Smaldino rightly distinguishes genuinely emergent group organization from mere aggregation, pointing to the active collaboration of individuals with different capacities as the key form of cooperation. We offer two friendly but important extensions. We incorporate group cognition into Smaldino's framework; and we see the conditions under which structured differentiation is beneficial as more fragile than he acknowledges. These modifications encourage integration of work on the cultural evolution of group-level traits with substantial research traditions on distributed cognition, organizational psychology, and collaborative recall.

On standard views in cognitive science, cognition is strictly an individual-level achievement. “Social” cognition is thought to occur when people think about social phenomena or when social stimuli trigger cognitive processes. But cognitive processes are conceived in nonsocial terms. If group cognition is countenanced, it is understood atomistically, as the aggregate output of individual cognition plus social processes of combination. Little attention is paid to collaborative interdependence as the hallmark of emergent group-level cognition, as expressed in the Gestalt maxim that “the whole is greater than the sum of its parts.”

This attractive but puzzling concept is clarified in a large body of rigorous experimental research on group problem solving in social and organizational psychology (Laughlin Reference Laughlin2011) that is surprisingly neglected both in Smaldino's presentation and in cognitive science at large. In one key work, Steiner (Reference Steiner1966) distinguished five types of group tasks. In additive and compensatory tasks, group members do not interact in producing an outcome. In additive tasks, the group outcome is indeed the sum of the member contributions. In compensatory tasks, the group outcome is a statistical average of individual solutions. The proper aggregation of estimates or predictions can yield greater information gains than the sum of individual contributions (Bettencourt Reference Bettencourt2009), as in the “wisdom of crowds” (Surowiecki Reference Surowiecki2004). But such crowd intelligence is largely aggregative in the sense of Wimsatt (Reference Wimsatt, Grene, Donagan, Perovich and Wedin1986), as Smaldino argues.

Although conjunctive, disjunctive, and complementary tasks require interactions among group members, only the latter are collaborative in a richer sense. Conjunctive and disjunctive tasks are end points on a spectrum of how many group members must succeed individually for the group to succeed. A disjunctive task, for example, would be a group working on a sudoku, where the group succeeds if any of its members solves the puzzle. Here, the role of social interactions is a matter only of recognizing and adopting a solution found by any one member. In complementary group tasks, on the other hand, members coordinate and combine their diverse knowledge, abilities, and cognitive resources into a collective, organization-dependent outcome that no individual could have produced alone. Psychological processes studied from this group-level perspective include problem solving (Larson & Christensen Reference Larson and Christensen1993), collective induction (Laughlin & Hollingshead Reference Laughlin and Hollingshead1995), the development of transactive memory systems (Wegner Reference Wegner1986), and creativity (Hargadon & Bechky Reference Hargadon and Bechky2006).

Likewise, the “distributed cognition” framework studies collaborative, dynamically evolving work practices mediated by the use of tools and representational instruments and carried out in environments that provide a rich organizational structure (Hutchins Reference Hutchins1995b; Perry Reference Perry and Carroll2003; Sutton Reference Sutton and Menary2010). It borrows from traditional cognitive science an emphasis on processes of creating, transforming, and propagating representational states, but views them as part of larger cognitive ecologies that involve the coordination of resources across people, tools, and shared environments. Complex collaborative tasks that have been analyzed as distributed cognitive systems include maritime navigation crews (Hutchins Reference Hutchins1995b), emergency rescue management (Garbis & Waern Reference Garbis and Waern1999), theatrical practices in Elizabethan drama (Tribble Reference Tribble2005), bioengineering labs (Nersessian Reference Nersessian2006), and crime scene investigation (Baber et al. Reference Baber, Smith, Cross, Hunter and McMaster2006). Expanding Smaldino's framework, we argue that these cases exemplify emergent group-level cognition (Sutton et al. Reference Sutton, Harris, Keil and Barnier2010; Theiner Reference Theiner2013; Theiner & O'Connor Reference Theiner, O'Connor, Corradini and O'Connor2010).

Our concept of emergent group cognition differs from the “assembly bonus effect,” when “the group is able to achieve something collectively which could not have been achieved by any member working alone or by a combination of individual efforts” (Collins & Guetzkow Reference Collins and Guetzkow1964, p. 58; cf. Larson Reference Larson2010). Firstly, an assembly bonus effect can occur without emergent group cognition. The “wisdom of crowds” critically depends on the lack of collaborative interdependence. It requires, ideally, that individual decision makers are connected only through suitable information aggregators such as market pricing, but otherwise do not influence each other's judgments.

Secondly, emergent group cognition does not necessarily produce assembly bonus effects. This is why Smaldino's claim that “structured differentiation is often beneficial to group success” (sect. 3, para. 7), while appropriately correcting standard views, needs some qualification. For example, though shared remembering in dyads or groups is a ubiquitous human activity, experimental studies of collaborative recall find that groups often remember less than the sum of their parts (Harris et al. Reference Harris, Barnier and Sutton2013; Weldon & Bellinger Reference Weldon and Bellinger1997). The most common explanation for such collaborative inhibition is that hearing other people recall disrupts individuals' idiosyncratic mnemonic strategies (Basden et al. Reference Basden, Basden, Bryner and Thomas III1997). The fact that collaborative facilitation is surprisingly hard to find experimentally (but see Harris et al. Reference Harris, Keil, Sutton, Barnier and McIlwain2011; Meade et al. Reference Meade, Nokes and Morrow2009) suggests not that emergent group cognition does not occur, but that the conditions under which it is beneficial are surprisingly fragile. As Smaldino notes, the history of group organization matters, as does the structure of differentiated expertise. A further key factor is the fine-grained nature of the communicative interactions in active collaboration among group members (Sterelny Reference Sterelny2012; Sutton Reference Sutton2013). This point can be neglected in social combination approaches in small group research, which tend to be output- rather than process-oriented and concerned mostly with intellective tasks where comparison with traditional baseline models (e.g., truth-wins, better-than-best-member) makes sense. Apart from lacking ecological validity, such narrow focus can lead us to misconstrue the functions of real-world group cognition. An important function of shared remembering, for example, is to reinforce social bonds, by merging disparate memories into a stable rendering of shared past experiences (Barnier et al. Reference Barnier, Sutton, Harris and Wilson2008; Hirst & Manier Reference Hirst and Manier2008; Hirst & Echterhoff Reference Hirst and Echterhoff2012). Recognizing the beneficial effects of collaborative interdependence requires that we conceptualize “group success” more broadly.

This concept of emergent group cognition can centrally inform ecologically realistic studies of the coevolution of minds, groups, and cultures.