Social heuristics that make us smarter

I argue that an ecologically distributed conception of instrumental rationality can and should be extended to a socially distributed conception of instrumental rationality in social environments. The argument proceeds by showing that the assumption of exogenously fixed units of activity cannot be justified; different units of activity are possible and some are better means to independently given ends than others, in various circumstances. An important social heuristic, the mirror heuristic, enables the flexible formation of units of activity in game theoretic situations, including collective units where these are instrumentally effective. In effect, the mirror heuristic makes the formation of units of activity endogenous to instrumental rationality. Moreover, the mirror heuristic is a conditional metaheuristic, which depends on mind reading of the heuristics of other players rather than on predictions of their behavior. Such mind reading can be regarded as emerging from an arms race between behavioral mimicry and ever smarter behavior reading. Even though unilateral mind reading may have benefits, the mirror metaheuristic illustrates that mutual mind reading has distinctive functions in responding to the challenges of social complexity. If simple heuristics can make us smart in the right environments, then social heuristics can make us smarter still. Keywords: RationalityCollective ActivityExogenous UnitsMirroringMind ReadingHeuristicsCooperationSituated Cognition Acknowledgments Thanks to Ralph Adolphs, Philipp Beckmann, Michael Bratman, Andrew Colman, Natalie Gold, Christoph Lumer, Elizabeth Pacherie, Hannes Rakoczy, Christopher Woodard and the anonymous referees for discussions of and comments on earlier drafts of this material. This article expands on and revises a chapter entitled "Rational agency, cooperation, and mind-reading" in Natalie Gold, Ed. (Citation2005), Teamwork: Multi-disciplinary perspectives (pp. 200–215), Basingstoke, Hampshire: Palgrave Macmillan; I am grateful to the editor for permission to draw on it here. Notes (and from 2006, also at Department of Philosophy, University of Bristol) See and compare CitationBratman (1992, p. 327), on how shared cooperative intentions can be intermediate between separate ultimate reasons and separate proximate means. See Kahneman, Slovic, and Tversky (Citation1982), Nisbett and Ross (Citation1980), CitationBargh (1999, Citation2005), Bargh, Chen, & Burrows (Citation1996), Bargh and Chartrand (Citation1999), Chartrand and Bargh (Citation1999), etc. For present purposes, heuristics and biases include confirmation bias, the representativeness heuristic, the base rate fallacy, uncertainty aversion, the Wason effect, attribution errors, chameleon effects, and so on. By, for example, CitationSterelny (2003, p. 25); cf. Gigerenzer (Citation1997) for skepticism about claims of social complexity. Natural selection may do plenty of "outwitting" in phylogenetic time, but not in decision-making time. Such boundaries can be understood by reference to an organism at a time, or in spatial or other terms. For a related discussion, see CitationDanielson (1992, p. 197). The exogenously specified units don't have to be individuals; they could for example be nations in an arms race. But individual units are most commonly assumed. By 'cooperation' I usually refer to the cooperative move in a PD, in contrast to defecting. At times I use the term more generally, to mean, roughly, doing one's part to obtain result A rather than result B, where the players agree in preferring A to B. Cooperation usually suggests the players only agree in their preferences about some possible results but disagree about others, whereas coordination suggests they agree about all possible results. See CitationHurley (1989, pp. 153–154), for an example that involves voting strategy in a corrupt electoral system; see also CitationColman (2003, p. 145) for a response to another tempting but unsuccessful way to avoid the Hi-Lo problem. See also CitationDanielson (1992, p. 16), and CitationSchmitt and Grammar (1997, pp. 94, 96, 106), for similar emphases on instrumental assessment. This does not entail that collective mental states or collective subjects are possible; see the end of this section for an elaboration of this point. Roughly, a tendency to give too much weight to initial evidence, and not to correct sufficiently in the light of later evidence that points another way. Here, I disagree with Colman's response (Citation2003) to my commentary (Hurley, Citation2003c); he assumes that collective activity depends on preference aggregation and thus faces the obstacle of social choice results. But in my view, collective activity can serve non-aggregated individual preferences and goals, as in Howard's Mirror Strategy for the Prisoner's Dilemma, which does not depend on preference aggregation (although it does depend on some degree of overlap in individual preferences: in PDs, e.g. the players agree in their judgments about which outcomes are second best and third best, despite their disagreement about best and worse outcomes). I am here alluding to Elster's (Citation1979) metaphor for pre-emptive strategies. In Newcomb's Problem, an agent faces a transparent box containing a $1 000 and an opaque box containing either $1 000 000 or nothing. She can choose to take either the opaque box only, or both boxes. She has been told, and believes, that a highly reliable predictor of human behavior has already put $1 000 000 into the opaque box if and only if it has predicted that she will take only the opaque box. Taking the opaque box has a better expected utility than taking both boxes, and it is also good news, since it indicates that you are the type of person who takes one box, which the predictor is likely to have predicted; but taking both is a dominant strategy since it yields an extra $1 000 whether or not $1 000 000 has already been put in the opaque box. For references and further discussion see CitationHurley (1991, Citation1994). Related rival hypotheses could explain the choice of the one box strategy in Newcomb's Problem. I follow the current usage of 'mind reading' instead of 'theory of mind' for the generic capacity to understand the mental states of others, since 'theory of mind' may suggest that the theory theory rather than the simulation theory of this capacity is correct. The latter is a substantive issue that should not be biased by terminology (see also CitationNichols & Stich, 2003, p. 2; CitationWhiten, 1997, pp. 150ff.). For some of the large body of literature on these claims, see Carruthers and Smith (Citation1996), Davies and Stone (Citation1995a, Citation1995b), Hare (Citation2001), Hare, Call, Agnetta, & Tomasello (Citation2000), Hare, Call, Tomasello (Citation2001), Heyes (Citation1998), Nichols and Stich (Citation2003), Povinelli and Vonk (Citation2005), Tomasello (Citation1999), CitationTomasello and Call (1997, Citation2005), Tomasello and Carpenter (Citation2005), etc. See and compare Call and Tomasello (Citation1999), Heyes (Citation1998), Heyes and Dickinson (Citation1993), Hurley (Citation2003a), Povinelli (Citation1996), and Povinelli and Vonk (Citation2005). See and compare Carruthers and Smith (Citation1996), Davies and Stone (Citation1995a, Citation1995b), CitationHare et al. (2000, Citation2001), Heyes (Citation1998), Povinelli and Vonk (Citation2005), and Tomasello and Call (Citation2006). For related points in a different context see CitationSterelny (2003, pp. 67ff.); cf. CitationWhiten (1996, Citation1997). Note that the relevant mind reading here is of intentions rather than of beliefs—an ontogenetically and phylogenetically earlier capacity (see CitationWhiten, 1997, p. 167; CitationTomasello, 1999; CitationTomasello & Call, 1997). So-called greenbeard genes could produce genetically determined analogs of such imitative free riding (CitationDawkins, 1982, p. 149). But the evolution of a general capacity for selective imitation would make it possible to gain the advantages of free riding without the need to evolve genes for specific behaviors. In this way the ability to turn imitation on and off would be a Machiavellian social advantage; on Machiavellian intelligence, see Byrne and Whiten (Citation1988), and Whiten and Byrne (Citation1997). Krebs and Dawkins (Citation1984) describe an arms race between signaling and deception that leads to mind reading. Here I make a similar, but more specific claim about how an arms race between signaling and imitation could lead to metaheuristic solutions to games of cooperation and coordination, such as the mirror metaheuristic in PDs. As in Dijksterhuis' (Citation2005) "high road" to imitation, which involves copying general patterns of behavior associated with character traits and stereotypes found in one's social environment, rather than copying specific observed behaviors (see also CitationGambetta, 2005). See CitationHurley (in press) for an extended version of this argument; see also Hurley (Citation2003b). Here I am indebted to discussion with Ralph Adolphs. See Adolphs et al. (Citation2000), and Gladwell (Citation2002). Note that some of the most compelling recent evidence for some limited mind reading abilities in apes relates to competitive rather than cooperative contexts (CitationHare, 2001; CitationHare et al., 2000, Citation2001; CitationTomasello & Call, 2006). Perhaps our own ancestors faced especially intense pressures to achieve flexible cooperation across a range of circumstances not amenable to genetic solutions. In Hurley (Citation2003a,Citation2003b) I defend the pointfulness of understanding as rational agents animals who do not in turn attribute rational agency to others. I am focusing here on early mind reading skills, which involve attributing agency, rather than more advanced mind reading skills, which involve, for example, attributing false beliefs to others where appropriate. Additional informationNotes on contributorsSusan Hurley (and from 2006, also at Department of Philosophy, University of Bristol)