The Outer Workings of Interacting Minds

Image above: Portraits from left to right of Nancy J. Cooke, Robert L. Goldstone, and Suparna Rajaram.

How collaborative remembering shapes individual and collective memory • The special role of specialized roles in group coordination • Discovering team cognition outside of the heads of team members • Toward collective cognition

Editor’s note: This column is written by speakers who will be featured in a plenary panel on Collective Cognition at the upcoming 2025 APS Convention in Washington, D.C. For more information on the convention, visit this page.

---

Virtually all of the work on fundamental cognitive processes, such as perception, attention, learning, memory, and problem-solving, has focused on internal mechanisms at the individual level. Even when studying social processes, participants in psychology experiments are often confined to their own isolated computer or survey and presented with social stimuli such as faces, behaviors, opinions, or stereotypes to judge.  

The tacit assumption that psychology seeks to understand individual minds has made work on the group dynamics of interacting minds relatively rare. And yet, many of the structures most important to humanity, such as relationships, family, governments, corporations, schools, religious communities, science teams, online interest forums, musical groups, and sports teams, owe their very nature to the dynamic patterns created by people simultaneously influencing each other. Physicists can’t understand large-scale patterns such as snowflakes, quartz crystals, and clouds simply by understanding the internal workings of particles. Instead, physicists have developed models of the interactions among the particles. Likewise, understanding group patterns requires psychologists to uncover and model the interactions across people. Fortunately, recent methodological and theoretical innovations have allowed psychologists to work out several of the mechanisms involved in creating group-level patterns, going beyond the inner workings of the human mind to reveal the outer workings that interrelate multiple minds. 

For the study of cognition, this leads to a number of questions: Is remembering in groups simply the sum of remembering alone, effecting no change in the individual or group-level memory? Is decision-making by a group the same as individual decision-making? If so, then why are there group-level decision biases that manifest? Is group problem solving just a faster version of individual problem solving? And how do group interactions and communication play a role in the product of the group, including changes to the individuals’ cognition? 

View past presidential columns

In this essay, we present three related explorations into group dynamics and processes that we view as key to understanding group cognition. In her work on social memory, Suparna Rajaram has developed an integrative framework to capture the different cognitive mechanisms that come into action during collaborative remembering and produce changes in both group-level and individual-level memory. Robert Goldstone has designed behavioral experiments and computational models to better understand the mechanisms by which the members of a group coordinate by taking on different roles. To understand how teams take on cognitive tasks, Nancy Cooke has developed a theory in which team interaction is central. 

How collaborative remembering shapes individual and collective memory 

Memories connect us with our past and shape our narratives and identity. We reminisce with friends and family about past experiences, we develop joint understandings of concepts and goals at school and work, and we build collective narratives of the past as communities and nations. Despite the ubiquity of social remembering, for decades psychological scientists have studied the nature of human memory by focusing on people performing memory tasks alone. 

This individual memory approach has many advantages. It allows us to uncover cognitive principles that govern how a person learns and remembers, while controlling for extraneous influences, including the influence of others. Indeed, this individual memory approach has yielded a wealth of findings and theories about human memory that we can leverage moving forward. 

Nonetheless, memory scientists have long acknowledged the importance of social remembering. For example, Bartlett (1932) explicitly emphasized it by studying the changes when one person recalls a story told to them by another person. Against this backdrop and decades of research devoted to studying individual memory, there is now a growing interest among cognitive scientists to systematically study social influences (Barnier & Sutton, 2008; Hirst & Rajaram, 2014; Roediger & Wertsch, 2022; Weldon, 2001), a paradigm shift that embodies the social memory approach (Rajaram, 2024). 

Watch a related webinar (available to APS members and webinar registrants): Science for Society: Smart Choices, Confidence, and Technology: The Science of Decision Making

In our work, we have described a dynamic interplay of specific cognitive processes that kick into action when people remember in groups compared to working alone (Rajaram & Pereira-Pasarin, 2010; Rajaram et al., 2022). Some group processes improve the amount and accuracy of memory. For example, one group member’s contribution may expose the other group members to something they had forgotten but will now remember going forward (re-exposure benefits); information recalled by one group member may cue others to recall something they might not have otherwise remembered (cross-cuing); group members may correct one another’s memory errors (error pruning).  

Counterintuitively, other group processes reduce recall. Collaborating groups recall less than an equal number of people recalling alone, a well-established deficit called collaborative inhibition (Weldon & Bellinger, 1997). We have found that some of this unrecalled information can remain inhibited, impairing later recognition memory (Barber et al., 2015). Other ways in which group recall impairs memory are more intuitive. For example, group recall suffers because members can pass on information they did not study earlier, producing memory contagion (Maswood & Rajaram, 2019; Meade & Roediger, 2002).  

These and other cognitive processes during group recall interact in dynamic ways that cannot occur in individual recall and produce group effects that cannot be inferred from individual performance alone. In this way, collaborative recall changes each member’s memory, and it changes group-level memory, making it more overlapping and collective, or more diverse and expansive, depending on the situation. Strikingly, when collective representations emerge, these representations embody overlaps not only in what people remember, that is, the contents of memory, but also in how people synchronize their recall, that is, the structures of collective memory (Greeley & Rajaram, 2023). As memory structures aid future learning, such collective organization can guide former collaborators towards similar learning, with important implications for education and teams. 

At the same time, the individual also influences the group. Group members bring to the task their own preexisting cognitive structures developed over the course of life, through education, information consumption, and prior social interactions. When people encounter new information, these preexisting cognitive structures filter learning and facilitate a new cognitive structure for this information. These idiosyncratic memory structures influence how each group member propagates information to shape the group’s memory.  

In brief, Rajaram and colleagues’ research on social remembering reveals a dynamic reciprocity between the collective and the individual (Rajaram, 2022). The nature of this dynamic interplay in groups and its impact on memory outcomes depends on many features of groups, for example, who is in the group, who talks to whom and how often, who knows what, how members interact to exchange information, what information is being shared (autobiographical? emotional? prone to false remembering?), and the task demands.  

“Studying collective cognition through the lens of collaborative remembering can thus reveal something fundamental, universal, and important about the nature of memory both at the individual level and at the collective level.”

Nancy J. cooke, Robert L. Goldstone, and Suparna Rajaram

These common characteristics of groups vary across situations; probing their power to shape memory reveals the consequences of collaborative remembering on group performance, on changes in individual memory, and on the emergence of collective memory. Studying collective cognition through the lens of collaborative remembering can thus reveal something fundamental, universal, and important about the nature of memory both at the individual level and at the collective level. 

The special role of specialized roles in group coordination 

When people think about what makes a group well-coordinated, they might imagine its members closely synchronizing their behaviors, as with soldiers marching in lockstep or dancers moving in unison. However, often a group is coordinated because its members adopt different, complementary roles. For example, surgeries are successful because the different members of the medical team assume specialized roles: scrub nurse, circulating nurse, anesthesiologist, and surgeon. In this case, these roles are predefined, and people dedicate thousands of training hours to honing specialized skills for the specific roles. 

“…early communication efforts pay continued dividends in terms of effective coordination.”

Nancy J. cooke, Robert L. Goldstone, and Suparna Rajaram

Impressively, even if a group of individuals is thrown together without predesignated expertise or roles, they will spontaneously organize themselves into roles. Undifferentiated members of a group will take it upon themselves to differentiate themselves, turning a mere group into a genuine, task-oriented team. One highly effective tool for achieving this is communication. People can explicitly make proposals for how they will divide up the work. Empirical results show that people who can freely communicate proposals, expectations, and promises regarding role-based resource management are more likely to come up with efficient, fair, and stable coordination schemes. In research that Goldstone has pursued with the Nobel Prize-winning economist Elinor Ostrom, groups of people forage on a two-dimensional terrain for virtual plant resources that slowly grow back after being harvested (Janssen et al., 2008). When people can vote to implement rule systems for divvying up the terrain into private properties to avoid overharvesting the plants to extinction, they collectively gather more resources compared to a condition that does not allow communication. Furthermore, having voted to implement property rights on an early round of the experiment, these groups continue to coordinate to gather more resources on later rounds even when property rights are no longer implemented. As such, early communication efforts pay continued dividends in terms of effective coordination. 

Coordination is greatly expedited by communication but does not always require it. Sometimes, people simultaneously adapting to their environment and to each other suffices for them to successfully coordinate even in the absence of communication. In a simple group-behavior paradigm designed to quantify the spontaneous emergence of roles, Roberts and Goldstone (2011) developed a game in which a mystery number between 51 and 100 is randomly generated by a computer and each group member submits their contribution between 0 and 50 without communication. The contributions are summed together and compared to the mystery number, with the same feedback given to each participant. “Your collective sum is too low [too high] by 7 [or whatever the exact amount is].” On each subsequent round, participants can adjust their contributions with the goal of having the group’s sum exactly match the mystery number. A group demonstration of this experiment can be run by selecting “Group Binary Search” at the Percepts and Concepts lab website. 

This game is challenging because it requires participants to coordinate on who will change their guess and by how much, without explicit communication. As the size of the group increases, so does the average number of rounds required. Roles in this game can be understood in terms of reactivity to feedback, with some participants dramatically increasing their contributions when told that the group was too low on the last round, and other participants not changing their contributions at all. Quantifying reactivity to feedback provides a useful operational definition of a role—a player adopting behavior that is consistent over rounds (e.g., reacting to feedback by a similar amount on each round) and differentiated from other players (e.g., if one player in a group reacts a lot to feedback, others will react little to avoid overcorrection). Equipped with this operationalization of roles, the results showed that (1) participants spontaneously differentiate themselves into different reactivity roles over the rounds within one game and become more consistent in their reactivity over the course of the five games that make up the entire experiment session, (2) participants in large groups show more differentiation into roles compared with those in small groups, and (3) groups that show more differentiation into roles are faster to converge onto the mystery number, particularly for larger groups. 

In another effort to understand how people learn to divide a task into components, dyads were tasked with finding a unicorn hidden somewhere under a grid of tiles (Andrade-Lotero & Goldstone, 2021). The two players converge on uncovering complementary sets of tiles, with different teams eventually attracted to splitting the grid top vs bottom, left vs right, or inside vs outside. Even though players almost never start with these strategies, dyads settle on these stable divisions by strategically complementing each other’s actions. 

Across these and other collective behavior paradigms, effective group coordination arises not from all members acting in unison, but from a division of labor that emerges during group interactions (Andrade-Lotero et al., 2022; Goldstone et al., 2024). The computational models that tend to match human group behavior the best and also solve a group’s task the best have mechanisms to (A) communicate intentions, plans, and proposals across group members, (B) adapt to feedback provided by the environment, (C) repulse a member’s behavior from others, (D) create plans at multiple temporal scales, and (E) infer other members’ knowledge and intentions from their behaviors. 

Discovering team cognition outside of the heads of team members 

In 1988 the USS Vincennes mistakenly shot down an Iranian Airbus killing all on board. The Navy launched programs to study the cause of the incident. One program examined the flawed decision-making of the team that shot down the civilian plane. A program called TADMUS: Tactical Decision Making Under Stress resulted (Cannon-Bowers & Salas, 1998). Cooke was approached by industrial–organizational psychologists working on this program (Eduardo Salas and colleagues) as they realized that they needed input from cognitive psychology. At that time, Cooke was working on the cognitive expertise of individuals and knowledge-elicitation methodologies. This was applied research at the individual level (Cooke, 1994). Questions quickly arose around “team cognition.” What is it? Where is it? How does it work? How can it be measured? How can it be improved? 

Early attempts to address these questions focused on the idea of shared mental models, adopted from the concept of individual mental models (Johnson-Laird, 1989). Researchers reasoned that teams, a special kind of group with differing roles and responsibilities, would consist of team members, each having a mental model of the task or team. The extent to which these models were shared or similar, the more effective the team cognition (Cannon-Bowers & Salas, 2001; DeChurch & Mesmer-Magus, 2010; Langan-Fox et al., 2000). Cooke and others set out to apply knowledge-elicitation methodologies to measuring individual mental models of teammates working in synthetic task environments and to examine their overlap or similarity (Cooke et al., 2000). This approach was aligned with the tacit assumption that cognition resides in the individual and not the group or team. 

However, research in Cooke’s lab and others yielded mixed empirical evidence for the idea of shared mental models. Teams would, for instance, improve their performance over time without concomitant convergence of mental models. Further, the idea of shared mental models in heterogeneous teams began to be questioned. Should we expect that knowledge of the task (or the mental model) is the same or even similar for a team that is heterogeneous (e.g., surgeon, nurse, anesthesiologist)? What about the fact that the model is a static snapshot in time even though teams make decisions and solve problems over time? And should we expect the union of very similar knowledge to be reflective of the team’s thinking? Further examining the data also started to reveal the importance of team interactions, often in the form of communication. When performance changed, mental models were not becoming more similar, but rather communication and coordination patterns were changing. Cross-training team members on the tasks of others did not produce superior performance as the shared-mental-model view would predict. However, disrupting or perturbing the teams’ interactions resulted in more adaptive, superior teams. These and other data led to the development of the theory of interactive team cognition (ITC; Cooke, 2015; Cooke et al., 2013). 

“…at least for action-oriented teams, the sharing of information through communication at the right time and to the right person (i.e., coordination), is essential for team effectiveness.”

Nancy J. cooke, Robert L. Goldstone, and Suparna Rajaram

ITC theory holds that team interactions—often in the form of explicit communications—are the foundation of team cognition. It assumes that team cognition is an activity, not a static property or product, and it is highly tied to the context of the task. It also assumes that team cognition is best measured and studied when the team is the unit of analysis. ITC theory has several advantages over shared mental models in addition to being more predictive of team performance. First, ITC theory requires that measurements focus on team interaction dynamics. This has resulted in some real-time, automated measures of team cognition that would not be possible if examining knowledge at the individual level (Cooke & Gorman, 2009; Gorman, Amazeen, & Cooke, 2010). In fact, whereas individual knowledge is not directly observable, team cognition is measured through interactions that are observable. These measures also enable us to measure team cognition when we have nonhuman team members (i.e., artificial intelligence) that interact with humans. ITC also suggests new ways of training teams (e.g., perturbation training; Gorman, Cooke, & Amazeen, 2010) and designing for teams. 

Does this mean that teams do not rely on individual cognition? No—in fact, individual cognition (taskwork, teamwork, knowledge, and skills) is foundational for teamwork to take place. In addition, there is certain information that should be held in common (e.g., task goals). However, at least for action-oriented teams, the sharing of information through communication at the right time and to the right person (i.e., coordination), is essential for team effectiveness. In many cases the cognition on the outside seems more critical to team cognition than the cognition inside each teammate.  

Toward collective cognition 

These three research lines converge on an understanding of cognition that transcends the individual human. The memories, search strategies, and problem-solving capabilities of groups are not deducible from their members’ independently considered capabilities. They also depend on interactions across the members, such as communication, intention-reading, interference, exposure, and alignment. A team’s flexible adaptation to its environment and tasks is shaped by—but also reciprocally shapes—its members’ cognitions. These bidirectional influences between the team and its individuals add complexity beyond standard individual-centric approaches to cognition but also offer the possibility that psychological scientists will be able to develop more sophisticated accounts of cognition than any of us could possibly have constructed on our own. 

Back to top

Feedback on this article? Email apsobserver@psychologicalscience.org or login to comment.