scores of people and sent more than 160,000 villagers fleeing to underprepared evacuation camps.

1010101010

Heuristics flourish in the face of complexity

• Computational intractability– “Choice in social interaction harbors a level of complexity

that makes it unique among natural decision-making problems.... These iteratively nested levels of complexity render many social decision-making problems computationally intractable” (Seymour & Dolan, p. 667)

• Competing goals– The social world adds a class of goals to decision making

that are important for creating and maintaining social structure and cooperation, such as transparency, fairness, and accountability

• Incommensurable reasons in social environments– Sacred values preclude comparisons and tradeoffs

(Tetlock et al., 2000) • Time pressure (Todd, 2001)

Hertwig & Herzog (2009). Social Cognition, 27, 661-698.

Presenter

Presentation Notes

Optimization is, in general, impossible for problems with multiple criteria. One cannot maximize several criteria simultaneously (unless one combines them by, say, a linear function)

Taxi drivers in the U.S. fall “victim to more deadly violent assaults (184 per 1,000) than any other occupation with the exception of police” (p. 1).

Social learning of cues to trustworthiness or lack thereof (e.g., a fully zipped, bulky coat) rather than individual (trial-and-error) learning.

11

Research questions: Simple heuristics in a social world

I. What heuristics do people use in social environments?

The descriptive study of the adaptive toolbox

II. When are these heuristics successful? The normative study of ecological rationality

III. How to design heuristics, strategies, decision systems, and social environments?

Intuitive design

12Hertwig, Hoffrage, & the ABC Research Group (in press). Simple heuristics in a social world. New York: Oxford.

The mini-ultimatum game

500

500

Proposer

Responder800

200

Offer 1 Offer 2

Presenter

Presentation Notes

Extensive Form representation of a two proposal ultimatum game. Player 1 can offer a fair (F) or unfair (U) proposal; player 2 can accept (A) or reject (R)

1414

Social preference models are as-if models

• The established, as-if modeling approach in economics extends the utility function by introducing psychological variables such as– Inequity aversion (e.g., Fehr & Schmidt, 1999), intention-based

reciprocity (Rabin, 1993), or a combination of both (Falk & Fischbacher, 2006)

• This approach resembles the parameterized as-if models in the risky choice literature– Models such as cumulative prospect theory accept the

Bernoullian framework and “inject” psychology into these functions (e.g., inverse S-shaped weighting function)

– But there are heuristic alternatives, e.g., priority heuristic (Brandstätter, Gigerenzer, & Hertwig, 2006, 2008)

1515

What would a psychologically plausible approach look like?

• Classification decision: Does this allocation belong to the class of allocations that I reject/accept?

• Subcategory of heuristics: Fast and frugal classification trees (e.g., Martignon, Vitouch, Takezawa, & Foster, 2003)– Sequential processing– Noncompensatory– Limited information (a “pruned” tree)– Provide a good descriptive account of decisions in

medicine and law, among other domains (e.g., Green & Mehr, 1997)

Hertwig, Fischbacher, & Bruhin (in press). In Simple heuristics in a social world.Fischbacher, Hertwig & Bruhin (2011). Manuscript submitted for publication.

1616

His payoff

My foregone payoff

His foregone payoff

Proposer

Status My payoff

Mirror Kindness

Responder

What social criteria might responders use?

Positive payoff

Presenter

Presentation Notes

The theory of moral sentiments. Sympathy" was the term Smith used for the feeling of these moral sentiments. It was the feeling with the passions of others. It operated through a logic of mirroring, in which a spectator imaginatively reconstructed the experience of the person he watches:As we have no immediate experience of what other men feel, we can form no idea of the manner in which they are affected, but by conceiving what we ourselves should feel in the like situation. Though our brother is on the rack, as long as we ourselves are at our ease, our senses will never inform us of what he suffers. They never did, and never can, carry us beyond our own person, and it is by the imagination only that we can form any conception of what are his sensations. Neither can that faculty help us to this any other way, than by representing to us what would be our own, if we were in his case. It is the impressions of our own senses only, not those of his, which our imaginations copy. By the imagination, we place ourselves in his situation.

1717

How we proceeded

1. People made decisions in 12 mini-ultimatum games (and 12 dictator games)

2. Via a finite mixture model analysis of their decisions, we sorted participants into four classes of responders (due to heterogeneity)

3. We constructed four fast and frugal trees to describe the heterogeneous decisions

4. We made response-time predictions from these trees and tested these against empirical response-time data

1818

More than nothing?

no yes

100%

The selfish tree

Reject Accept

1919

The priority tree

Accept

As good or better than his?

no yes

AcceptReject

100%

80%97%

Kind?

no yes

2020

The priority+1 tree

Accept

As good or better than his?

no yes

Accept

Kind?

no yes

Would I have?

no yes

Reject Accept

100%

94%

46% 70%

2121

Would I have?

no yes

88%

The mirror tree

Reject Accept

46%

22

Being selfish is being fast

All decisions (in sec): 3.5 vs. 5.2Acceptances (in sec): 3.5 vs. 5.0

Time

23

Examining more priorities takes time

23

Priority tree (in sec)1. Status: 3.52. Kind: 5.0

Priority tree+1 (in sec)1. Status: 4.52. Kind: 4.73. Mirror: 6.1

ns

Time

24

Simple heuristics in social games

• The response-time data suggest a sequential reasoning process among respondents who rely on social criteria

• This reasoning process can hardly be captured by as-if models (such as inequity aversion)

• How well do the trees predict behavior (relative to as-if models), and can we generalize our modeling approach—i.e., fast and frugal classification trees—to other social games?

25

Conclusions

• Three aspects of bounded rationality in a social world, each of which represents a very rich set of research questions:– descriptive study of the adaptive toolbox– normative study of ecological rationality– intuitive design

• Simple heuristics can be an alternative to af-if models in economics and game theory (and beyond), focusing on cognitive processes and behavioral outcomes rather than merely outcomes

• Simple heuristics interacting with social and non-social environments can give rise to complex behavior (e.g., allocation heuristics, pedestrian heuristics)

Presenter

Presentation Notes

Moussaid, Helbing Theraulaz 2011 The heuristics generate pheniomena of self-organization phenomena such as the formation of unidirectional lanes and stop and go-waves. Moreover, the cominbation of pedestrian heuristics with body collisions generates crowd turbulance

Documents

Simple heuristics in a social world - ETH Z · – Sequential processing – Noncompensatory – Limited information (a “pruned” tree) – Provide a good descriptive account of