Varieties of Group Cognition, Theiner 2024

1. Introduction

This chapter reviews contemporary developments of the idea of group cognition. This discussion serves a two-fold purpose:

• By discussing how group cognition can be operationalized, Theiner seeks to retain some central theoretical insights of the “group mind” thesis without succumbing to its spooky metaphysics.
• Theiner seeks to provide a generalizable taxonomy that helps to make sense of the concepts and resources available to disparate research programs, from social psychology, biology, cognitive science, and philosophy.

During the late 19th and early 20th century, many social scientists employed the notorious “group mind” idiom to express the sensible idea that groups can function as the seats of cognition, intelligence, and agency in their own right (Allport 1968; Wilson 2004). In their quest to stress (rightly) that group phenomena are something “over and above” the sum of individual contributions, a fondness for vitalist metaphors led them to believe (wrongly) that genuine group cognition must be the result of tapping into individualistically inaccessible, “holistic” forces.

2. From individual to group cognition

This section is foundational scaffolding and builds the taxonomic machinery for organizing and understanding group cognition. The subsequent sections are applications of this scaffolding.

In his social psychology work on group productivity, Steiner (1966, 1972) developed a taxonomy of group tasks. Steiner’s question was practical: when do groups outperform individuals and why?

Steiner classified tasks along three dimensions:

• Divisibility: can the task be broken into subtasks and assigned to different people?
• Output type: does the task call for maximizing quantity (maximization) or matching some standard of quality (optimization)?
  • Laughlin (1980) refines this into a continuum from intellective tasks (demonstrably correct answers—math, logic) to judgmental tasks (no single correct answer—hiring decisions). This distinction recurs in §3.
• Combination scheme: how are individual contributions combined into a group output? Theiner layers atop the schemes a spectrum of cognitive interdependence. (Listed below from least to most cognitive interdependence.)
  • No interaction required (Passes Wimsatt’s test—not group cognition):
    • Additive: sum individual outputs (volunteers filling sandbags during a flood; admissions officers independently reviewing applications)
    • Compensatory: average individual judgments (crowd estimating the ox’s weight)
  • Minimal interaction required (Ambiguous wrt Wimsatt’s test):
    • Disjunctive: group succeeds if any one member succeeds (a search party for a lost hiker)
    • Conjunctive: group succeeds only if every member succeeds (hiking group, can only travel as fast as slowest member)
  • Substantial interaction required (Fails Wimsatt’s test—genuine group cognition):
    • Discretionary: group chooses its own combination method (jazz trio improvising)
    • Complementary: members pool diverse skills into a unified output none could produce alone (orchestra, crime investigation)

Process losses (Steiner): Actual productivity = potential productivity minus process losses.

• Two types: motivation losses (loafing, free riding) and coordination losses (failing to recognize or adopt a correct solution).
• Groups almost always underperform their potential. But, if the task is divisible: a conjunctive task can be converted into a complementary one by assigning subtasks to match members’ strengths. This conversion is itself an organizational achievement that can push groups past their expected potential.

Assembly bonus effect (Collins & Guetzkow 1964): A group produces an outcome no member could have produced alone and that couldn’t be produced by summing independent efforts. Lives at the complementary end of the spectrum — it’s the empirical payoff of high cognitive interdependence.

Theiner’s definition of group cognition: collaborative performance of cognitive tasks producing group-level outcomes, where the defining feature is collaborative interdependence — emergent modes of organization by which individual actions, ideas, and resources are coordinated into a group-level outcome. Not defined by the quality of the output, but by the organization that produces it.

Wimsatt’s test supplies the diagnostic for genuine emergence.

• Sand in a bag: rearrange the grains, remove some, split into two bags — total weight doesn’t care. That’s aggregative. An orchestra: rearrange the seating, remove the oboist, regroup strings and winds — qualitatively different performance. That’s emergent.
• Wimsatt’s test formalizes this. A property is merely aggregative if it is (1) invariant under rearrangement of parts, (2) qualitatively similar under addition/subtraction of parts, (3) invariant under regrouping, (4) unaffected by interactions among parts. Group cognition fails these conditions — the output depends on who’s there, where they sit in the process, how they’re grouped, and how they interact.
• Steiner gives the taxonomy of combination schemes. Wimsatt gives the test for which ones produce something genuinely emergent. That line is where group cognition begins.

3. Groups as decision-making units

3.1 Transactive memory (Wegner)

Transactive memory systems (Wegner 1986): Groups expand collective memory through division of mnemonic labor—cooperatively storing and recalling information among members.

• Crucially, the system requires transactive memories — higher-order meta-knowledge about who knows what and who to trust on what.

Collaborative inhibition: Groups remember more than their best single member, but often less than the same number of non-interacting individuals.

• Hearing others recall disrupts one’s own storage and retrieval strategies. This can be mitigated by strategies such as consensus-based decision schemes.

While most studies measure groups on amount or accuracy of recall, which rewards differentiated TMS—everyone remembering different things, it also serves as identity-consolidation—a shared narrative that reinforces social bonds.

TMS formation is driven by two processes (Hollingshead 2001):

• cognitive interdependence: how much each person’s contribution depends on others’
• convergent expectations: shared conception of who should do what. These map onto Theiner’s two-track account of group memory — differentiation for performance, assimilation for identity.

3.2 Collective information sampling (Stasser & Titus)

Hidden profile effect (Stasser & Titus 2003): Groups are biased toward discussing already-shared information and consistently fail to surface unshared information.

• Stronger for judgmental (e.g., hiring someone) than intellective (e.g., math) tasks, and worse when members feel they lack sufficient information.

The bias reflects the dual function of group discussion: informational and normative.

• Informationally, groups should sample unshared information.
• But the normative drive toward consensus, closure, and social validation counteracts this. Shared information serves as a proxy for competence: if you say what everyone already knows, you look credible.

Interventions that help: designating “experts” on specific topics, leveraging senior members’ higher social status, emphasizing dissent, priming counterfactual thinking, etc.

3.3 Collective induction (Laughlin)

Groups can actually think better together, not just remember.

Collective induction: cooperative search for generalizations or rules through cycles of hypothesis generation and testing. This is a divisible, complementary task—the kind where assembly bonuses can appear.

Laughlin’s card-sorting experiments (1999): Groups induce a rule partitioning playing cards into examples and non-examples. The task has both intellective and judgmental components—wrong hypotheses can be demonstrated wrong, but plausible ones can’t be demonstrated uniquely correct. With enough time and evidence, groups matched the best individual.

• Key finding: groups sometimes formulate emergent hypotheses—ones no member proposed individually.
• More evidence helped more than more hypotheses. Positive examples provided better evidence than negative ones.

Letter-to-number experiments (Laughlin et al. 2002, 2003): Groups of three outperformed the best individual in equivalent-sized sets. This is the assembly bonus realized.

• The task was highly intellective—correct answers were demonstrable through multiple reasoning methods (arithmetic, algebra, logic).
• Critical point: groups succeeded via complementary combination—integrating different reasoning strategies—not via disjunctive selection of the best member’s answer.
  • This is the empirical evidence that complementary interaction produces something qualitatively different from picking the best individual contribution.

4. Groups as distributed cognitive systems

Edwin Hutchins distributed cognition: cognition extends beyonds one’s brain, stopping where information flow stops. When a ship navigation crew works together the thinking is spread across the people, the tools, and the workspace.

§4. Groups as distributed cognitive systems (Hutchins)

§3 showed groups doing cognitive tasks through organizational mechanisms. §4 goes further—the group is a cognitive system. The move is to redefine what cognition is (information propagation, substrate-neutral) so that the group-level description is as legitimate as the individual-level one.

Distributed cognition (Hutchins 1995): Redefines cognitive processes as propagation of representational states across media—where “media” includes both internal representations (in heads) and external ones (speech, gestures, tools, displays).

• If cognition is information transformation, the substrate doesn’t matter.

Three dimensions of distribution (Hollan et al. 2000):

• Across people: Social organization is cognitive architecture—it determines how information flows.
• Across neural, bodily, and environmental resources: Tools and artifacts are parts of the cognitive system. The blind person’s cane, the biologist’s microscope, the mathematician’s calculator.
  • This is the extended cognition thesis (Clark 2008).
• Across time: Culture as cumulative cognitive scaffolding—each generation inherits cognitive resources from the prior.

Hutchins’ cognitive ethnography: Lab experiments strip away what matters for distributed cognition—the workspace, tools, informal communication, etc.

• Studies include ship navigation crews (Hutchins 1995), emergency response (Garbis & Waern 1999), Elizabethan theater (Tribble 2005), etc.

5. Groups as complex adaptive systems

§3 used designed procedures. §4 used distributed cognitive architecture. §5 introduces a qualitatively different mechanism: self-organization from simple local rules, with no centralized control or designed procedures at all.

Self-organization: Collective behavior emerges from large numbers of agents following simple local rules, without centralized control or designed procedures.

Ant foraging: An ant finds food, drops pheromones on the return path. Other ants are attracted to the trail, find the food, reinforce the trail. Positive feedback amplifies the signal. Negative feedback (pheromone evaporation, food depletion, forager availability) stabilizes it.

Flocking/schooling (birds, fish)—the paradigmatic case of direct interaction:

• Local rules: near-range repulsion (personal space), long-range attraction (group cohesion), alignment with nearest neighbors.
• Through behavioral coupling, the flock becomes a distributed perceptual system whose range exceeds any single member’s.
• Context-sensitive: under threat, tighter alignment increases collective vigilance (at the cost of false alarms). During migration, looser coupling dampens noise and individual errors.

Human complexity: Human cognitive sophistication means behavioral conventions emerge faster, in more varied settings, and are culturally sensitive—adding a layer of complexity absent in animal cases.

6. Group agency (List & Pettit)

§§3–5 operationalized group cognition empirically. §6 is the most purely philosophical section—it asks whether groups can be agents with genuine beliefs, desires, and rational coherence.

Judgment aggregation impossibility (List & Pettit 2002): Analogous to Arrow’s impossibility theorem for preferences. No aggregation function can simultaneously satisfy the below. Groups seeking coherent collective attitudes must relax at least one condition:

1. Universality: accepts any possible profile of member attitudes.
2. Rationality: produces consistent, complete group output.
3. Anonymity: gives all members equal weight.
4. Systematicity: group attitude on each issue depends only on members’ attitudes on that issue (independence), applied uniformly across all issues (neutrality).

List and Pettit’s solution: Drop systematicity (especially independence). This is the discursive dilemma—vote on premises, let conclusions follow logically, rather than voting on every proposition separately.

• This means group attitudes can’t be derived proposition-by-proposition from member attitudes. The group may end up endorsing positions most members individually reject.
• Dropping anonymity further allows distributed decision-making—different subgroups fix different judgments.

The upshot: Individual and group attitudes come apart in principled, logically necessary ways. This establishes the theoretical autonomy of group agency—group-level attitudes are not reducible to individual-level attitudes.

• The logic is: if group attitudes could always be derived from individual attitudes, then “group agency” would be shorthand for “what the individuals collectively think.” You wouldn’t need group-level explanations because individual-level explanations would do all the work. Instead, group agency is irreducible to individual attitudes.