8
Onwards and Upwards with the Extended Mind:
From Individual to Collective Epistemic Action
Georg Theiner
In recent years, philosophical developments of the notion of distributed and/or scaffolded cognition have given rise to the “extended mind” thesis. Against the popular belief that the mind resides solely in the brain, advocates of the extended mind thesis defend the claim that a significant portion of human cognition literally extends beyond the brain into the body and a heterogeneous array of physical props, tools, and cultural techniques that are reliably present in the environment in which people grow, think, and act (Clark and Chalmers 1998; Clark 1997, 2003, 2008; Wilson 2004; Rowlands 1999, 2012; Menary 2007; Theiner 2011). However, as commentators who are friendly to the idea of distributed cognition have pointed out, the philosophical debate over extended cognition has predominantly focused on the impact of tools on our thinking while somewhat neglecting the distinctively social and cultural dimensions of cognitive scaffolding (Sterelny 2004, 2010; Caporael 1997a, 1997b; Smith and Semin 2004; Wilson 2005; Barnier et al. 2008; Sutton et al. 2010; Theiner, Allen, and Goldstone 2010).
To reorient the reigning paradigm, Hutchins (2010, 445) has recently proposed the “hypothesis of enculturated cognition” (HEnC) as an alternative to Clark’s (2003, 2008) largely individualistic vision of the extended mind. According to the HEnC, the “ecological assemblies of human cognition make pervasive use of cultural products” and are typically “assembled … in ongoing cultural practices” (ibid.). Cultural practices, for Hutchins, are essentially “the things people do in interaction with one another” (ibid., 440). My goal in this chapter is to follow up on Hutchins’s call to “spur the program forward” (ibid., 445), by generalizing Kirsh and Maglio’s (1994) distinction between pragmatic and epistemic actions from the level of individuals to the level of groups. The concept of a collective epistemic action refers to the ways in which groups of people actively change the structure of their social organization, with the epistemic goal of reshaping and augmenting their cognitive performance as integrated collectivities. By placing a renewed emphasis on the interactions between people, rather than between people and their tools, I hope to reconnect the cognitive-scientifically-driven “extended mind” thesis with complementary areas of social-scientific research in which groups are analyzed as the seats of action and cognition in their own right. In particular, the literature to which I aim to build a bridge in this paper is, on the one hand, certain segments of social and organizational psychology, (cf. Larson and Christensen 1993; Hinsz et al. 1997; Mohammed and Dumville 2001), and, on the other hand, theories of collective and institutional action (cf. Ostrom 1990; List and Pettit 2011).
What Is the Social Dimension of Extended Cognition?
A central trait of the human species, which has become increasingly complex in socially and technologically advanced societies, is the organized division of cognitive labor in groups. Being able to think in groups is a premier amplifier of our cognitive abilities since it means that no individual needs to be able to do all the thinking that can be collectively achieved by the group as a whole. What exactly does it mean to say that cognition is socially distributed in this sense? In this section, I compare several answers that have recently been given to this question from an “extended mind” perspective and consider some of their limitations. As a way of setting the stage for my subsequent discussion, I begin with a brief review of Kirsh and Maglio’s (1994) notion of epistemic action.
On Distinguishing Pragmatic and Epistemic Actions
In an influential study of situated cognition, Kirsh and Maglio (1994) investigated human problem solving in the context of the video game Tetris. To succeed in Tetris, a player has to determine under time pressure whether two-dimensional blocks of various shapes, displayed in falling motion on a computer screen, fit into open slots in an emerging wall at the bottom of the screen. Whenever a player manages to create a complete horizontal line of blocks without gaps, the line disappears, thus furthering the player’s ultimate goal to score as many line clears as possible. Traditional information-processing approaches to problem solving are predicated on the assumption that rational players always carry out actions that bring them closer to their desired goal state. Contrary to this prediction, Kirsh and Maglio found that expert players of Tetris regularly performed moves that seemed to be physically disadvantageous and less rational than the actions of novices from a traditional viewpoint. For example, experts would often rotate blocks on the screen just to see whether they fit into one of the slots—an action which, in case the block doesn’t fit, obviously cannot bring the player any closer to the goal. In contrast, novices usually did not perform any overt actions at all before they decided where to put a tile, attending solely to mental images of the falling blocks which they rotated in their heads.
In order to make sense of their observations, Kirsh and Maglio introduced a new category into their information-processing analysis. They argued that in addition to regular physical actions (e.g., filling a hole in the wall), people also perform epistemic actions. Such actions are performed not with the primary goal of bringing about the desired physical changes in the environment but with the aim of changing one’s informational state as a problem solver. Kirsh and Maglio showed that the experts’ method of on-screen rotation is a faster and less error prone way of solving spatial reasoning problems than the rotation of mental images in one’s head. What the Tetris experts were so prolific in doing was maximizing the effects of epistemic agency. Hence if we consider that cognitive agents operate in a joint state space that includes physical as well as epistemic goals, it turns out that expert players are more rational problem solvers, after all.
Kirsh and Maglio offer several, nonidentical glosses of the cognitive benefits that people can accrue from epistemic actions—calling them “actions that use the world to improve cognition” (ibid., 513), or actions that are “used to change the world in order to simplify the problem-solving task” (ibid.), or actions that can be used “to unearth valuable information that is currently unavailable, hard to detect, or hard to compute” (ibid., 515). Despite the subtle differences between these formulations, the upshot of their study is that difficult perceptual and cognitive problems are often solved more quickly, more easily, and more reliably by performing epistemic actions in the world than by performing the corresponding cognitive operations in one’s head.
Socially Extended Individual Cognition
Playing off Kirsh and Maglio’s study, Clark and Chalmers (1998) argued for the “extended mind” thesis on the grounds that epistemic actions performed directly in the world play a functionally similar role for the guidance of intelligent behavior as do brain-bound cognitive processes. One of their much-discussed examples is the case of Otto, an imaginary person suffering from Alzheimer’s disease who relies on his trustworthy notebook to take over the causal-informational role usually played by our biological long-term memory. Their reasoning is based on the premise—which has since become known as the parity principle—that “[I]f, as we confront some task, a part of the world functions as a process which, were it done in the head, we would have no hesitation of recognizing as part of the cognitive process, then that part of the world is (so we claim) part of the cognitive process” (1998, 8). As Clark (2005, 2) put it more recently, the point of the parity principle is to erode an unreflective “bio-chauvinistic prejudice” that unjustifiably favors location over functionality when it comes to demarcating the bounds of cognition (cf. Theiner 2011, chapter 2, for a review of the debate over the role of the parity principle).
Clearly, the parity principle leaves the door open for other people to function as external stand-ins for one’s internal cognitive processing. Clark and Chalmers briefly entertain the possibility that cognition might also be socially extended: “Could my mental states be partly constituted by the states of other thinkers? We see no reason why not, in principle. In an unusually interdependent couple, it is entirely possible that one partner’s beliefs will play the same sort of role for the other as the notebook plays for Otto. What is central is a high degree of trust, reliance, and accessibility” (1998, 17). In similar vein, they suggest that a secretary’s beliefs might be considered part of the extended memory of the boss, and that the recommendations of the waiter at a customer’s favorite gourmet restaurant might be construed as part of her extended beliefs, or perhaps even extended desires for good meals (ibid.).
What strikes me as objectionable about this suggestion is that it essentially treats the putatively “social” extension of cognition as just another capacity-enhancing tool of individual minds. One way to highlight the limited cognitive role that Clark and Chalmers are willing to grant to people with whom we interact is by considering their account of the conditions under which a certain environmental resource (e.g., a notebook, a calculator, a PDA) is sufficiently causally integrated with a person’s cognitive apparatus to be considered part of his or her extended mind. In the cases which they discuss, those “coupling conditions” boil down to the “high degree of trust, reliance, and accessibility” (1998, 17) that is typical of our interactions with well-calibrated, customized instruments. The underlying intuition here seems to be that the good, old-fashioned secretary is basically an imperfect version of a ubiquitous computing device. The technophilic assimilation of people and artifacts under the ambit of highly accessible auxiliary equipment confirms the suspicion, previously raised by Sterelny, that “cognition, for [Clark], remains paradigmatically a solitary vice, though one prosthetically enhanced by wideware” (2004, 246). Unlike the deeply mutual, continuous interactivity that is characteristic of socially distributed cognition, especially in close interpersonal relationships, my interaction with artifacts is extremely lopsided. For instance, a properly functioning measuring tool does not—and should not—take into account, and be responsive to (the rest of) my mental life, lest it undermine its epistemic value, which consists entirely of its accuracy. Consequently, I don’t have to coordinate my actions, beliefs, goals, and intentions with my tools to engage in joint actions. In fact, my tools and I don’t do anything together. If we aim to capture the distinctively collective aspects of socially distributed or extended cognition, we ought to take a different route.
The Social Manifestation Thesis and the Group Mind Thesis
Looking at the psychology of memory through the lens of the distributed-cognition framework, Barnier et al. (2008) usefully distinguish among three ways in which the process of remembering—or any other cognitive activity—can be socially distributed. According to (what they call) the triggering thesis, social phenomena can be a valuable source of input, trigger, or context eliciting individuals to engage in certain forms of psychological processes inside the head. For the bulk of research done in many areas of cognitive but also social psychology that tend to have a strongly individualistic bent, the triggering thesis succinctly summarizes a kind of standard view about the place of “the social” in cognition. As an alternative to the rampant methodological individualism in mainstream psychology, Barnier et al. appropriate the so-called social manifestation thesis (SMT) that has been developed in a series of papers and books by Robert Wilson (2001, 2004, 2005). The SMT asserts that at least some psychological processes can be manifested only insofar as the individual engaged in that process forms part of a social group of a certain kind. For instance, people experience emotional reactions like joy, anger, or sadness in response to events that concern a group to which they feel strongly attached (e.g., a family, a favorite sports team, a nation) just like they would experience events pertaining to their personal lives (Smith, Seger, and Mackie 2007). Similarly, social constructionists have argued that our ordinary practices of remembering are deeply relational, insofar as our ability to form and maintain unified, persisting memories of complex events in the past is heavily scaffolded by our reliance on the company of others (Campbell 2003; cf. Sutton 2009). Central to the SMT is the idea that individuals incorporate certain aspects of their social and cultural environment so deeply into their mental functioning that those aspects become constitutive features of cognition. Hence the SMT can be seen as advocating an “active” form of social externalism, insofar as the individual-bound portion of a socially manifested cognitive process is not seen as metaphysically sufficient for the psychological phenomenon in question to occur (Wilson 2001, 2004).
Finally, Barnier et al. dub as the group mind thesis (GMT) the claim that groups themselves, over and above the individuals who compose those groups, can engage in psychological processes or have psychological abilities in their own right. The somewhat pretentious locution of a “group mind” should be taken with a grain of salt. Contemporary proponents of group cognition typically restrict their claims to particular kinds of psychological predicates which they take to be shareable by individuals and groups (Theiner and Wilson 2013). Such predicates can be drawn from folk psychology (e.g., belief, intention, rational agency), may refer to classical mental faculties (e.g., memory, decision making, or problem solving), or involve more theoretically driven notions (e.g., adaptive information processing). Nevertheless, the notion of group cognition retains a commitment to the older emergentist idea that a group as a whole can have cognitive properties that none of its members has, properties that are irreducible to the properties had by those members (Theiner, Allen, and Goldstone 2010).
Conceptually speaking, the SMT and the GMT are clearly two separate theses. As Wilson (2001, S266) points out, the SMT could be true without entailing the truth of GMT if there are no psychological states at the group level. Conversely, the GMT might be true without entailing the SMT if the relevant individuals comprising the group mind did not have any psychological states of their own. The latter case points to an interesting distinction between two types of group-level (psychological or other) traits. Using Wilson’s terminology (ibid., S265), a group mind is a multilevel trait if it is said to exist at both the level of the group and the level of the individuals comprising the group; by contrast, a group mind is a group-only trait if it is claimed to exist only at the level of the group. According to Wilson, the “collective psychology” tradition, which played an important foundational role in the development of sociology (Durkheim 1898/1953) and also social psychology (McDougall 1920), was committed to the former. It viewed social groups of humans as autonomous entities with their own distinctive mental features that were supposed to be emergent from, and not reducible to, the psychological states of individual members. For Wilson, the latter possibility is perhaps most uncontroversially manifested by social insect colonies, which are often considered as a kind of “superorganism” (Wheeler 1920), capable of performing complex biological functions such as food collection and distribution, nest construction and maintenance, and reproduction of the hive. Since at least some of these capacities are arguably psychological in nature, the cognitive behavior of superorganisms would support a group-only reading of the GMT since individual bees, ants, or wasps lack the requisite capacities to perform any of these functions in solo.
As conceived by Wilson, the SMT is meant to provide a fertile “middle ground between an individualistic psychology and the group mind hypothesis” (Wilson 2001, S271). On the one hand, the SMT is sufficient to break down the strongly individualistic bias of contemporary psychology and cognitive science, while on the other hand, it stops short of positing that groups as a whole can be the bearers of psychological properties. I am entirely sympathetic to Barnier et al.’s pluralistic approach, which suggests that each of the three theses may well be true of a different range of socially distributed cognitive phenomena. My own goal in this chapter, however, is to explore an aspect of socially distributed cognition that goes beyond the SMT and is more naturally understood in terms of the GMT. Fleshing out the substance of my proposal, which is the task to which I shall now turn, will shed further light on the nature of the relationship between individual and group cognition.
From Individual to Collective Epistemic Action
The first step to consider is that the current debate over extended cognition is not primarily a dispute over whether certain extraneural tokens of bodily behavior or bioexternal artifacts are “intrinsically” cognitive or not. The deeper, but methodologically far more consequential, project is to reorient the study of cognition toward a greater appreciation of the significance that complex interactive couplings between brain, bodies, and their social and cultural environments have for the production of cognitive outcomes (cf. Clark 2008; Sutton et al. 2010). However, there are two importantly different ways of developing the idea of extended cognition once we elevate our unit of cognitive analysis from the level of individuals to that of groups.
According to one conception, our focus continues to rest on the external props, tools, and resources which scaffold the social interactions of people who collaboratively carry out certain cognitive tasks. Many studies of technologically regimented work environments (e.g., cockpits, navigation bridges, traffic control centers) conducted from a distributed-cognition perspective have been greatly concerned with understanding how “cognitive technologies” (Norman 1991) enable us to distribute cognition in space, time, and among people (Hutchins 1995). In his recent treatment of “scaffolded” cognition, Sterelny (2010) seems to have a similar conception in mind when he distinguishes between individual and collective cognitive resources. His main examples of the latter are cultural products such as language, maps, and writing systems, but also the socially organized transmission of expertise—that is, resources that have been created and modified by the cumulative activities of many generations and are typically used in cooperative situations. Sterelny’s notion of collective cognitive resources is interesting in its own right but remains centered on the social mechanisms by which human-made resources are culturally transmitted and the transformative effects those resources have on subsequent generations of individual learners.
As an alternative conception of collectively extended cognition—which is the one I want to promote here—we unambiguously direct our attention to the mechanisms by which groups of people actively change the structure of their own social organization, with the epistemic goal of reshaping and augmenting their cognitive performance as integrated collectivities. For this purpose, the dynamic, interaction-centered notion of epistemic action that we have encountered earlier in the work of Kirsh and Maglio (1994) seems to provide a more promising starting point than the static, object-centered notion of a cognitive resource. As a way of articulating a distinctively collective dimension of the “extended mind” thesis, I thus propose that we distinguish between individual and collective epistemic actions. On a first pass—paraphrasing Kirsh and Maglio, and thus also using three closely related, but nonidentical formulations—we can say that collective epistemic actions are actions that people perform to improve their cognitive performance as a group, or, alternatively, actions by which groups change the world in order to simplify their problem-solving tasks, or, as a final variant, actions that are performed by groups to unearth valuable information that is currently unavailable, hard to detect, or hard to compute. The reasoning behind this generalization from individuals to groups is an instance of the principle of social parity (Theiner and O’Connor 2010; see also Tollefsen 2006; Gallagher and Crisafi 2009). Building on Clark and Chalmers’s (1998) original parity principle (discussed above), the principle of social parity suggests that if, in confronting some cognitive task, the members of a group collaboratively interact in a process which, were it done in the head, would be accepted as a cognitive process, then that group (as a whole) is performing that cognitive process. Social parity can be viewed as a heuristic template to analyze the dynamics of group processes in terms of various cognitive functions (e.g., learning, memory, decision making), which serves to highlight certain important information-processing patterns in groups that are also observed when the same function is performed by individuals (Larson and Christensen 1993; Hinsz et al. 1997). In the rest of this section, I want to put some flesh on this abstract, generic characterization of collective epistemic action with the help of a few examples from collective decision making and judgment formation.
Example 1: Collective Decision Making
In real-world groups, information is typically unevenly distributed among its members because it is likely that each individual knows certain things that others don’t. Hence when groups must decide on a course of collective action, it would clearly seem to be in their best interest to pool the diverse, unshared knowledge of its members about relevant decision alternatives in order to make better decisions than its members could make if acting alone. The promise of such an assembly bonus (Collins and Guetzkow 1964) is often cited as the rationale to entrust groups with difficult decision-making processes, and for encouraging group members to exchange information in the form of group discussions, brainstorming, briefings, memoranda, or position papers. However, the literature on so-called “hidden profile” tasks is apt to challenge this popular opinion (Stasser and Titus 1985; for reviews, see Stasser 1999; Stasser and Dietz-Uhler 2001).
In the context of collective decision making, the concept of a hidden profile refers to situations in which the relevant information supporting the best decision alternative is unevenly distributed in a group in such a way that from the perspective of individual members, it is outweighed by shared information which favors a worse decision alternative. Hence only if a group is able to combine the unshared pieces of information can it discover the hidden profile, and thus collectively make the best possible decision. The persistent failure of groups to discover hidden profiles has spawned lots of studies aimed at explaining and possibly reversing this disappointing and prima facie counterintuitive trend (see Stasser and Titus 2003 for a brief history).
First, the collective information sampling model by Stasser and Titus (1987) predicts that if shared and unshared items are equally memorable, and all members have the same base rate of recalling and mentioning items, information that is more widely distributed is by default much more likely to enter into group discussion. Second, raising unshared information bears higher social costs for an individual because no other member of the group can vouch for its accuracy and importance. These costs can be reduced by explicitly labeling people as experts in the domain which is associated with their unique information, designating discussion leaders, or appointing members with relevant task experience who are more likely to bring up or repeat otherwise unshared information. Third, and spurred by Wegner’s (1986) idea of a transactive memory system, the success rate of exchanging unshared information can also be raised by assigning responsibility for specific categories of information to specific members. Assigning such expert roles cuts down the sampling advantage enjoyed by shared information because it limits each member’s attention to items (shared or unshared) that lie within his or her area of expertise (Stasser, Stewart, and Wittenbaum 1995). This effect can be strengthened if groups are required to rank-order decision alternatives, rather than simply picking the best, and it also helps when the decision problem is framed as an intellective task for which there is a known best answer, rather than as a purely judgmental task.
In sum, small groups are seemingly much better equipped to identify information that is already shared among its members than they are at integrating diverse yet complementary bodies of information. These results have significant implications for our discussion of group cognition. First, they show that it would be mistaken to treat the realization of an assembly bonus as a conceptually necessary condition for the manifestation of group-level cognition. This can be seen from the simple fact that sometimes group-level information processing can be adaptive if it produces an “assembly malus,” that is, if it yields a less-informed cognitive outcome than what would potentially be contained in the sum of its parts. A nice example of this sort is the widespread use of focus groups, because their function—as McQuarrie and McIntyre (1988) have argued—is precisely to filter out idiosyncratic perspectives and thus help to identify the most popular views in a population. Second, they also remind us that an unbridled collectivist enthusiasm for the superior decision-making capacities of groups is equally out of place as an opinionated individualistic preconception that human beings are poorly equipped to make decisions in groups. A better understanding of why groups frequently fall short of their cognitive potential can help us to devise procedures and interventions by which we can at least partly overcome these limitations.
Example 2: Collective Judgment Aggregation
My second example concerns the procedures by which multimember groups form judgments on the basis of individual judgments made by their members (see List 2005, 2008; List and Pettit 2011 for reviews). To illustrate how a group’s choice of an “aggregation procedure” can affect the rationality of its judgments, consider an expert panel which is summoned to decide whether Japan ought to extend the evacuation zone around the Fukushima nuclear power plant in the aftermath of its nuclear disaster. Let us assume that in the course of their deliberations, the experts have to take a collective stand on the following set of propositions on which they individually disagree: (p) The average level of radioactive radiation in some contaminated area A exceeds 100 mSv; (p → q) If the average level of radioactive radiation in A exceeds 100 mSv, then residents of that area have an increased risk of cancer; (q) Residents of A have an increased risk of cancer. The experts have further agreed to use majority voting to determine their judgment as a panel, with their individual judgments on the relevant issues being distributed as shown in table 8.1.
Table 8.1
A majoritarian inconsistency
|
p
|
p → q
|
q
|
|
Expert 1
|
True
|
True
|
True
|
|
Expert 2
|
True
|
False
|
False
|
|
Expert 3
|
False
|
True
|
False
|
|
Majority
|
True
|
True
|
False
|
The Fukushima example illustrates a situation—sometimes called a discursive dilemma (Pettit 2003)—in which the collective rationality of the group stands in direct conflict with the individual rationality of its members. As a result of adopting the seemingly sensible practice of majority voting, the group as a whole ends up endorsing an inconsistent set of propositions in its judgments, even though each member is consistent in his or her individual judgments. The discursive dilemma is a more general version of Condorcet’s more famous paradox of voting because it applies to any set of propositions rather than just preference rankings. Condorcet’s paradox shows that majority preferences can “cycle” (i.e., fail to be transitive), even though the preferences of individual voters do not. In both cases, the paradox is generated because different propositions are supported by different majorities, but the intersection between those majorities is not itself a majority in the group. The possibility of such scenarios demonstrates a principled upper bound on the rationality of groups.
A more general result, shown by List and Pettit (2002, 2004), is that no aggregation procedure which jointly satisfies the conditions of universality, anonymity, and systematicity can generate complete and consistent collective judgments.1 In this context, universality means that the procedure accepts as admissible input the domain of all logically possibly combinations of individual judgments, anonymity implies that the procedure gives equal weight to the judgments of all individuals, and systematicity requires that the collective judgment for each proposition depend only on individual judgments on that proposition and that it do so in a uniform pattern for all propositions. If we consider the satisfaction of all three conditions as a minimal requirement of rationality, no group can be perfectly rational in its judgments. However, by relaxing one or more of these conditions, groups can strategically circumvent their cognitive limitations to ensure that they produce collective judgments which satisfy at least a bounded kind of rationality (cf. List 2008).
First, by restricting the universality of its domain, a group might simply try to steer away from any combination of individual judgments that would lead to an undesirable lapse into collective inconsistency. This solution could work for groups in which the amount of disagreement among members is limited or if there are other mechanisms in place for reducing disagreement before a majority vote is taken (e.g., by going through iterated rounds of mandatory group deliberation). However, there is no guarantee that this method will work in general, especially since disagreements on evaluative issues are bound to arise and remain significant in many real-world cases. A second solution is to give up anonymity and adopt a kind of “dictatorial” procedure by which the group judgment is invariably determined by the judgment of some privileged member. Adopting a method of dictatorship plainly has the distinctive drawback that it entirely forgoes the epistemic benefits of socially distributed cognition, which involves the pooling of disparate but often complementary bits of information. A third solution is to treat certain propositions as “premises” and others as “conclusions” of a logical inference, thereby assigning different epistemic weights to the different propositions. This violates the principle of systematicity. As a result, the collective outcome of this method depends on which of two main types of aggregation procedures is adopted by the group (Pettit 2003; List 2006). Using a conclusion-centered procedure, a group would only take a single majority vote on the conclusion, thus withholding collective judgment on the premises. This procedure, which yields a collectively espoused view on q that is maximally responsive to individuals’ views on q, avoids collective inconsistency, but only at the price of violating not only systematicity but also completeness. Using a premise-centered procedure, a group would first form a collective judgment on each premise (e.g., by taking a majority vote) and then let logic decide its collective stance on the conclusion. In our Fukushima example, this would mean that the group collectively accepts that q, despite the fact that there is no majority among its members in their individual judgments that q. Pettit (2003) has argued that the choice of a premise-centered procedure most saliently represents a situation in which groups act as cognitive agents in their own right, and that purposive groups which are regularly confronted with discursive dilemmas will be under substantial pressure to “collectivize” reason in this manner in order to remain credible and effective promoters of their assumed purpose.
A fourth option, which maintains all three of the above conditions of rationality, is to permit incompleteness simply by refraining from issuing collective judgments on certain propositions for which there is too much disagreement among individual members (List 2006). For instance, a group might decide to make a judgment just in case its members unanimously endorse a certain proposition. A somewhat weaker method, which still falls short of completeness, would be to use supermajority voting. However, this is hardly an option if the judgment in question is precisely the issue for which the group has been designated to take a determinate stance, such as the question whether the evacuation zone around Fukushima ought to be extended or not.
In this context, it is worth pointing out that violations of anonymity, as they occur in the case of dictatorship, are not necessarily detrimental to group cognition. In particular, a group which is prepared to relax not only systematicity but also anonymity can often profit from employing a distributed premise-centered procedure (List 2008). In this case, the group not only assigns different epistemic weights to certain premises (as opposed to a conclusion) but also assigns different group members (“specialists”) to judge different subgroups of premises on its behalf. This can be an attractive way for a group to collectivize reason, while dodging the above impossibility result, if the cognitive labor is divided in accordance with the distribution of relevant expertise. On the other hand, the reliance on relatively small subsets of privileged members that inevitably follows from the violation of systematicity opens the door for individuals with an agenda to assume a disproportionate influence on the group judgment. This can be clearly seen when it comes to the choice of which propositions count as the relevant premises. For instance, a “lobbyist” group member who wants to avoid the extension of the evacuation zone around Fukushima at all costs would have an incentive to demand that the group ought to consider p and q as premises. Knowing that there is already a majority support for p, the “expert” lobbyist who judges q to be false can thus easily put pressure on the entire group (which follows a distributed premise-centered procedure) to reject the proposition p → q.
The Janus-faced nature of this option also demonstrates the sometimes neglected point that the agenda-setting influence of political parties, which is often considered as a dysfunctional if not outright immoral part of public policy-making, may in fact be vital to overcome Condorcet’s paradox as a central theoretical limitation of democracy. By endorsing certain candidates or policies while strategically sweeping others under the rug, political parties indirectly help to reduce the likelihood of situations in which the general public’s electoral choices would endlessly “cycle” and thus lead to irrational policy outcomes. When viewed from a purely epistemic perspective of collective decision making, political vices can in principle be cognitively virtuous.
Summing up our discussion, in what sense can we view the choice of an aggregation procedure as collective epistemic action that is performed by the group as a whole? First, the capacity of a group to function as a collective cognitive system is manifested in part by its ability to form aggregate judgments that adhere to certain standards of rationality. Pettit (2003) goes so far as to argue that groups can be considered as institutional persons if they actively bind themselves to the discipline of reason in ways that make them epistemically responsible for failures to achieve a certain rational unity in their collective judgments, intentions, and actions, at least in the relevant regards. Moreover, as the above example clearly shows, the rationality of a group with respect to its collective judgments is crucially determined by its aggregation procedure. Different aggregation procedures can lead to different collective cognitive outputs for the same combination of individual cognitive inputs. The judgments of a collectivity are thus logically discontinuous from the judgments of its members, even though they supervene on the latter if we also factor in the shared dispositions of individuals to adopt a certain aggregation procedure as members of the group (Pettit 2003). Second, the strategic choice of a group’s aggregation procedure is clearly an epistemic action. Its purpose is not to achieve some jointly espoused physical goal (e.g., to evacuate people around Fukushima) but to ameliorate its cognitive performance as a group by modifying the organizational structure on which it depends. In particular, the decision of a group to retain certain conditions of rationality while rejecting others from the theoretical perspective can be seen as a collective epistemic action to work around the constraints of the impossibility theorem on judgment aggregation. Unlike in Kirsh and Maglio’s original Tetris example, the environment that is epistemically manipulated here does not concern the relationship between an individual and an artifact, but the social.
Conclusion: Moving Beyond the “Social Manifestation” Thesis
Our discussion of collective epistemic agency has been framed against the backdrop of our thesis that human social groups can form cognitive systems in their own right. As collective cognitive systems, they include the cognitive systems of individual human beings among their proper parts. At least prima facie, such a claim goes beyond the SMT and is more naturally understood in terms of the GMT. What now remains to be clarified is the relationship between these two conceptions of socially distributed cognition, and whether human group cognition ought to count as a multilevel trait or a group-only trait in the aforementioned sense of Wilson (2001).
At certain times, Wilson (2004, 2005) has proffered the SMT as a kind of reductive analysis of group cognition. For instance, commenting on Hutchins’s (1995) celebrated analysis of ship navigation crews as collective cognitive systems, Wilson presses the point that “[t]he statement ‘the crew saw the oncoming ship and decided to change direction’ might be made true simply by individual-level psychological facts, together with other, nonpsychological facts about social organization” (2004, 291). In response, it may help to alleviate Wilson’s metaphysical qualms to note that our version of GMT does not deny that any group-level psychological facts are metaphysically determined and, thus, “made true” by the totality of the latter facts. This is because contrary to some older incarnations of the “group mind thesis,” we do not consider group cognition to reside in an ontologically emergent, preternatural realm (Theiner and O’Connor 2010). Group-level psychological properties could not be realized if it weren’t for the interactions among people who are capable of engaging in individual cognitive processes—processes which themselves are partly constitutive of the realization of the collective cognitive system. At a minimum, we can therefore grant the weak supervenience thesis that any two groups which are composed of the exact same members, participating in all the same social interactions (in a given environment) cannot differ in their group-level psychological properties. Consequently, as I shall now argue, it would be a mistake to pit the SMT against the GMT as if those two theses were necessarily in conflict. If human group cognition is a multilevel trait in the sense I have outlined, the GMT is presumably most plausibly defended in conjunction with the SMT.2
To drive this point home, let us consider a philosophically uncontroversial case of intelligent group-level behavior—the stupendous self-assembly of fire ants into a sturdy, waterproof raft that can stay afloat on water for months (Mlot, Tovey, and Hu 2011). We can see how this impressive display of social coordination is a colony-level adaptation in the native habitat of fire ants, the Brazilian rainforests, because it allows entire colonies to migrate collectively to drier land when their nest is being flooded. How is this amazing feat accomplished? It is fairly easy to see how a single ant can stay afloat simply due to its small size, its hydrophobic exoskeleton, and the right amount of surface water tension. However, it is far less clear how the ability to stay afloat propagates upward from single ants to a large, densely knit conglomeration of ants, clinging together for their lives with their mandibles, claws, and the adhesive pads on the ends of their legs. As the study reports, it turns out that ants assembled in raft formation are in fact more water-repellent as a whole than they are as individual ants. The ants on the bottom of the raft, which is partly submerged in water, create a solid textured surface with their hairy bodies entangled, while trapping a thin layer of air. Consistent with the Cassie-Baxter law, which says that increasing the roughness of a surface also increases its water-repellency, the enclosed layer of air plays a constitutive role for the capacity of the ant colony to float together as a unit. It prevents water from sticking, adds buoyancy to the raft, and enables the ants at the bottom to breathe. The same feature also accounts for the great resilience of the ant raft in turbulent waters. For instance, when the raft is poked with a twig and pushed underwater, the assembly of ants tightens up and becomes even more rigid, thereby trapping large bubbles of air which cause the submerged raft to bob back to the surface. Mlot et al. show how the coordinated self-assembly of the raft emerges from a random walk on the level of individual ants. During the construction process, an ant which happens to arrive at the current raft edge either turns around and keeps walking, or is forced to the bottom by other ants that are pushing it from behind and then walking on top of it—a little reminder that seemingly “altruistic” behavior often relies upon social coercion.
Why do we find it natural to consider the ant colony, in particular in its raft-shaped form, as an adaptive biological unit of its own? The answer, I suggest, is because the ants are causally coupled so as to form an integrated system with functional gains (Theiner, Allen, and Goldstone 2010). Following a related discussion by Wilson (2010), we can break down this complex notion as follows. First, two (or more) elements are coupled just in case there are reliable, two-way causal connections between them. For instance, in order to form a raft, the ants need to be capable of reversibly attaching to each other, and also of climbing over one another such that ants at the raft edge can be trapped by their neighbors, thus slowly expanding the base of the raft. Second, two (or more) coupled elements form an integrated system in situations in which they operate as a single causal whole within the causal nexus—with causes affecting the resultant system as a whole, and the activities of that system as a whole producing certain effects. In our case, the relevant group-level activity is the capacity of the colony to stay afloat in raft formation. Third, an integratively coupled system shows functional gain just when it either enhances the existing functions of its coupled parts or manifests novel functions as a whole relative to those possessed by any of its parts.
An example of the first kind of functional gain, which is manifested at the level of parts, would be certain types of behavior displayed by individual ants that we can broadly subsume under a raft-building capacity. An example would be the ability of individual ants to grasp onto one another using their claws, mandibles, and adhesive pads. The raft-building talents of individual fire ants are plausibly an instance of a “socially manifested” capacity in the sense of Wilson (2001). This is because it is at least conceivable that a perfect molecule-by-molecule duplicate of the body of a fire ant could have the same morphological structure but is not part of a colony that is collectively capable of aggregating into living rafts. Accordingly, we would not attribute a raft-building capacity to the behavior of this imagined doppelgänger of a real fire ant.
An example of the second kind of functional gain, which is manifested at the level of the group as a whole, is the increased water-repellency of the ant raft, which enables the colony to remain together and stay afloat. Note that the physical property of water-repellency as such is a multilevel trait that is manifested, at one and the same time, at the level of individual ants as well as at the level of the entire colony. However, importantly, the gain in water-repellency by the colony is not simply due to an unstructured aggregate of the water-repellency of single ants, in the sense in which the mass of the entire colony is an aggregate of the mass of individual ants. If the ants would scurry out of their nests in the event of a flood and try to escape individually, it might be sufficient for them to rely on the physics of their bodies (relative to the ecologically normal surface tension of the water) to stay afloat. However, this would mean that the colony would probably not be able to survive together as a superorganism, and the individual ants would thus be doomed to perish. The nifty raft-assembling mechanism by which this group-level capacity is physically realized involves, as we have seen, the stable, tightly knit base created by the entangled bottom ants, which causes the entrapment of a thin layer of air. Consequently, we can say that the group-level capacity of the ant colony to morph into a waterproof raft is partly constituted by the socially manifested raft-building capacity of single ants. Applying the words of Wilson (2001, S272) to our present example, we might say that the two levels are “metaphysically entwined.”
In conclusion, I would argue that the instances of human group cognition that we have discussed can be understood along the very same lines. In each of those cases, the simultaneous and mutual influence of people on each other in their collective performance of a cognitive task (e.g., decision, memory) makes it a moot issue—if not impossible—to single out any one individual member as the “prime cognizer” and treat the other members as mere cognition supporters. As the extent of the epistemically relevant social interactivity among the members of a group increases, so does the impetus to ascribe emergent cognitive properties to the group as a whole. In previous work (Theiner 2010; Theiner and O’Connor 2010; Theiner, Allen, and Goldstone 2010), we have argued that the relevant sense of emergence in this context, which is meant to capture the degree of organization dependence, can be defined as a failure of “aggregativity” in the sense of Wimsatt (1986). The coordinated division and functional integration of cognitive labor may be the result of explicit organizational decisions by some group members, as in some examples of collective epistemic actions that we have discussed. It is indeed plausible to think that the conscious, deliberate efforts by individuals that are directed at shaping the social organization of a group may be indispensable for the formation and maintenance of maximally cohesive “social integrates” (Pettit 2003) such as firms, parties, or courts of law to which we impute relatively sophisticated levels of collective agency and intentionality. However, perhaps more frequently, the cognitive benefits of collective information processing in groups can arise as an unintended “side effect” of repeated social interactions among individual agents—not unlike the self-assembly of ant rafts (Ball 2004; Goldstone and Gureckis 2009).
Acknowledgments
I wish to thank Linnda Caporael, James Griesemer, William Wimsatt, and Olle Blomberg for their insightful suggestions and helpful comments on earlier versions of this chapter.
Notes
References