Interactive Team Cognition

Authors


should be sent to Nancy J. Cooke, Arizona State University Polytechnic, Cognitive Science and Engineering, TEIM, Santa Catalina Hall, 7271 E. Sonoran Arroyo Mall, Mesa, AZ 85212. E-mail: ncooke@asu.edu

Abstract

Cognition in work teams has been predominantly understood and explained in terms of shared cognition with a focus on the similarity of static knowledge structures across individual team members. Inspired by the current zeitgeist in cognitive science, as well as by empirical data and pragmatic concerns, we offer an alternative theory of team cognition. Interactive Team Cognition (ITC) theory posits that (1) team cognition is an activity, not a property or a product; (2) team cognition should be measured and studied at the team level; and (3) team cognition is inextricably tied to context. There are implications of ITC for theory building, modeling, measurement, and applications that make teams more effective performers.

1. Introduction

Work today is increasingly cognitive and decreasingly physical. It is often the case that task complexity precludes a single individual from accomplishing the task, as no individual has access to all of the information, cognitive skills, or time required. Indeed, there are few tasks in today’s technological society that require an individual to perform as an isolated unit, uncoupled from an ever-expanding world of information. Examples of task domains with high cognitive complexity range from coordination of medical care to military command-and-control to interdisciplinary scientific collaborations. Teams of individuals have been adopted as a solution for meeting the challenges of task complexity; from waging warfare to serving fast food, teams are ubiquitous. In this article, we use “team” to refer to two or more interdependent individuals that adaptively interact to reach a common goal (Salas, Dickinson, Converse, & Tannenbaum, 1992).

The term “cognition” used in the team context refers to cognitive processes or activities that occur at a team level. Like the cognitive processes of individuals, the cognitive processes of teams include learning, planning, reasoning, decision making, problem solving, remembering, designing, and assessing situations—like individuals, teams may or may not exhibit intelligent behavior. Unlike cognitive processes of individuals, however, many instances of team cognition can be readily observed. For example, the process of decision making in distributed teams (Massey, Montoya-Weiss, & Hung, 2003) and the distribution of responsibility for remembering certain information among business partners or spouses (i.e., transactive memory; Hollingshead, 1998) are both readily observable. In these instances, the cognitive processes carried out by teams involve individual, heterogeneous perspectives of an event, which require integration.

Under high levels of cognitive complexity, it is impossible for any individual to have complete awareness or to take on all perspectives of the task or situation. It is even less likely that all individuals would share a common perspective. Phenomenologically, the environment is the same for all members of a team, yet their perspectives are heterogeneous, because each individual has different knowledge, skills, histories, or even positions in physical space. This confers upon individual team members a unique perspective on the situation, event, or problem to be solved. However, as teams process a situation, problem, or event, team members must dynamically integrate those perspectives to more completely comprehend the situation, event, or problem (Gorman, Cooke, & Winner, 2006). Indeed, team cognition provides a context for individual behavior. In this light, understanding how teams process information, coordinate, and behave as a unit is increasingly important.

The focus of this article is on cognitive activity carried out by teams—team cognition, and the team, rather than the individual team members, as the foundation for observing and understanding team cognition. The theory described here pertains to the cognition of teams in distributed or co-located complex task environments (e.g., military command and control) in which team members are specialized, yet highly interdependent, performing tasks synchronously or asynchronously. It therefore extends work in collective behavior and joint action that focuses on groups of relatively independent members (Goldstone & Gureckis, 2009) or tasks that tap a common coding for simple cognitive or motor skills across people (e.g., Knoblich & Jordan, 2003; Sebanz & Knoblich, 2009; Sebanz, Knoblich, & Prinz, 2003). Unlike the work on common coding or joint action, however, we are concerned with team cognition, which consists of two or more people with diverse training and background knowledge that must be coordinated dynamically and adaptively to accomplish a task.

We posit that team member interaction, typically in the form of explicit communication (e.g., e-mail, phone, talking face-to-face), is team cognition. In this framework, team interaction is not an intervening variable between individual-level (team member) cognitive inputs and team-level performance outputs. Rather, teams are cognitive (dynamical) systems in which cognition emerges through interactions. This theory is supported by data and unfolds in terms of three premises: (1) team cognition is an activity, not a property or a product; (2) team cognition should be measured and studied at the team level, and (3) team cognition is inextricably tied to context. In the following sections, the importance of understanding and studying team cognition is first highlighted, followed by a description of the prevailing theoretical view, Shared Cognition, and empirical data at odds with this view, and then an emerging theory of team cognition as team member interaction is described.

2. The importance of understanding team cognition

Team cognition plays an important role in team performance or effectiveness and in the effectiveness of many sociotechnical systems (Cooke, Gorman, & Winner, 2007). There are several notable examples of catastrophic system failures that are tied to breakdowns in team cognition, such as the 1988 shooting down of an Iranian Airbus full of passengers by the USS Vincennes (Collyer & Malecki, 1998), the events leading to the Challenger shuttle disaster (Vaughan, 1996), and the untimely response to Hurricane Katrina (Leonard & Howitt, 2006). Each of these examples can be, at least partially, attributed to deficits in team cognition as evidenced in faulty or limited team interactions. For example, failures in team decision making can result from deficiencies in communication and coordination among team members. Similarly, instances of effective team cognition and successful team performance can be seen in the U.S. Airways Flight 1549 aviation accident, the so-called the Miracle on the Hudson, in which all passengers and crew survived a bird strike that compromised both engines in flight (McFadden, 2009) and in the teamwork evident in the preparation for flooding that occurred along the Red River of the North around the Fargo, ND, area in 2009. By formulating testable hypotheses from a theory of team cognition, researchers will be able to construct a theoretical basis for interpreting team successes and failures. A theory of team cognition will improve predictions of when and why successes and failures occur in complex sociotechnical systems and provide guidance on improving team cognition and ultimately team effectiveness.

A good theory of team cognition has a number of implications for science and practice. First, a firm theoretical underpinning of team cognition can help to improve understanding of cognition at the individual level. There are a number of research questions that concern connections between the team and individual levels of analysis (e.g., What cognitive capacities within individuals are necessary and sufficient for being a good teammate?) that would benefit from such a theory. In addition, an adequate understanding of the links between individual and team cognition should allow for the modeling of synthetic teammates that can interact with human teammates in various capacities (teammate, coach, assistant).

Second, there are a number of applications that rely on a solid understanding of team cognition (Cooke, Gorman, & Winner, 2007). Team training in cognitive arenas can benefit from better assessment of team cognition and diagnosis of deficits and strengths. Assessment methods should tie directly to measures of team cognition that are suggested by theory. Models and simulations, an outgrowth of observation, that involve the team level are also dependent on an adequate understanding of team cognition. A good understanding of team cognition will lead to good models and simulations of teams that are cognitively plausible and that predict team behavior. Armed with an understanding of team cognition, technology can be designed to facilitate teams engaged in cognitive activities.

In the upcoming sections, the predominant perspective on team cognition, which we label Shared Cognition, is discussed. This perspective is followed by a discussion of a series of team experiments within an Uninhabited Aerial Vehicle–Synthetic Task Environment. Results of these experiments cannot be explained or satisfactorily interpreted using the prevailing shared cognition approach and have led to the new theory of Interactive Team Cognition (ITC).

3. The shared cognition perspective on team cognition

Theoretically, the Shared Cognition perspective is heavily influenced by an information-processing model of individual cognition and is the predominant theoretical perspective on team cognition. This theoretical perspective, as described in this article, actually comprises many related concepts, theories, and methodological frameworks that have similar assumptions. In this section, we describe the features and assumptions common to this general perspective. Shared Cognition proponents extrapolate the information-processing framework for studying individual cognition to teams by treating the collective knowledge of team members as inputs subject to team process behaviors (e.g., communications), which ultimately affect team outcomes. The resulting model of team cognition research (the paradigm) follows an Input–Process–Output framework (I–P–O; Hackman, 1987). More important, proponents of this perspective posit team cognition as the shared knowledge of team members—the inputs—where shared is defined in some cases as common (overlapping) knowledge and in others as complementary knowledge.

The Shared Cognition perspective aggregates team cognition from individual mental models, making team members (rather than the team) the primary unit of analyses (Cooke, Gorman, & Rowe, 2009). A mental model is generally defined as a stored representation of the environment that allows an individual to describe, explain, and predict the environment (Rouse & Morris, 1986). The mental model construct has been promoted as the knowledge structure of interest within—and shared across—the individual team members.

Shared Cognition constructs such as Team Mental Models (TMM) and Shared Mental Models (SMM; Cannon-Bowers & Salas, 1990; Cannon-Bowers, Salas, & Converse, 1993) are conceptualized and measured in terms of mental model complementarity and overlap, respectively. SMMs, however, are not characterized as stored team-level representations, but rather as emergent states within team members’ mental models. DeChurch and Mesmer-Magus (2010b, p. 33) state that “Team cognition is an emergent state that refers to the manner in which knowledge important to team functioning is mentally organized, represented and distributed within the team and allows team members to anticipate and execute actions.” For example, to measure a TMM, knowledge is elicited from individual team members via a method such as conceptual sorting or relatedness ratings and then aggregated (Langan-Fox, Code, & Langfield-Smith, 2000). Subsequently, amounts of knowledge overlap, sharedness, and accuracy among team members are assessed and the results are related to team outcomes.

The idea behind the Shared Cognition perspective, as evidenced in the related methodology, is that individuals with similar and complementary mental models will be able to anticipate the needs of others. It is believed that through the development of a shared set of expectations, team performance is improved via nonverbal interaction, common ground, and implicit coordination by eliminating the need for overt interaction between teammates under high cognitive complexity or during stressful circumstances (Entin & Serfaty, 1999; cf. Gersick & Hackman, 1990). Thus, it is assumed that team members who have conceptual structures that closely resemble a referent or “expert” team structure with respect to positional and interpositional expectancies will perform well (Smith-Jentsch, Campbell, Milanovich, & Reynolds, 2001).

Similar to TMMs, team situation awareness, defined as the union of responsibility (i.e., overlapping plus complementary knowledge) for maintaining awareness of a dynamic environment, is hypothesized to be based on the team’s shared understanding of a specific situation (Bolstad & Endsley, 2003). These theoretical concepts have been extended to explain why shared knowledge among team members is important to team performance, what type of knowledge must be shared, and to further develop methodologies for measuring shared knowledge (Cannon-Bowers & Salas, 2001; Cooke, Salas, Cannon-Bowers, & Stout, 2000; Langan-Fox, Anglim, & Wilson, 2004; Langan-Fox et al., 2000).

Empirical support for the Shared Cognition perspective is mixed. Shared knowledge or understanding is sometimes predictive of team performance (DeChurch & Mesmer-Magnus, 2010a; Mathieu, Goodwin, Heffner, Salas, & Cannon-Bowers, 2000), but in other cases, it is not. For example, Stout, Cannon-Bowers, Salas, and Milanovich (1999) found that teams who made plans as a group (i.e., a team process behavior) had a higher degree of mental model similarity. However, Stout et al. (1999) also found that under high workload conditions, better planning influenced communication and performance independent of whether team members held a high level of shared knowledge about others’ informational requirements. In some cases, therefore, team process behaviors such as planning and communication seem to fully mediate the relationship between SMM and team performance (Marks, Sabella, Burke, & Zaccaro, 2002; Mathieu et al., 2000).

Despite these mixed results, interventions with the goal of enhancing team performance outcomes are typically geared toward mental model convergence. This is demonstrated in work that has explored SMM convergence through team member cross-training in which team members are explicitly trained for each other’s roles and responsibilities (Blickensderfer, Cannon-Bowers, & Salas, 1997; Cooke et al., 2003; Volpe, Cannon-Bowers, Salas, & Spector, 1996). The consensus is that cross-training is responsible for increasing SMM convergence among team members, which in turn leads to improved implicit coordination, and ultimately better team performance outcomes (Entin & Serfaty, 1999).

In addition to mixed empirical support, there are logical and pragmatic limitations of Shared Cognition. First, the focus on knowledge structure within and across team members, rather than interaction processes, methodologically restricts the Shared Cognition perspective to the individual as the unit of analysis. However, there is no straightforward mapping between individual team members and the team in action (but see Kozlowski & Klein, 2000). Furthermore, the focus on individuals’ cognitive structure lends itself to dealing with task dynamics in terms of static snapshots (often retrospectively reported), missing much of the actual team cognition that unfolds only over time as the situation unfolds.

Second, inherent in the Shared Cognition approach is the assumption that, cognitively speaking, the whole (team) is equal to the sum of the parts (individual team members); that is, team knowledge is equivalent to the collective knowledge of a team’s members. A physical analogy is the Ringelman effect, in which collective group performance in a rope pulling task is substantially less than the sum of the individual efforts (Latané, Williams, & Harkins, 1979). On the other hand, when individual team members have very different lifting strengths, team lifting strength may exceed the sum of individual lifting strengths (Lee, 2004). Certainly, these are physical demonstrations of the problem associated with equating teams to collections of individuals. Even in studies of collective intelligence; however, it is clear that information is not a linear aggregate of parts (Bettencourt, 2009).

Third, and related, Shared Cognition also tends to be coupled with the idea (and supported by the methodology) that the parts of a team—the team members—are cognitively homogeneous and thus similarity metrics are often used to operationalize shared knowledge. The assumption of homogeneity breaks down when teams consist of members with differing skills, knowledge, and abilities. Thus, similarity metrics do not reflect the knowledge of heterogeneous team members.

Fourth, the assumption that shared cognition is a positive indicator of team performance breaks down when teams become very large. In such settings, it is impossible for everyone to have shared understanding of the same information; it is really more important to coordinate heterogeneous knowledge. For example, in operational settings such as air traffic control operations and Air Operations Centers, hundreds of people can form a team, or teams of teams, to successfully accomplish goals without necessarily having a high level of overlapping knowledge. Aggregating across subcomponents cannot result in reasonable predictions of team cognition when the subcomponents are specialized.

Fifth, mismatches between relatively static shared cognition and dynamic task demands could lead to undesirable outcomes. In highly dynamic, high-stakes environments, team member interactions are more fluid and adaptable than individual or SMMs (Gorman et al., 2006). Mismatches, when they do occur, can lead to tragic outcomes, regardless of the individual expertise of team members, as seen, for example, in the untimely launch of the Space Shuttle Challenger (Vaughan, 1996) and the Operation Provide Comfort friendly fire incident (Snook, 2002).

The ITC theory differs from the Shared Cognition perspective in its focus on cognitive processes (i.e., interactions) at the team level as opposed to cognitive structure. ITC recognizes the contribution of individual team member knowledge to team cognition, but it does not posit an emergent knowledge state on the part of the team. Because knowledge at the individual level is considered a prerequisite for team cognition, ITC would predict a relationship between knowledge (operationalized as a SMM) and team performance; however, as teamwork develops, ITC predicts that team interactions should account for more of the variance in team performance than knowledge.

We do not take the Shared Cognition perspective lightly. Indeed, we started our investigation into team cognition using the I–P–O and Shared Cognition approach. However, we came to realize that aggregating individual knowledge (whether through composition or compilation; DeChurch & Mesmer-Magus, 2010b; Kozlowski & Klein, 2000), rather than taking team interaction as the fundamental unit of analysis, provided inconsistent and unsatisfying explanations of our own empirical results. ITC holds that team interactions aggregate individual knowledge in real time as needed, and thus, no knowledge aggregation or emergent team knowledge state is required. This is not to say that shared cognition has no explanatory merit, but that it may be better suited for certain kinds of teams (i.e., smaller, homogeneous teams) performing knowledge-oriented tasks.

In the next section, we describe an empirical paradigm that we used to test our initial Shared Cognition predictions and that, in the process, led us to an alternative theory of team cognition.

4. The uninhabited aerial vehicle–synthetic task environment

The UAV-STE (Uninhabited Aerial Vehicle–Synthetic Task Environment) provides a testbed for the study of team cognition for a three-person heterogeneous team (Cooke & Shope, 2004). In this testbed, three participants coordinate to “fly” a simulated drone plane in order to take photographs of reconnaissance targets. The UAV-STE task was modeled after the team task components of the United States Air Force Predator ground control station (Cooke & Shope, 2004). In the simulated task, three participants are each assigned to a different role: pilot, photographer, or navigator. Individuals are first trained on the tasks specific to their roles and, after attaining proficiency in their individual roles, come together to work as a team to complete multiple 40-minute reconnaissance missions in which the team photographs a series of stationary ground targets.

Each participant in the UAV-STE task is seated in front of two computer monitors that display unique role information, as well as common vehicle information (heading, speed, altitude). Team member interaction occurs over headsets using a push-to-talk intercom system for voice communication, which allows recording of speaker and listener identities, as well as timing. A number of team and individual measures have been designed and validated and are embedded in the task software and collected apart from the task (Cooke & Gorman, 2009). To gauge team performance, a composite outcome score is computed for teams at the end of each 40-minute mission based on number of targets successfully photographed, number of route violations, and amount of fuel and film used. In addition, individual- and team-level knowledge, team process, team situation awareness, and team coordination are measured. Many of the measures (interaction-based measures) rely on patterns of communication flow and content as indices of team coordination (Cooke & Gorman, 2009) and team situation awareness (Gorman et al., 2006). Data have been collected from eight experiments in the UAV-STE, many results of which are reported elsewhere (Cooke et al., 2004; Cooke, Gorman, Duran, & Taylor, 2007; Cooke, Gorman, Pedersen, et al., 2007; Cooke, Kiekel, & Helm, 2001; Cooke, Shope, & Kiekel, 2001). In the next section, we highlight results from experiments on team skill acquisition and retention, team experience, and team training in the UAV-STE that have led to an interactive theory of team cognition as an alternative to the Shared Cognition perspective. Overall, these results demonstrate the importance of team interactions and their ubiquitous influence on team performance.

5. Results from the UAV-STE and their implications

5.1. Acquisition and retention of team skill

Across a series of eight UAV-STE experiments, we have found that team skill acquisition, as reflected in changes to the team performance score (a composite score based on outcomes at the team level), follows the log-law of skill acquisition (Cooke, Gorman, Duran, Myers, & Andrews, in press; Cooke, Kiekel, et al., 2001; Cooke, Shope, et al., 2001; Duran, 2010). Fig. 1 exemplifies typical team learning in this setting. As teams gain experience in the UAV-STE, the team performance score increases in a log-linear fashion, reaching asymptotic levels of performance by the fourth mission, similar to the power law of learning (Newell & Rosenbloom, 1981). Furthermore, over the eight experiments, we observed improvements in team performance that are accompanied by improvements in team interactions (team process behaviors such as communication and coordination), but not individual, complementary, or shared knowledge. When knowledge changes do occur, they tend to take place in the earlier stages of team development (e.g., between missions 1 and 2; Duran, 2010). Taken together, these results suggest that team learning curves parallel those of individuals and that team skill acquisition in the UAV-STE involves changes in team process, as well as knowledge.

Figure 1.

 Mean three-person team performance and standard error (N = 8) across 10 simulated UAV-STE missions with breaks after Mission 3 and Mission 7 (from Cooke et al., in press).

Just as teams acquire skill, they can also lose it. Declines in team performance have been observed after a retention interval greater than 7 weeks long (Fig. 2; Cooke et al., in press; Cooke, Gorman, Pedersen, et al., 2007). Retention of team skill for the UAV-STE is best predicted not by individual skill loss or knowledge decay (forgetting) but by decrements in interaction-based measures, including team coordination and process behaviors (Cooke et al., in press). In other words, team interaction deficits are better predictors than knowledge or individual performance deficits of declines in team performance after a retention interval. This pattern is observed even though the team performance score (i.e., the criterion) shares some components with individual performance scores (see Cooke & Gorman, 2009, for an elaboration on differences between component scores and interaction-based measures).

Figure 2.

 Mean team performance and standard deviation for teams that returned for Session 3 (Missions 8–10) after a short or long interval. Mission 6 is used here as a baseline (from Cooke et al., in press).

In a skill-retention experiment, in which retention interval and team composition (i.e., teams remained intact or changed team membership after the retention interval) were manipulated, we observed an expected decline in team performance after the retention interval for teams who changed membership (Gorman & Cooke, 2011). But, unexpectedly, the mixed teams who suffered a performance decrement following the interval continued to improve in terms of interactions after that interval (e.g., they were better at dealing with change in the environment). The newly composed teams developed more flexible and stable interaction patterns (i.e., timing of information passing at target waypoints) after the break compared with intact teams whose interaction patterns did not change, remaining rigidly fixed, after the break (Gorman, Amazeen, & Cooke, 2010). We speculated that the newly composed teams benefited by being exposed to a wider repertoire of interaction patterns by virtue of the new team members. Those process changes ultimately translated into increased team effectiveness. Overall, these results demonstrate that teams dynamically acquire and lose skill over time and that a critical factor at the basis of these developments is team interaction, operationalized as observer ratings of team process, timing of information exchange, or adaptation to change.

5.2. Team training

Training and instruction are critical to learning, retention, and transfer of cognitive skill. A UAV-STE experiment was conducted to compare three types of training with the goal of training for skill retention and transfer to novel post-training conditions. The three training approaches were procedural training, cross-training, and perturbation training (Gorman, Cooke, & Amazeen, 2010). In procedural training, teams were trained to follow a rigid team coordination script on when to pass critical information at target waypoints, and deviations from the script were corrected through feedback. Members of cross-trained teams were given training in all positions, a commonly used procedure to promote mental model similarity or convergence and proposed to enhance implicit coordination (Blickensderfer et al., 1997; Cooke et al., 2003; Volpe et al., 1996). Finally, perturbation training is a form of process training, in which teams are presented with perturbations (i.e., brief disruptions) during task acquisition that force teams to coordinate in new ways. This latter manipulation was inspired by findings in the motor and verbal learning literature, suggesting that introducing practice variability to the learner leads to transfer of skill to novel, but related, post-training conditions (Schmidt & Bjork, 1992). In addition, this condition was inspired by the finding that mixing team members provided varied interaction experience, ultimately leading to flexible coordination patterns (Gorman, Cooke, et al., 2010).

Subsequent to training, and after a retention interval, teams in all conditions exhibited some degree of skill loss. However, perturbation-trained teams exhibited significantly better performance in critical test missions that introduced novel conditions requiring adaptation and performance under increased workload (Fig. 3; Gorman, Cooke, et al., 2010). Procedural training led to the least adaptive teams, who required a significantly longer amount of time to respond to novel changes in the task environment, and the benefits of cross-training (significantly greater shared knowledge) appear to have broken down as the task became nonroutine. By introducing coordination variability during task acquisition, interaction-based perturbation training led to flexible interaction processes that teams transferred to novel task conditions. More important to the present discussion, these results indicate that training focused on interactions can be more effective for learning transfer than training based on enhancing shared cognition.

Figure 3.

 Mean UAV team performance for each training condition over missions (from Gorman, Cooke, & Amazeen, 2010).

5.3. Experience

What happens to teams as they become highly experienced in a specific context? Does this experience transfer to related contexts? This question was addressed in a UAV-STE experiment that brought teams of three individuals who regularly performed command-and-control team tasks together (e.g., the Internet-based video game, Counter Strike) into the UAV-STE context to compare them to newly composed teams who had never worked together. Cooke, Gorman, Duran, and Taylor (2007) found that the teams who had prior experience in command-and-control-type tasks had superior team performance scores (acquisition, performing under high workload) than teams who had no prior experience working together (Fig. 4). Teams experienced working with each other in command-and-control-type tasks, but who had no direct experience with UAVs or the UAV-STE, surpassed asymptotic performance levels of the other teams within the first two missions, yet exhibited no superiority in terms of individual or shared knowledge. The divergence in performance resulted from differences in how teams interacted to process UAV-STE reconnaissance targets and not differences in the constructs associated with Shared Cognition (i.e., shared knowledge or mental models). Consequently, the performance superiority of experienced command-and-control teams cannot be explained in terms of shared cognition, but it can be attributed to interaction advantages that dynamically emerge over time by repetitively coordinating with each other over a wide range of problems in different command-and-control contexts.

Figure 4.

 Mean team performance across Mission by Team Experience with 95% confidence intervals. High workload means are also shown (High Wkld = high workload; from Cooke, Gorman, Duran, & Taylor, 2007).

6. Theory of interactive team cognition

Our experiments on teams suggest that team cognition is an emergent, dynamic activity that is not attributable to any one component of the team, nor the shared cognition of the team members, but to the team as a whole as it interacts in the face of a changing, uncertain environment. We submit that team cognition is team interaction. All cognitive activity at the team level requires some form of team interaction, although the reverse is not the case because physical interactions among team members would not necessarily constitute team cognition. ITC theory converges with psychological theory that takes an active, ecological stance on what constitutes primary psychological phenomena (Cooke et al., 2009) and with recent trends in cognitive science that allow cognition to occur outside of the head (e.g., distributed cognition; Hutchins, 1995a; radical embodiment; Chemero, 2009; activity theory; Engestrom, Miettinen, & Punamaki, 1999; Nardi, 1996). ITC theory posits three premises regarding team cognition: Team cognition (1) is an activity not a property or product; (2) should be measured and studied at the team level, and (3) is inextricably tied to context. In the remainder of this section, we will discuss each of these premises in turn.

6.1. Premise 1: Team cognition is an activity, not a property or a product

The roots of ITC theory may be traced to the functionalist psychology of the late 19th and early 20th centuries (e.g., Angell, 1907; Dewey, 1896; James, 1890/1950). Psychological functionalism involves the study of the utility of psychological processes as a person adapts to his or her changing environment (Green, 2009). William James (1890/1950) emphasized the stream of consciousness as primary in the study of psychological phenomenon, as opposed to breaking the stream apart to inspect its component properties. Because the value of any component property (concept) is dependent on its context within the stream, the inspection of any one component devoid of its historical and functional context has a proclivity to mislead the inspector, resulting in a version of the “psychologist’s fallacy” (James, 1890/1950). ITC theory proposes that team cognition exists in the dynamic flow of team member interaction, and retrospective accounts outside of functional and historical context (e.g., offline queries or introspective reports) do not adequately capture team cognition. We digress from team cognition (briefly) in order to discuss what we mean by cognition as activity.

Physical properties of cognitive systems may include brains (neurons and glial cells), sensory apparatuses (receptor cells; neural pathways), and functional properties, including memory, perception, etc. Although certain properties are understood to be properties of cognitive systems, those properties alone do not constitute cognition. We argue that cognition is an activity that is realized when the physical system (e.g., a nervous system) comes into active contact with the information in its environment, such that the cognitive system exhibits functional properties. Simply put, cognition is not a thing or a place in the body or the environment; it is an interaction between the two. This is not to deny the inner phenomena associated with cognition, such as memories and dreams. Just as cognitive artifacts linger in the environment (e.g., tools, language, books) so can cognitive artifacts linger in the body (e.g., memories). Nevertheless, these artifacts are retrospective and devoid of meaning without some original reference to their source: the dynamic intersection of body and environment.

Interestingly, the view that teams cannot have cognition is often based on the assumption that cognition is a thing that needs a location or a place in the brain, and this is problematic when there is not a team brain. We assert that team cognition is not a property of the individual team members or the products produced by the team; team cognition is the interactions of the team members, and this assertion is counter to traditional approaches to studying team cognition. For example, the I–P–O approach focuses on the relation between the input properties of the team members (e.g., a mental model from each team member) and the products of team member interaction (the output of the team; e.g., a decision or a plan). In the I–P–O framework, team member interaction is the process part of this equation, which is poorly linked to the input and output (Kozlowski & Ilgen, 2006); and when interactions are analyzed, they are not associated with cognition (e.g., Stachowski, Kaplan, & Waller, 2009). To the degree outputs do not measure up to the (potential) inputs, the mismatch may be attributed to “process loss” during the intervening interaction process (Steiner, 1972). Hence, when interactions are analyzed, they are often associated with individual processes that can interfere with the reconciliation of potential input/output relationships (e.g., mindguarding; Janis, 1972). In this way, the I-P-O framework presents something like a teleological approach in which the outputs we can actually observe at the team level should be contained in the cognitive properties of the individuals producing them. In the I–P–O framework, the output is effectively “encoded” in the inputs, but the inputs must be transmitted via noisy interaction processes; it is as if the intervening processes are the messy details of producing the correct output given an input. Counter to this teleological approach, ITC theory holds that team cognition is precisely the details of the intervening interaction processes, such that a new cognitive function is gained during the interaction process—namely team cognition.

Our premise is reminiscent of what Ryle (1949) called dispositional factors. For example, on its own, glass does not contain the property of brittleness any more than a hard floor contains the property of making things brittle. But glass has the disposition to shatter when it dynamically interacts—collides with—a hard surface such as a floor. However, attributing the behavior of “shattering” solely to the properties of the glass is an example of a category mistake (Ryle, 1949). Furthermore, the resulting shards of glass, although products of the emergent “brittleness property,” do not contain the brittleness property any more than the glass or floor does. In a similar way, we argue that team cognition may be characterized as a dispositional factor of individuals; one that emerges during the process of team interaction. Team members, themselves, do not contain the property of team cognition. Individually, they only fill out different a priori roles in a team task (e.g., to fill out a specified division of labor). When they interact, however, team members may be compelled to think in ways not defined in their a priori roles to meet team-level goals. In this way, to meet team-level goals, teams may exhibit cognitive properties that individual team members do not. Again, we hypothesize that this new, team cognition property emerges only from team member interaction and is not encoded a priori in team member roles (e.g., team member selection). Furthermore, we argue that this is rooted in the concept that cognition is not a priori a property of neurons and glial cells, but a person with a normal and healthy nervous system has the disposition to exhibit new cognitive functions as they functionally encounter the information in their environment. Furthermore, similar to the shards of glass in the example above, the products of cognitive activity, such as a plan or decision, do not contain cognition any more than the brain, or a person’s environment, contains cognition.

If team cognition is not contained inside the heads of team members, then how do we observe it? Based on the evidence described above, team cognition, as it relates to team outcomes, is rooted in team interaction. If team cognition is team interaction, as proposed in ITC, then it is directly observable in terms of dynamic communication and coordination patterns. From an ITC viewpoint, for example, a metric based on communication flow (i.e., who talks to whom) patterns among team members is an example of a measure of team cognition, whereas overlapping knowledge structures (assessed ex vivo) are, at best, indirect measures of team cognition. This is a critical distinction that needs to be made explicit. Assessing the knowledge of individuals and then taking the union of that knowledge may be important in some respects (e.g., assessing team potential), but it does not assess cognitive properties that emerge during team interaction. An analogy can be made to illustrate this point.

Great basketball players are generally tall and muscular and have long arms and long legs. But does that mean a person with those properties will be a great basketball player? Not necessarily. If a team is made up of members with the right kind of properties, does this mean they will be a successful team? No, because what each team member does affects what the other team members will do. The active component of team-level functioning—interaction—is cognitive function on the team level. In short, a team consisting of individual experts fails if they do not interact or those interactions are not coordinated with the changing environment.

To the degree a team must be able to process a novel situation, a team must be able to adapt its interaction patterns to meet changing environmental demands (Gorman, Amazeen, et al., 2010; Manser, Howard, & Gaba, 2008) or else become rigid and ineffective (see also “groupthink”; Janis, 1972). This involves team cognition of the first order because such interaction processes must scale with team-level goals rather than individual-level goals. For instance, Gorman, Amazeen, et al. (2010) identified long-range dependencies (viz. long-range correlations) in team coordination events over time, wherein interaction patterns unfolded over longer time scales (on the order of dozens of minutes) than the characteristic actions of team member inputs (on the order of minutes). The results indicated that teams with less rigid dependencies adapted their interactions to novel situations more successfully, although individuals were operating on vastly different timescales. Thus, there is an overriding temporal component to team cognition because team interactions unfold over time. It follows from this premise that team cognition is inherently dynamic. Thus, computational and/or mathematical methods for describing dynamic processes provide the means for examining these principles of changing coordination and can be contrasted with methods that rely on aggregate snapshots over long periods of task performance (see also Arrow, McGrath, & Berdahl, 2000).

To summarize, we argue that team cognition is not a property of the individual team members or the products produced by the team. Team cognition is the process of team members interacting to complete a cognitive task. Interestingly, the interactions between team members can be directly observed and subjected to analytical approaches developed for assessing change over time dynamics. Indeed, most of the action, when it comes to team cognition, is happening at the team level.

6.2. Premise 2: Team cognition should be measured and studied at the team level

Because many of their properties arise through component interactions, rather than a priori component-level properties, complex systems can exhibit behavior that is greater than the sum of their parts. Simon (1962) argued that complexity often takes the form of a hierarchy in both physical and social complex systems. For example, molecules contain atoms that contain the subatomic components protons, neutrons, and electrons, among many others. Similarly, social organizations such as businesses contain subcomponents such as departments that contain teams that contain individuals and possibly many other subcomponents. But a scientist must pinpoint an appropriate level of component interactions in the hierarchy to understand how the complex system of interest works, both top-down and bottom-up.

Our second premise argues that the appropriate level of analysis for understanding teams is, of course, the team level of behavior and not exclusively the teammate-knowledge level (i.e., aggregation of measures taken at the individual level), as traditionally studied in the Shared Cognition literature. Indeed, team components are important to team cognition, just as neurons are foundational to ganglion or cognitive subsystems (i.e., visual vs. auditory encoding, etc.) are important to understanding neural pathways the brain. However, to understand how teams process information, we must study teams as a system, rather than solely focusing on the subcomponents.

Interaction, a hallmark of team behavior, introduces variance that is unique to team activity. Interaction and related team behaviors such as coordination, communication, and team situation assessment inherently involve more than one individual and cannot be studied or measured meaningfully at the level of individual team member knowledge. Hence, ITC theory focuses on the variance that is found at the team level of analysis.

In terms of complexity, we assume that psychological phenomena at the team level do not uniquely reduce to psychological phenomena at the individual level. For example, the distributed nature of team cognition implies that similar individual-level activity can be recruited for very different team-level functions. Different team cognitive activities, such as team coordination with sequential dependencies, constitute different modes of cognitive activity, such as team collaboration of the brainstorming variety, although team members retain their same role on the team as the team engages each of these modes. In that case, the knowledge of team member roles would not be sufficient to distinguish between coordination versus collaboration because the difference is in the interactions. The idea that different team-level modalities (e.g., coordination vs. collaboration) differently constrain team member interactions means that team cognition is not only irreducible to individual knowledge, but that team cognition provides a changing context for individual-level behavior.

This premise also suggests that there are aspects of team cognition that are not accessible when we limit our study to the aggregate of individual team member cognition. The idea that team cognition is not equal to the sum of its parts has much in common with the concept of emergence in complex systems. Again, we will digress briefly from the primary topic of team cognition to consider a metaphor: emergent pattern formation and complexity in a physical system as a metaphor for complexity and emergence in team cognition.

The concept of emergence is that pattern formation at the system level is not determined by the aggregate properties of the system’s components. Rather, patterns emerge from the interaction of components under stable states of system behavior. For example, as oil in a dish is heated from below, there is a shift from a stable conduction state at low temperatures in which molecules collide randomly to a stable convection state at high temperatures in which molecules move in coordinated convection rolls to dissipate heat in the most stable fashion (Nicolis & Prigogine, 1989). Thus, as the conditions of the surrounding environment change, the molecular components are constrained to interact in novel ways to seek new, stable system-level patterns, though the execution of those patterns is not encoded in the molecules themselves.

Social psychologists and sport scientists have equally recognized that group or team behavior at the global level arises from nonlinear interactions at the local level; it is self-organized (Arrow et al., 2000; McGarry, Anderson, Wallace, Hughes, & Franks, 2002; Passos et al., 2008). Such complexity blurs the distinction between individual agency versus what emerges through interaction. Similarly, we suggest that team cognition emerges as team members interact in the context of a changing task environment. Similar to the heated oil, pattern formation at the team level is not “encoded” in the team members but is both driven by and drives team member interaction. This reciprocal cause and effect was termed reflexivity by Simon (1954), and in social systems, it is referred to as social reflexivity (Morgan, Morgan, & Ritter, 2010). Emergence does not imply that team cognition is an epiphenomenon, however. Emergent phenomena have causal efficacy that an epiphenomenon does not. Team cognition exerts downward causality by introducing constraints on team member interaction, while interaction at the individual level exerts upward causality by driving pattern formation at the team level. Thus, as an emergent phenomenon, team cognition emerges from and also constrains the interactive behavior of individual team members.

This premise implies that understanding cognition at the team level will aid in our understanding of behavioral constraints at the individual level. This can be conceptualized in terms of the “nesting” of levels of analysis in team cognition. In particular, the individual level is nested within the team level. Although not detailed here, the team level may also be nested within a multiteam system (DeChurch & Mathieu, 2007), nested within an organization level, and so forth. The most important concept here is that behavior at the individual level is constrained by emergent patterns at the team level. This is very similar to the concept of constraint hierarchies (Vicente, 1999) and the stratification of levels of activity and the concept is also echoed in Chen and Kanfer’s (2006) work on motivational processes in teams.

Newell’s (1990) timescale of human action (Table 1) is an example of a stratified nesting of human activity in terms of timescales of activity. The cognitive band is nested within the rational and social bands via being embedded within a larger timescale of human activity. We suggest, similarly, that individual team member cognitive activity is nested within larger timescales of team cognitive activity. Using the bands in Table 1 as a rough guide, the appropriate timescale of team cognition is certainly more than milliseconds but also almost certainly less than years. Conversely, the appropriate timescale of individual cognitive activity is almost certainly less than days. Table 1 gives no indication, however, of how different levels interact; that is, how lower levels simultaneously drive and are driven by higher levels. However, an understanding of the complementary aspects of levels—for example, how individual cognitive activity is temporally nested with team cognitive activity—would enrich our understanding of cognition in general. The multilevel modeling approach (Kozlowski & Klein, 2000) and arguments for “crossing levels of analysis” (Hackman, 2003, p. 919) are consistent with this perspective. Multiple levels of analysis and how they interact, therefore, are central concerns in ITC theory.

Table 1. Newell’s (1990) timescale of human action [from Newell, 1990, p. 122]
Scale (s)Time UnitsSystemWorld (Theory)
107Months Social Band
106Weeks 
105Days 
104HoursTaskRational Band
10310 minTask
102MinutesTask
10110 sUnit taskCognitive Band
1001 sOperations
10−1100 msDeliberate act
10−210 msNeural circuitBiological Band
10−31 msNeuron
10−410 μsOrganelle

Fig. 5 abstractly illustrates our concept of cross-level interaction. Three levels of analysis are shown: organization, team, and individual; the systems comprising each of these levels are illustrated. For example, the team level is nested within the organization level and is composed of a system comprising people interacting with technology and other people. Similarly, the individual level is nested within the team level and comprises a person interacting with technology. There are not only interactions within levels but interactions across levels. The cross-level interactions identify the simultaneous upward level influence of lower levels (e.g., individual) and downward causation from higher levels (e.g., organizational). Knowledge of the higher level constraints can inform our understanding of lower levels, as described above. Similarly, knowledge of lower level influences on the higher levels can help researchers explain the patterns of interaction that occur at the higher levels. The third column of Fig. 5 identifies some sources of variability at each level. Note that there are other influences on each level beyond the components depicted in Fig. 5. These contextual influences are described in the next section. For studying teams, it is important to recognize that although team cognition is not determined by aggregates of individual-level properties, there is some influence from the individual level (similarly, at the individual level there may be some influence from even lower level properties; e.g., a person’s physical or mental health). Most important for team cognition, however, it is not individual-level variability that is the primary driver of variability at the team level. Rather, there is variability specific to the team level that must be identified to better understand team-level phenomena. Although, generally speaking, levels can be arbitrarily defined, variability that is specific to the team level is primarily driven by team member interactions, which are therefore the focus of ITC theory. Nevertheless, these levels suit our purpose in defining ITC theory, just as the levels in Table 1 suit an individual theory.

Figure 5.

 Levels of analysis and their interactions; team level is nested within organization level and individual level is nested within team level, resulting in different sources of variability at each level.

The levels depicted in Fig. 5 make it clear that teams do not operate in a vacuum. They are influenced both by organizational and individual-level constraints from above and below, respectively (see also Vicente, 1999). These other levels make up part of the context in which teams are embedded. In addition, there is an even richer context of artifacts, culture, and social considerations that impact team cognition (e.g., Hutchins, 1995a; Nardi, 1996).

6.3. Premise 3: Team cognition is inextricably tied to context

Another important functionalist idea is that the flow of cognitive activity occurs within a historical and functionally meaningful context. To inspect cognition devoid of context, leads to an impoverished, and mistaken, interpretation of cognition. As the activity theorist James Wertsch decried, “the [alternative] assumption is that cultural and social issues can be incorporated as additional variables once the basic forms of mental functioning in the individual have been isolated and understood” (Wertsch, 1991, p. 85). Counter to making such an assumption, we argue that variance in cognitive activity cannot be isolated from context. This derives from the idea that variation in a psychological “state” results from the co-adaptation of variance due to both “internal” (e.g., person) and “contextual” (e.g., environmental) sources.

Accordingly, ITC theory has roots in perspectives such as ecological psychology, situated cognition, and activity theory that recognize the interaction between internal and contextual sources of variation as a fundamental unit for studying psychological phenomena. Gibson’s (1966, 1979) information-based theory of perception led to an ecological psychology that emphasized the dynamic animal–environment relationship as the basic unit of analysis, as opposed to the analysis of individual agents isolated from their environment. Also, Jean Lave’s research into situated cognition revealed that algebra skills of participants as gauged in a pristine classroom setting actually change significantly (improved) in a more familiar agent–environment coupling, for example, shopping for groceries (Lave, 1988). Along these same lines, team cognition unfolds in context and is, therefore, inextricably connected to that context. An ITC theory of team cognition must account for the ubiquitous surrounding context of team member interactions and a dynamic environment (Cooke & Gorman, 2009). In the UAV-STE task, for example, specific types of interactions are to be expected in the context of target waypoints, but those interactions are only meaningful in that context.

As a sociocultural theory of human behavior, activity theory (Engestrom et al., 1999; Nardi, 1996) provides additional insight into the contextual nature of mind and cognition. Activity theory identifies dynamic action and interaction of human agents-in-environment as the fundamental unit of psychological analysis. According to activity theory, the human mind is to be considered primarily by the work (i.e., activity) of a person in a society, which is understood to shape both the internal processes of the person and simultaneously the person’s external relations to his or her social niche. Indeed, from this perspective, the trajectory that human evolution has taken, including higher mental powers such as language and abstract thought, may derive from the uniquely human social activity of work organized around implements (e.g., bricks for building, food for cooking, weapons for hunting, etc.; Vygotsky, 1960; Vygotsky & Luria, 1992).

According to Vygotsky (1960), each higher psychological function appears according to a general genetic law; first as an element realized within the constraints of social interaction and then as an inner psychological process beyond actualized social constraints (Wertsch, 1991, 1994). This idea is perhaps most pronounced in Bakhtin’s (1986) linguistic conception of dialogicality as central to all-human activity, including introspective thought. According to this view, the utterance is neither wholly the property of a grammatical system nor of the speaker. Rather, utterances are inextricably imbued with social history and, therefore, “inhabited by the voices of others” (Cheyne & Tarulli, 1999, p. 17). In this regard, Bakhtin espoused “multivoicedness,” in that the utterances we speak, speak to, regard, defend, assert, etc., are with respect to all of those utterances that have come before it or will come after it. By speaking an utterance we make it our own, adding our own voice to the “multivoiced” sociocultural milieu of utterances which have been and are yet to be spoken. Similar to ITC, cognition is not localized in the present mental state of a cognitive agent, it must be interpreted in light of the past and future history of the social niche in which that agent operates.

In activity theory, the concepts of division of labor and mediational means are fundamental to socially reciprocated dialog: Individual agents and their means for social activity inexorably form a single unit of analysis (Wertsch, 1994), such that individuals and their contributions are shaped by the activity of doing work within a field of practice (Wenger, 1998).

The distributed cognition approach (Hutchins, 1995a) also emphasizes the surrounding context of ubiquitous team member interactions and artifacts in the dynamic task environment. For instance, social aspects of cognitive processes within the cockpit of a commercial airliner are highlighted by Hutchins (1995b) as critical to cognition at the team level. The distributed cognition perspective also posits that cognitive processes are not confined to the individual.

On the surface, ITC appears to be distributed cognition applied to teams; however, there are ways in which the two theoretical perspectives differ. First, the theory of distributed cognition applies the concepts of the information-processing paradigm (e.g., representations, processes) to distributed cognitive systems (Rogers, 2006). Essentially, cognitive systems are treated as large information-processing systems in Hutchins’ theory of distributed cognition, in which the critical characteristic is that information processing is now taking place within a system that comprises individuals, the history of their vocational niche, and their environment.

Whereas distributed cognition translates the classical constructs of cognitive science (i.e., input→process (representations)→output), ITC does not. Rather, ITC does not assert that team members receive inputs from the environment, represent those inputs, process them, and then output team behavior in a linear (feedback) process (e.g., Ilgen, Hollenbeck, Johnson, & Jundt, 2005). Instead, teams are viewed as cognitive (dynamical) systems in which team cognition emerges through interactions of heterogeneous individuals. As such, measurement at the team level is a critical component to understanding team functioning. By only focusing on the internal structure of team members and how they process inputs into outputs, and not taking these overt interactive behaviors into account, we are missing important information about the ability of teams to perform cognitive activities.

To summarize, team cognition unfolds in context, and there are implications for such a tight connection to context. First this implies that context that includes other team members and the task dynamics needs to be taken into account in the measurement of team cognition. Second, it greatly expands the notion of cognition beyond the individual human. Team cognition certainly requires cognition on the part of the individual, but team cognition ultimately occurs outside the head of the individual, namely, in the unfolding interactions of team members in a dynamic environment. Finally, the implication of studying teams as systems inextricably tied to their context is consistent with the team as the unit of analysis, as discussed under Premise 2.

7. Discussion

Teams are complex systems that exhibit cognitive phenomena such as making and executing plans, making decisions, acquiring and retaining skills, and solving problems. Teams process information as a unit and can provide output as a unit, somewhat analogous to the information processing attributed to individuals, but team cognitive processes are inherently interactive and dynamical, unfolding in real time and within a unique historical (team member dependent) context. In this article, we have described a theory of team cognition, Interactive Team Cognition, which stands in sharp contrast with the prevailing view that we have labeled Shared Cognition. According to ITC, and in contrast with Shared Cognition, team cognition is an activity that needs to be addressed at the team level and inextricably embedded in a surrounding context. A consideration of teams as cognitive entities, and their interactions within a rich and dynamic context, addresses empirical findings in team cognition that cannot be explained by a Shared Cognition view. Thus, ITC has the potential to lead to better predictions of team planning, team decision making, team problem solving, and team performance.

We emphasize that ITC, with its focus on team member interactions, does not deny the fact that team members have knowledge about the task and team that can be shared to varying extents. In addition, it is clear that without some prerequisite levels of knowledge, team members will not be effective and will likely have negative impacts on team performance. However, ITC goes beyond team knowledge by locating team cognition in team interactions and postulates that these interactions are cognitive processes that are more critically linked to team effectiveness than knowledge. That is, team members can have a suitable distribution of knowledge, but if team members do not interact or fail to coordinated effectively, then the team fails. Thus, the divide between ITC and Shared Cognition can be interpreted as one of emphasis (i.e., knowledge inputs vs. process), but this emphasis leads to a number of novel implications across multiple facets of cognitive science, including theory building, modeling, measurement, and application concerning team cognition. We describe the implications in these areas in the remaining sections.

7.1. Implications of ITC for theory building

There are a number of aspects of ITC that have implications for theoretical development in areas other than team cognition. First, the concept that team cognition is conceived as interactions that are largely observable through explicit communication opens a window through which to directly view cognitive processing. What is learned about cognitive processing at the team level through this direct inspection may have implications for theories of individual cognition which are not so transparent. For instance, can our understanding of individual situation assessment or planning be corroborated by the same processes carried out at the team level?

Also, even though ITC advocates a focus on the team level, it is acknowledged that this is only one of many nested levels. The individual and organizational levels are also impacted by and impact the team level (Fig. 5). The exact nature of the relationship between these levels thus represents an important research question. For instance, if individuals must dynamically adapt to continuous states of the other team members and the surrounding environment, then how is individual-level cognition constrained by team-level functions? In turn, the team and environment are also affected by individual behaviors. But what are the individual capacities required to be an effective team member? What is the exact nature of the relationship between the individual and team cognition? How do team cognitive processes affect individual cognitive processes? For example, how do coordination dynamics, such as long-range correlations (see Gorman, Amazeen, et al., 2010), affect memory retrievals or task strategies of individual team members? An added challenge is that, though they exhibit seemingly volitional action in the team context, individual team members are not necessarily aware of the dynamic, emergent, and continuous state of team cognitive processes, at least in a declarative sense (e.g., Gorman & Cooke, 2011). Answers to questions like these will advance cognitive theory by simultaneously building across multiple levels of analysis.

7.2. Implications for modeling

Computational models of individuals behaving within a team can be used to rigorously test a theory of team cognition (i.e., computational modeling), as can models of teams. In fact, there are a number of complementary approaches to modeling teams (Zachary, Bell, & Ryder, 2009). We elaborate here on the implications of ITC for modeling (1) cognition of an individual teammate and (2) dynamic interaction patterns of teams that unfold over time.

Regarding the modeling of teammates, or “synthetic teammates,” if the cognitive capacities at the teammate level of analysis are necessary and sufficient for producing adequate interactions between teammates, then developing synthetic agents with the appropriate cognitive capacities and placing them in all-agent or agent–human teams should produce a critical test for synthetically producing team-level cognitive phenomena. And, furthermore, coordination dynamics in all-agent or agent–human teams similar to those produced by all-human teams (i.e., team-level learning and retention) would provide additional feedback on the validity of the synthetic agent. Thus, computational modeling can simultaneously inform ITC and provide a test of ITC. More specifically, because ITC posits that team cognition is in the interactions of individual teammates, models of cognitive activity at the team level should be successful to the extent that they focus on interactions. In recent modeling work, it is precisely these interaction processes (e.g., language comprehension and production; situation assessment) that have been identified as those most needed to develop a computational cognitive model that acts as a full-fledged teammate (Ball, Myers, Heiberg, Cooke, Matessa, Freiman, & Rodgers, 2010).

Beyond computational modeling of basic interaction processes, there are other subtleties that ITC posits for effective teaming. For instance, effective coordination in the UAV-STE requires that teammates get the right information to the right person at the right time. This goes beyond language comprehension and generation and even an understanding of the task. Teammates need to know who needs what information, precisely when it is needed, and whether the information should be delivered as is or reframed. Timing is important so that teammates are not overwhelmed with too much information at once, are not interrupted in their workflow, and yet have the information, delivered in an appropriate way, that they need to do their tasks. Much of this subtle interaction competency is probably not stored in a knowledge repository but is an adaptive response to the interactions of fellow teammates (e.g., who is overloaded and individual workflow differences).

Recent social modeling efforts have focused on the rich context of other teammates in interactions and the mutual effects of teammates on one another, or social reflexivity (Morgan et al., 2010). According to Morgan et al., participation (vs. hesitation) is a social phenomenon that can be modeled “independent of explicit agent knowledge” (Morgan et al., 2010, p. 246). In this case, the computational cognitive model is able to display emergent group behavior through its focus on reflexivity of individual teammates. Not only does synthetic teammate work advance theories of individual and team cognition, but validated computational cognitive models acting as full-fledged teammates can be incorporated into trainers, operations, and team-based experiments.

Team cognition and the content of its associated social phenomena (e.g., reflexivity) unfold within a context that is both task specific and historically relevant. The content of a historical context provides the temporal context in which team cognition unfolds (i.e., exhibits dynamical hysteresis). Therefore, we further submit that team interaction—the primary mechanism for intelligent team behavior—is an inherently dynamic activity. Dynamical systems modeling offers a promising approach to capturing the evolving interaction patterns that propogate as teams think. Indeed, investigators of joint action have recently acknowledged the important role of dynamics and hysteresis in understanding the mechanisms of interpersonal coordination and the emergence of self-organization in groups and teams (Marsh, Richardson, & Schmidt, 2009; Moussaid, Garnier, Theraulaz, & Helbing, 2009; Shockley, Richardson, & Dale, 2009). For both joint action and team cognition, dynamical concepts and models can provide a methodological substrate with which to describe complex histories of interactions, which are fundamental to both of these phenomena.

Gorman, Amazeen, et al. (2010) recently applied principles and methods of nonlinear dynamics to assess changes in team membership status and their impact on team coordination dynamics and adaptability. That study illustrates the generality and power of the dynamical approach for studying team cognition: Nonlinear dynamic models used to describe the stabilization of an inherently unstable system (e.g., maintaining upright posture) were successful in detecting the nature of the difference in team interaction dynamics due to changes in team membership, whereas changes in the static measures of shared knowledge were not. Furthermore, several of the dynamical parameters used to examine team coordination dynamics in that study (e.g., the Hurst and Lyapanov exponents) succinctly describe the dynamic history of patterns of team coordination. Recent developments indicate that they have potential to be monitored in real time (e.g., Gorman, Cooke, Amazeen, & Fouse, 2012), opening the door for real-time intervention and training manipulations.

7.3. Implications for measurement

ITC implies a fundamental change not only in the conceptualization of team cognition but also how team cognition is measured and assessed. Traditional (shared cognition) measures focusing on static knowledge apart from the context of the task (e.g., card sorting tasks) are antithetical to the tenets of ITC. According to ITC, team cognition unfolds in real time and is, therefore, context-dependent. Specifically, ITC suggests that team cognition needs to be measured in the context of the unfolding task with the focus on activity occurring at the team level, not relatively static (ex situ) knowledge representations.

Measures of team process behavior that are based on expert opinion of targeted team behaviors in context (e.g., Brannick, Prince, Prince, & Salas, 1995) are somewhat in keeping with the ITC perspective, but to the extent they are based on a retrospective judgment across a sequence of events, they too may be relatively static. However, the rich nature of team behavior affords other approaches. Cooke and Gorman (2009) described a number of new measures based on the objective measurement of unfolding team interactions. In the context of the UAV-STE and similar tasks, team interactions tend to be largely explicit verbal communications. Thus, leveraging communication data and interaction analysis methods has significant potential for describing team interactions. Furthermore, it is clear that communication behavior is necessarily occurring in context and evolving over time, which is in keeping with the ITC. Indeed, Hutchins and Johnson (2009) have described communication as a “collective cognitive activity.” As an additional bonus, if communication data collection and analysis were to be automated (e.g., Foltz & Martin, 2009; Weil et al., 2008), then team cognition could be directly assessed, in real time, as it unfolds.

We have developed interaction-based measures for team coordination and team situation assessment (Cooke & Gorman, 2009). Team coordination is assessed by examining interactions that occur in the context of the scenario (Cooke & Gorman, 2009; Gorman, Amazeen, et al., 2010). In that context, coordination, in terms of information passing to and from teammates is expected to happen at certain points in the scenario (e.g., at and around target waypoints in the UAV-STE). Furthermore, constraints on information passing sequences and timings may be used to generate a normative model of event sequences (e.g., in the UAV-STE, the navigator provides target information to pilot; then the pilot and photographer negotiate vehicle and camera settings; then the photographer provides feedback that a good photo was taken). A coordination metric is then developed that describes the timing and sequence of these seminal coordination events. Dynamical systems analysis can then be applied to a series of coordination scores to model changes in a team’s coordination over time. Dynamical parameters can be used, for example, to quantify coordination flexibility and stability in response to various environmental and task perturbations (Gorman, Amazeen, et al., 2010).

Although coordination dynamics is not typically in the measurement toolbox of scientists who focus on shared cognition, the measurement of situation assessment on the part of the team (or team situation awareness—a focal construct in the shared cognition literature) can be defined in new ways using this approach. Team situation awareness has been traditionally measured in a knowledge-based way (i.e., team members’ shared knowledge of the situation). ITC, however, is at odds with team situation awareness measured as static snapshots of team knowledge of the situation. Instead, team situation awareness is considered the timely and adaptive responding on the part of the team—via their interactions—to detect and accommodate unexpected changes in the task environment (Gorman et al., 2006).

Accordingly, to measure team situation assessment with an ITC focus, we have examined team interactions that occur in the presence of a critical change in the environment that has the potential to threaten future outcomes. For team situation assessment, there needs to be (1) joint (two or more team members) perception of the change (but not necessarily by everyone on the team, which would often be unnecessary and inefficient); (2) coordinated interpretation and perception of the change (i.e., team members communicate to “connect the dots” to relate their different perspectives on the change); and (3) coordinated action is taken to overcome future negative impacts of that change by one or more team members. Therefore, assessment of team situation awareness involves observation of team behavior and communication in the face of unexpected change, namely environmental uncertainty. For effective team situation assessment, it is not only critical that teams correctly assess the state of the environment and take action, but how this is accomplished (how many involved, timing, and coordination of information) is just as important. The ITC approach, accordingly, focuses on explicit, observable coordination on the part of the team within a dynamic context.

7.4. Implications for application

What does the ITC perspective have to say about applications of team cognition? Focusing on interactions rather than team member knowledge affords a new set of approaches to problems such as lack of team situation awareness and the need for virtual collaboration. Applications that focus on facilitating team member interactions and the timely and adaptive sharing of information are predicted by ITC to be more effective than those that distribute content to team members (e.g., a common operating picture) or attempts to present more information to more team members (see also Bearman, Paletz, Orasanu, & Thomas, 2010). ITC implies that not all team members need all information, but instead specialization with well-coordinated information passing (i.e., an effective transactive memory system; Hollingshead, 1998) is what is essential for team effectiveness.

The measurement of team interaction by monitoring communications or interactions that are mediated by information technology also opens the door to real-time monitoring of teams. Coordination patterns can be monitored, and when anomalies are identified, interventions to further assess or improve the situation can be triggered.

Considerations of ITC applied to training arise from the results of the aforementioned research which demonstrated the importance of widening experience with other teammates to foster team flexibility when unforeseen circumstances occur. This kind of “perturbation” training (Gorman, Cooke, et al., 2010) can be contrasted to traditional forms of knowledge-based training, such as cross-training, which is inspired by shared cognition, or rote procedural training (i.e., drills), which is inspired by traditional behavioral learning paradigms that value repetition to acquire new skill. For example, one way to ensure accurate SMM across all team members is to cross train all team members on all tasks. However, this becomes impractical fiscally (and functionally; see above) as teams grow in size and task diversity. ITC and associated approaches, like perturbation training, offer alternatives to these traditional approaches. Furthermore, synthetic teammates can also be enlisted to serve a training function and to “model” behavior of appropriate coordination dynamics in a team setting.

7.5. Limitations

ITC is proposed as a useful way to conceptualize team cognition that has a number of implications for theory building, modeling, measurement, and application. Although supported by data, it is not without limitations and boundary conditions. First, the theory is meant to apply to complex team tasks in which members are heterogeneous, yet interdependent. It would therefore not apply to small group decision making (e.g., juries or management teams). That said, most of our data supporting this theory have been collected in a single task context (team UAV control with mostly co-located teams and synchronous interactions) and questions of theoretical extensibility beyond this setting await ongoing experimentation. This experimentation is taking place in new team contexts such as mission planning, software design, emergency medical procedures, cyber defense, and distributed (team) perceptual-motor coordination in remote environments.

8. Conclusion

Teams engage in cognitive activity as a unit, and this activity extends beyond the knowledge that each team member carries within his or her head. As a cognitive function, team cognition is located in the interactions among team members rather than the static properties of their shared knowledge structure. We contend that this is cognitive processing at the team level. In this article, we have argued that team cognition is an activity and should be measured and studied at the team level in the context of the team environment, as the team task unfolds. This view is distinct from the Shared Cognition perspective and, indeed, the general perspective gained from much of the group and organizational literature. It is aligned, however, with recent perspectives in cognitive science such as embodied cognition, situated action, joint action, and dynamical systems theory, as well as the tenets of ecological psychology. Furthermore, ITC accounts for a number of empirical findings that Shared Cognition does not, and it is also scalable to large heterogeneous teams common in today’s workplace. There are implications of ITC for theory building, modeling, measurement, and applications that make teams more effective thinkers. We believe the prospect of increased effectiveness in today’s cognitively demanding work environments will be improved through the realization of these implications.

Acknowledgments

We thank numerous students who have assisted with data collection and analysis through the years, including Christy Caballero, Olena Connor, Janie DeJoode, Pat Fitzgerald, Rebecca Keith, Harry Pedersen, F. Eric Robinson, Leah Rowe, Amanda Taylor, and Jennifer Winner. We also thank our colleagues, Nia Amazeen, Dee Andrews, Jerry Ball, Brian Bell, Peter Foltz, Kevin Gluck, Preston Kiekel, Eduardo Salas, and Steven Shope, who have provided the inspiration for our theoretical perspective and for sponsors of the associated empirical work: Air Force Office of Scientific Research, Air Force Research Laboratory, and the Office of Naval Research.

Ancillary