Conference on Collective Wisdom Collège de France 22

Conference on Collective Wisdom, Collège de France, 22 and 23 May 2008 Dan Sperber & Hugo Mercier Institut Jean Nicod (CNRS/EHESS/ENS) Reasoning as a Social Activity

The conference’s question “What are the nature, causes, and mechanisms of the emergent phenomenon by which a group can be more likely to solve its problems, obtain a more accurate information or make more accurate predictions than any of its members on its own, including any experts in the group? ”

Collective Wisdom? The good, the bad, and the ugly

Collective Wisdom? The good, the bad, and the ugly

• Different individual psychological dispositions are involved in different collective performances Hence: • A deep understanding of collective performance is not compatible with shallow psychology

• A deep understanding of collective performance is not compatible with shallow psychology

The disunity of human thought processes • The common view of human thinking as a unified process is based on – conscious access to our thoughts – our relative ability to integrate our conscious thoughts in discursive form • Empirical research suggest that – Conscious access to thought processes is very poor – thought processes are carried out by a variety of autonomous inference mechanisms

Inference • Inferences are processes that, given some input information, derive conclusions in a cognitively sound way Ø Inference is an essential ingredient of any cognitive system

Dual process theories of inference • Two distinct systems (or sets of systems) underly human inference • System 1 is automatic, unconscious, parallel, and fast • System 2 is controlled, conscious, serial, and slow.

An example of automatic inference Which of these two politicians is more competent? (Look carefully. You will have 1 second to decide)

?

?

Election to the US Senate, Wisconsin 2004 Russ Feingold Tim Michels 55% Elected 45%

Percentage of races correctly predicted on the basis of perceived competence • Senate (2000 -2002 -2004) 71. 6% (n = 95) • House of Representatives (2002 -2004) 66. 8% (n = 600) Todorov, A. et al. (2005). Inferences of competence from faces predict election outcomes. Science, 308, 1623 -1626.

Intuitive and reflective inference • Most of our inference are intuitive inferences: we come to believe their conclusion without attention to, or awareness of reasons • Some of our intuitions are about reasons. We come to believe that accepting premises P 1, P 2, …Pn, is a good reason to accept conclusion C • If we then infer C from P 1, P 2, …Pn , this is a reflective inference: we have attended to and are aware of the reasons for believing C

• You walk out of your house, you see fast moving dark clouds, feel a strong chilly wind, you spontaneously come to expect that it might rain. [Intuitive inference] (i. e. Unreflective, or spontnference] • You look at the barometer and at thermometer, see that both pressure and temperature have suddenly dropped, and you tell yourself: "this means that it might rain. “ [Reflective inference]

Reasoning Reflective inference is reasoning proper

A complex enthymematic argument • They said that [Lincoln] was unprepared and they implied that therefore he would not be a good president • Lincoln was a great president • Therefore the argument against Lincoln was not good • They say that Obama is unprepared and imply that therefore he would not be a good president • By parity of reasoning, the argument is not good against Obama

Why Do Humans Have Reasoning at All? • “Cartesian” view: To enhance individual cognition

Why Do Humans Have Reasoning at All? • “Cartesian” view: To enhance individual cognition • Evolutionary approach: To increase the acceptability of communicated information

An evolutionary argument: cooperation • Cooperation is advantageous to honest cooperators but even more so to cheaters • Hence, for cooperation to evolve, cheating must be prevented • In the human case, it is prevented by cognitive means

An evolutionary argument: communication • Communication is advantageous to honest communicators but even more so to dishonest ones • Hence, for communication to evolve, communicated information must be adequately filtered • In the human case, it is filtered in particular by means of reasoning

Ways to Filter Communicated Information: • Trust people who are benevolent and competent • Watch for behavioural signs of honesty or dishonesty • Check communicated information for consistency (both internal and external)

Ways to Persuade a Cautious Audience: • Do no directly assert what your audience won't accept on trust, let them infer it • Give evidence and arguments for the intended conclusion • Help your audience derive the intended conclusion • Highlight logical or evidential relations between premises the audience is willing to accept and the intended message

Argumentation • Arguing consist in displaying the consistency relationships between premises the audience is willing to accept and conclusions which they won’t accept on trust • Arguing is a relatively costly, hard-to-fake “honest display strategy” ,

The intuitive aspect of reasoning • Reasoning involves intuitions about logical and evidential relationships among representations • The resources for deriving these intuitions are activated in a dialogic context (actual, intended, or imagined)

Predictions • Classical view: – Reasoning should apply across the board – Its successes and failures should be accounted for by the limited capacity of reasoning and the difficulty of the tasks • Argumentative theory: – Reasoning should be triggered more easily in dialogic and argumentative contexts

When is reasoning efficient? • Decontextualised deductive reasoning? “it must be said that logical performance in abstract reasoning tasks is generally quite poor. ” (Evans 2002)

Explanations of poor performance • Classical view: – Cognitive limitations (e. g. poor untaught reasoning abilities, working memory limitations) • Argumentative theory: – The conditions for the full activation of the reasoning module are not met

When is reasoning efficient? • Decontextualised deductive reasoning? • Reasoning in argumentation?

Reasoning in argumentation People have been shown to be good at • making the explanation / evidence distinction (Brem & Rips 2000) • building counterarguments (Pennington & Hastie 1993) • understanding the macrostructure of arguments (Ricco 2003) • understanding burden of proof arguments (Baillenson & Rips, 1996; Rips 1998) • recognizing circular reasoning (Rips 2002) • using hypothetical thinking in argumentation (Green et al. 2006) • recognizing fallacies (Neuman et al 2006; Weinstock et al 2004; Neuman et al 2004)

When is reasoning efficient? • Decontextualised deductive reasoning? • Reasoning in argumentation? • Group reasoning?

Group reasoning: A form of ‘collective wisdom’ • The argumentative theory predicts that people should be better at recognising good arguments and justifications than at constructing them • Hence group processes may improve performances

Good arguments win • In mathematical and logical problems • In inductive problems (Laughlin & Ellis 1986; Laughlin 1999; Laughlin et al 2002)

Wason selection task Individual resolution Group resolution 18% Correct answers 80% Correct answers Moshman & Geil 1998

The ubiquitous confirmation bias (Nickerson 1998; Kunda 1990) • Preferential treatment for the hypotheses that support our beliefs (Wyer & Frey 1983) • Overweighing of the examples that fit our beliefs (e. g. Kunda 1987) • Finding (only) what we are looking for in data (Snyder 1981, 1984; see Nisbett & Ross 1980) • Problems with belief revision (e. g. Ross & Lepper 1980) • Problems with generating counter arguments against claims we hold (e. g. Kuhn 1990)

Explanations of the confirmation bias • Classical view: – Useful heuristic in some cases only • Argumentative theory: – Means to help us win arguments – Can be seen as a division of cognitive labor

• The division of argumentative labour is likely to succeed if people interacting have competing viewpoints and a common interest in truth (as in science) • If the people interacting have a common interest in defending a given single viewpoint, a runaway exaggeration of that viewpoint may ensue (as in political or religious fanaticism)

Average punitive awards given by 500 mock juries Statistical juries 409, 200$ Deliberating juries 1, 510, 000$ Schkade, Sunstein, &Kahneman. 2000

Conclusion • Reasoning is a mental mechanism with a primary social function: producing reasons to convince others and evaluating reasons given by others • This function helps explain the success and failures of individual reasoning • It helps explain how reasoning in groups can contribute to collective wisdom (or collective madness)

the end

• A man form Illinois was running for president. His opponents ridiculed him as unexperienced and woefully unprepared. His only governement experience had been servicing the Illinois legislature, plus two years as an obscure member of congress. He never held an excutive or management position of any kind. Yet this man was elected president. Twice. And they said he was unprepared.