Abstract
As society has come to rely on groups and technology to address many of its most challenging problems, there is a growing need to understand how technology-enabled, distributed, and dynamic collectives can be designed to solve a wide range of problems over time in the face of complex and changing environmental conditions—an ability we define as “collective intelligence.” We describe recent research on the Transaction Systems Model of Collective Intelligence (TSM-CI) that integrates literature from diverse areas of psychology to conceptualize the underpinnings of collective intelligence. The TSM-CI articulates the development and mutual adaptation of transactive memory, transactive attention, and transactive reasoning systems that together support the emergence and maintenance of collective intelligence. We also review related research on computational indicators of transactive-system functioning based on collaborative process behaviors that enable agent-based teammates to diagnose and potentially intervene to address developing issues. We conclude by discussing future directions in developing the TSM-CI to support research on developing collective human-machine intelligence and to identify ways to design technology to enhance it.
Keywords
Humans have had to function in groups and cooperate with others to enhance collective survival since the beginning of humankind (Morelli et al., 2014) because of their inability to possess all of the knowledge or resources they need to deal with the changing demands of their environment (Wilson, 1975). Traditionally, this cooperation developed via direct personal relationships in small groups, such as families or tribes (Hamilton, 1964), in which better coordination of resources to respond to changing demands led to increased survival. Fast forward to the 21st century, in which technology-enabled globalization has resulted in the use of groups in practically every sector of society, many involving contributors who have never met each other and, increasingly, technology-based teammates (Woolley, Gupta, & Glikson, 2023). Consequently, given the importance of groups in many areas of people’s lives, interest in the psychology of collectives and what enables this massively distributed cooperation to take place has grown significantly in recent decades (Mathieu et al., 2018).
One area of research that is highly relevant to understanding how these large distributed collaborations can coordinate effectively is the growing body of work on collective intelligence (CI). As work on the topic has expanded over the last few decades, at least two different streams have developed that use the term “collective intelligence” in qualitatively different ways. One stream uses the term as a noun to refer to the outcome or product of collaboration, such as an information repository like Wikipedia, a prediction based on a combination of forecasts, or an innovative solution produced by integrating the input of a large crowd, as in online contests or crowd-sourced science (Aristeidou & Herodotou, 2020; Kittur et al., 2009; Malone, 2018). Research in this vein focuses on the effect of how collaborators’ inputs are combined on the average output quality for all groups (Budescu & Chen, 2015; Centola, 2022; Kameda et al., 2022). A second stream of work uses the term to describe different groups’ capability level, building on the psychometric approach to intelligence in psychology (Spearman, 1904). Studies in this vein define CI as the ability of a group to perform a wide range of tasks or achieve a wide range of goals in different environments that vary in complexity (Gupta & Woolley, 2021; Legg & Hutter, 2007; Woolley et al., 2010). Related studies have examined how the characteristics of different groups—such as their composition or structure—account for variation in their CI (Aggarwal et al., 2019; Woolley et al., 2015).
Recent work has built on the latter perspective of CI as a description of general group capability and has integrated decades of research on individual and collective cognition across diverse areas of psychology to develop the Transactive Systems Model of Collective Intelligence (TSM-CI; Gupta & Woolley, 2021; Woolley, Gupta, & Glikson, 2023). The TSM-CI articulates how CI emerges and is sustained in social systems that vary in size and configuration, involving only humans or human–artificial intelligence (AI) collaboration. Among the goals of the TSM-CI is to integrate research and to guide the design of systems and technologies to enable higher CI (Gupta et al., 2023). We describe recent research contributing to the development and testing of TSM-CI and its potential contribution to a developing area of research on collective human-machine intelligence, or COHUMAIN (Gupta et al., 2023). After we review the research on the TSM-CI, we discuss potential future directions for further development.
Background
From individual intelligence to CI
Research on human intelligence, initiated more than a century ago, has been closely associated with the psychometric tradition for most of the history of psychology (Kail, 2000). Woolley and colleagues (2010) explored whether the psychometric concept of intelligence would generalize from describing capability in individuals to describing capability in groups. To examine this possibility, they recruited groups varying in size from two to five members to work together on a variety of tasks sampled from existing group-task taxonomies (Larson, 2010; McGrath, 1984; Steiner, 1972). In a factor analysis of all the groups’ scores, the researchers observed that the first factor accounted for 43% of the variance in performance across all the tasks, which is comparable with the 30% to 50% of variance typically explained by the first-factor batteries measuring individual intelligence (Chabris, 2007), supporting the conclusion of a general CI factor to describe a group’s capability (Woolley et al., 2010). Several studies have since replicated this finding and have demonstrated that measures of a group’s CI predict future group performance in a range of different environments (Aggarwal et al., 2019; Engel et al., 2014; Kim et al., 2017; Riedl et al., 2021).
An important related question concerns the basis of CI: What are the key ingredients or systems that enable CI to emerge and be sustained? Reviewing research on intelligence across different disciplines reveals some consistent themes suggesting that any intelligent system—whether biological, technological, or hybrid—needs to manage three essential functions: memory, attention, and reasoning (Deary, 2012; Hawkins & Blakeslee, 2004; Luria, 1973; Malone & Bernstein, 2015; Schweizer & Moosbrugger, 2004). In human cognition, a wealth of research supports the interdependent roles of memory, attention, and reasoning in supporting intelligence, such as work on the developmental cascades observed in normal child development, in which gains in attentional control enable advances in working memory, reasoning, and problem-solving (Fry & Hale, 1996; Masten & Cicchetti, 2010; Tourva & Spanoudis, 2020). Memory, attention, and reasoning functions also play a central role in work on AI, in which learned patterns stored in memory guide attention and enable adaptive reasoning and learning (Csaszar & Steinberger, 2022; Gugerty, 2006; Hawkins & Blakeslee, 2004; Young & Lewis, 1999). Indeed, many are surprised to learn that the concept of “working memory,” commonly associated with human cognition, was first introduced in work on AI by Newell and Simon (Miller et al., 1960; Newell & Simon, 1956) and only later generalized to psychology (Baddeley & Hitch, 1974), now recognized as playing a significant role in human intelligence (P. L. Ackerman et al., 2005).
Likewise, collective memory, attention, and reasoning play essential roles in CI in human and human-machine groups (Woolley, Gupta, & Glikson, 2023). However, as is common in many research areas, work on collective cognition has developed in parallel streams in different subfields, resulting in similar words being used to describe substantively different phenomena. For instance, work in cognitive and social psychology has defined collective memory (e.g., Vlasceanu et al., 2018) and shared attention (Shteynberg, 2015) according to whether the cognition of different group members is the same. By contrast, research in organizational psychology on collective attention (Woolley, Chow, et al., 2023) or transactive memory (Ren & Argote, 2011) has evaluated the strength of these collective cognitive systems according to the total capacity of information they can effectively manage, which is enhanced when the content of members’ cognition is mostly unique. On the basis of this latter perspective, collaborators in well-functioning systems focus on different areas of their shared work but also know others’ areas of responsibility to facilitate coordination.
An essential process supporting the development of collective cognition from individual cognition is “metacognition,” defined as people’s beliefs and knowledge about their own cognition and the cognition of others, including mental processes, states, and capacities (R. Ackerman & Thompson, 2017; Flavell, 1979). Early work on metacognition, such as studies by Flavell (1979), focused primarily on the role of metamemory in learning. Subsequent work has considered metacognitive components in other areas of cognition, such as work on metareasoning (R. Ackerman & Thompson, 2017) and meta-attention (Loper et al., 1982; Short & Weissberg-Benchell, 1989), all focused on how individuals reflect on their own cognitive processes independently and in relation to others. Metacognition plays a critical role in collective cognition, making it an essential foundation for the processes enabling collective intelligence.
TSM-CI
The TSM-CI articulates the emergence and mutual adaptation of three interrelated socio-cognitive systems underlying the emergence and maintenance of CI: transactive memory (TMS), transactive attention (TAS), and transactive reasoning (TRS) systems (Gupta et al., 2023; Gupta & Woolley, 2021; Woolley, Gupta, & Glikson, 2023; see Fig. 1). The intermember processes forming these systems are considered transactive because collaborators bring a particular set of cognitive resources that they combine and exchange with other collaborators to improve their collective capability to pursue mutually beneficial goals. Each of the three transactive systems underlying CI evolves through cycles of allocation, retrieval, and updating of the relevant cognitive and metacognitive resources, resulting in systems that optimize collective memory, attention, and reasoning. Below, we provide a brief overview and integration of the extant work related to the development of TMS, TAS, and TRS, respectively, and then discuss recent research that has examined how the three systems mutually coregulate to adapt to different environmental conditions.
Collective intelligence emerges from individual human and/or agent-based cognition (Level 1), which is guided by metacognition to form transactive-memory, transactive-attention, and transactive-reasoning systems (Level 2) that produce observable collaborative-process indicators (Level 3) to signal the functioning of the underlying systems.
TMS
Research on TMS, initiated by Wegner (1987), was based on the observed tendency in established couples to specialize in remembering different and complementary details for taking care of household tasks or recalling memories of past events together. Subsequent research observed similar tendencies for members of high-performing teams to specialize in remembering different details of shared work, and now dozens of studies have demonstrated the benefits of TMS for team learning and performance in a variety of settings (Ren & Argote, 2011; Yan et al., 2021). A TMS consists of members’ memory and metamemory, which is updated through interactions with others to facilitate allocating and retrieving knowledge to and from the most appropriate contributor. This process can unfold outside of conscious awareness and when operating effectively, results in expanded overall memory capacity resulting from collaborators’ storing different and complementary information.
More specifically, the development of a TMS involves three transactive processes that ensure effective utilization of members’ limited and distributed knowledge and skills stored in individual memory: (a) “Updating” of metamemory, or “who knows what,” occurs as the collective works on interdependent tasks and learns everyone’s competencies; (b) “allocation” of new information and tasks to members based on the understanding of expertise that has evolved through successive rounds of updating; and (c) “retrieval” of information and expertise that is made more efficient as the TMS develops (Brandon & Hollingshead, 2004; Wegner, 1987). With multiple updating, allocation, and retrieval episodes, members form shared beliefs about the extent to which they can credibly rely on a given member’s expertise, which can lead to increased specialization of members, a typical pattern in groups with a well-developed TMS (Brandon & Hollingshead, 2004; Ren & Argote, 2011).
The development and management of transactive memory can happen outside of members’ explicit awareness but can be facilitated by member diversity, particularly in areas relevant to joint work (Aggarwal et al., 2019). Although the majority of extant research on TMS development has focused primarily on human collaboration, recent studies have demonstrated it can also develop in the context of technology-mediated communication (Peltokorpi & Hood, 2019; Yan et al., 2021) and the availability of knowledge repositories and dashboards that identify expertise can facilitate the process, particularly when accompanied by direct communication between collaborators (Gupta & Woolley, 2018; Su, 2012; Yuan et al., 2010). Although technology could store all the information for the entire team, related research has demonstrated that relying too heavily on technology as a memory aid can be problematic, particularly if it displaces individual and collective cognition (Gupta & Woolley, 2018; Sparrow et al., 2011).
TAS
In addition to coordinating memory of relevant knowledge to accomplish joint goals, groups must also structure and allocate member attention. TAS is formed from individual attention, guided by meta-attention to combine via intermember, transactive processes that facilitate allocating and retrieving individual and collective attention (Gupta, 2022).
Whether in small face-to-face groups or larger distributed human-machine collaborations, the development of a TAS involves three intermember transactive processes that facilitate the efficient utilization of collective attention: (a) updating understanding of individual and collective focus areas and current demands on contributors’ attention, (b) allocation of one’s own attention as guided by the formal or informal structures and norms of the collective, and (c) retrieval of attention in response to urgent needs or coordinating changes in collective priorities and when synchronous attention is required. Groups with a robust attentional-retrieval mechanism exhibit organized patterns of synchronous attention, sometimes manifest as “burstiness,” in which longer stretches of independent work separate concentrated periods of interaction (Mayo & Woolley, 2021; Woolley, Chow, et al., 2023). As with TMS, the management of attentional processes can emerge organically in small and highly interdependent groups but typically requires more explicit structures in larger and less interdependent groups or networks to serve as a foundation (O’Leary et al., 2011). For example, although a small restaurant waitstaff might quickly self-organize to serve customers via more implicitly developed TAS, the waitstaff on a large cruise ship with multiple large dining rooms to serve simultaneously will require systems such as dashboards and written protocols to make the necessary signals visible and expected responses explicit to everyone to coordinate attention to tasks effectively. In addition to more standard and explicit systems to manage the larger group, smaller subgroups will manage attention to ancillary details among themselves via more implicit TAS-guided processes—such as noticing the relative workloads of different members or an urgent need of some customers and shifting their attention to help using the cues they have available.
It is also essential for collectives to find the appropriate balance between TMS-driven coordination, which allocates work according to members’ expertise, and TAS-driven coordination, which distributes work according to availability and efficiency. Relying too much on TMS-driven coordination can lead to bottlenecks if allocating tasks to the most knowledgeable contributor leads to uneven workloads, whereas TAS-driven coordination might increase efficiency by balancing workloads but at the expense of quality when not using members’ expertise (Gupta, 2022; Kaufmann et al., 2021). We theorize that this balance and the related trade-offs are governed by the TRS, as described in the next section.
TRS
Although TMS and TAS work together to ensure effective utilization of the available collective memory and attention resources, they do not guarantee that the collective is pursuing the right goals or that its members are highly motivated to go after the selected goals. For collaboration to remain viable, collaborators must perceive that their efforts pursuing goals with the group are more valuable than pursuing them on their own or elsewhere (Raveendran et al., 2020). Building on extant work on transactive-goal dynamics, integrated with research on negotiations and collective reasoning (Bacharach, 1989; Colman et al., 2008; Fitzsimons et al., 2015; Gupta & Woolley, 2021), researchers of recent work have described TRS as a dynamic system consisting of members’ knowledge of their own and others’ goals and the transactive processes enabling members to reason about and maximize joint rewards by negotiating and aligning collective goals (Gupta, 2022; Gupta & Woolley, 2021; Haan, 2023; Woolley, Gupta, & Glikson, 2023).
In parallel with TMS and TAS, the TRS also involves three intermember transactive processes that reduce experienced uncertainty by facilitating individual and collective reasoning: (a) updating of metareasoning related to self and others’ goals and motivations through observations of the environment and others’ responses to different opportunities (R. Ackerman & Thompson, 2017; Colman et al., 2008), (b) allocation of joint priorities occurring via negotiation of collective goals to ensure the pursuit of those that are most rewarding to individuals and the collective as a whole (Turan et al., 2013), and (c) retrieval of commitment to collective goals and associated effort in supporting their pursuit (Fitzsimons et al., 2015). When a group works together for a significant period, collaborators who remain engaged typically do so because they adopt a dense set of aligned goals, which results in higher effort and commitment and more collective resources to accomplish goals. More resources lead to a higher likelihood of success, increasing members’ goal persistence and commitment to the collective (Fitzsimons et al., 2015). As with TMS and TAS, the TRS can develop without explicit awareness, although discussing goals and confirming priorities help support TRS development and maintenance. Financial incentives and traditional performance-management goal-setting systems can serve as more formal mechanisms for establishing a foundation for TRS, particularly in larger groups and organizations. However, in collectively intelligent systems, traditional mechanisms will likely be augmented by TRS-guided development of shared values and priorities to achieve a high level of alignment and sustained commitment (Allen & Meyer, 1990; Buchanan, 1974). In contrast to the many technological tools that augment individual memory and attention, very few exist to help individuals identify their personal goals and priorities or align the goals and motivations of a group of collaborators. Given the importance of TRS development, this is an important area for future work to develop supporting technologies because ultimately, the goals identified by the TRS regulate the resources sought by the TMS and the priorities guiding the TAS for efficient execution.
Environmental conditions influencing CI
The mutual adaptation of the three transactive systems is central to the collective’s ability to accomplish two essential functions in all social systems: efficiency and maintenance (Arrow et al., 2001; Hackman, 1987; Taylor, 2012). The efficiency function is concerned with the effective use of resources to achieve the system’s goal, achieved via planning, organizing, and coordinating activities. The maintenance function monitors the system’s well-being via activities such as providing support, resolving conflict, and maintaining motivation. The relative importance of these two functions can vary depending on the context. For example, in a crisis, the need for a fast, coordinated response may make the efficiency function more important than the maintenance function. However, in an environment with heavy competition to attract collaborators and maintain their commitment, the maintenance function may become more critical in driving goal-setting and decision-making than the efficiency function.
Building on this reasoning, Gupta (2022) postulated that each transactive system contributes differently to a social system’s efficiency and maintenance functions. Specifically, Gupta theorized that TAS and TMS are key drivers of the efficiency function because they work in concert to coordinate collaborators’ inputs to use their knowledge and skills most effectively. On the other hand, the TRS provides the foundation for motivation and conflict resolution by aligning individual and collective goals, making it a key driver of the maintenance function. Furthermore, the collective decision to use efficiency-driven or maintenance-driven decision-making would be determined according to the demands from the environment as interpreted by the TRS, which is primarily responsible for guiding the selection and alignment of goals and priorities among collaborators and providing guidance to the TMS and TAS (Gupta, 2022; Gupta & Woolley, 2021).
Furthermore, suppose each transactive system is better suited to address different demands from the environment as theorized. In that case, one should observe that under different environmental conditions, high CI is associated with adaptation driven primarily by the transactive system(s) best equipped to respond. Testing whether this theorized relationship is reflected in a complex adaptive system’s behavior is challenging in an experiment or a field setting because capturing the complete set of inputs in an experiment or measuring them in the field would be practically impossible. However, such a question is well suited to computational methods, such as agent-based modeling, which allows researchers to examine the interrelated activities of a system and how they change over time according to changes in the environment (Kozlowski et al., 2013). Agent-based models (ABMs) enable researchers to articulate and test the underlying “process rules” that guide individual and social behavior and to validate that the rules implied by a theory result in the predicted patterns in simulated data (Carley, 1992). A goal in articulating the rules underlying the emergence of a phenomenon in constructing an ABM is not to be exhaustive but rather to determine the minimum rule specification necessary to produce the pattern, also referred to as “generative sufficiency” (Epstein, 1999; Kozlowski et al., 2013).
To explore whether the theorized adaptations occur in collectively intelligent systems, Gupta (2022) used an ABM to simulate the environmental changes and observe the behavior of the three transactive systems associated with high CI. According to the model, when there is a need for specialized expertise in a task environment, leading to high knowledge interdependence, the ABM results demonstrate that high CI is associated with TMS-led coordination (see Fig. 2). Moreover, when demand for production or output also increases, resulting in a larger workload, increased task interdependence, and high knowledge interdependence, the ABM results show that high CI is associated with coordination coregulated by both the TMS and TAS. Finally, when the environment in the model also includes agents with changing goals (and thus weak commitment, leading to high turnover), then CI is associated with strong contributions from all three systems—TMS, TAS, and TRS—underscoring the additional need for a strong maintenance function that balances the strong efficiency function when a collective is dealing with low commitment.
Schematic of agent-based modeling (ABM) results demonstrating the adaptation of the transactive-memory system (TMS), transactive-attention system (TAS), and transactive-reasoning systems (TRS) in response to changes in the environment to maintain collective intelligence. Based on Gupta (2022).
Collaborative-process indicators of CI
In modeling the TSM-CI, Gupta (2022) articulated explicit “rules” for how the systems emerge and coregulate as a basis for developing a model. However, as mentioned previously, these transactive systems operate in an undetectable manner to observers and often outside the explicit awareness of the participants themselves. Therefore, to provide insight into these processes, particularly for agent-based teammates, recent work has developed computational indicators of transactive-system functioning using observable collaborative-process behaviors (Gupta et al., 2023; Riedl et al., 2021; Woolley, Chow, et al., 2023).
Specifically, building on ideas from the normative model of team effectiveness (Hackman, 1987), recent studies have demonstrated three collaborative-process behaviors that can serve as diagnostic indicators of the functioning of transactive systems and, consequently, are predictive of CI: (a) the sufficiency of collective “effort,” (b) the efficiency of “task strategy,” and (c) the effective “use of knowledge and skill” of collaborators (Riedl et al., 2021; Zhao et al., 2023). Initial research on these processes using collaborators’ self-report ratings demonstrated their value as significant predictors of team performance (Hackman & O’Connor, 2004; Wageman et al., 2005). More recent studies have identified digital traces of behavior to use in computational measures of collaborative-process behavior (Riedl et al., 2021) that correlate with observers’ ratings of collaboration and differentiate teams that are high versus low in CI (Glikson et al., 2019; Ostrowski et al., 2022; Woolley, Chow, et al., 2023; Zhao et al., 2023). When used as real-time diagnostic indicators of CI, these metrics enable interventions by human leaders or agent-based teammates, such as an agent-based facilitator that prompts discussion or digital “nudges” that draw collaborators’ attention to weak collaborative processes (Gupta et al., in press). For instance, the group could receive real-time feedback via visual displays to make contributors’ engagement more transparent in remote collaboration and encourage everyone to contribute when their effort is low (Glikson et al., 2019). These collaborative-process indicators can also help identify, a priori, which groups would benefit from proactive interventions to improve task strategy, such as using structured, hybrid brainstorming techniques to generate creative ideas (Ostrowski et al., 2022).
Given the emergent and adaptive nature of CI, computational measures of collaborative processes, such as those presented here based on the TSM-CI framework, open up tremendous opportunities for leaders or even intelligent technological-support systems to evaluate collective functioning and provide corrective interventions when indicated (Gupta et al., 2023; Riedl et al., 2021; Zhao et al., 2023). Providing signals to enable agents to interpret human social behavior is essential to enabling true human-machine collaboration, a significant motivation driving work in emerging research areas such as COHUMAIN (Gupta et al., 2023).
Summary and Conclusions
Although the psychology of collectives has interested philosophers since antiquity (Aristotle & Keyt, 1995), recent decades have seen a rapid rise in the reliance on collaborative groups in every sector of society. As these collectives become larger and more loosely interconnected, views of groups need to be updated from systems structured to accomplish narrowly defined goals to their true nature as technology-enabled, fluid, complex, adaptive systems that can achieve a broad range of different and changing goals. Toward that end, the TSM-CI we have reviewed here has focused on the component systems underlying the emergence and adaptation necessary for CI, enabling groups to solve problems in various contexts. Although studies to date have focused on modeling and testing predictions based on the guiding logic of the TSM-CI, more research is needed to develop further theory and methods, including the role of technology, about all three of the transactive systems and in a fourth area related to their coregulation in response to the environment.
First, although there have been many developments in research on TMS in human systems over the past 3 decades (Yan et al., 2021), additional work is needed to refine the theory and methods for studying the role of technology in the process. Does an effective TMS that develops in a large, distributed, human-machine collective have the same characteristics as a TMS in a small human group? What new concepts and measures will be needed to understand how TMS develops and is maintained or when corrective intervention is necessary? Second, although research on TAS builds from a substantial body of work on attention at the individual and collective levels, the understanding of it as a transactive system is nascent compared with TMS. Future work can further develop theory and supporting measures of TAS in different contexts since its development is often implicit and happen outside of participants’ awareness (Gupta et al., in press), leading behavioral measures to be the most reliable indicators, as is true of studies of attention in general (Lindsay, 2020; Mancas, 2016).
Furthermore, researchers have only begun to examine the role of technology in supporting TMS or TAS. Technological support for individual memory is already quite common, and most individuals and groups regularly use tools such as calendars, task trackers, and search engines to externalize information and locate it as needed. Examining the effects of these and similar memory-related tools for TMS development can provide important insights into how best to use existing technology and ideas for developing new systems. By contrast, technology is often a significant source of difficulty in managing human attention. To effectively augment attention, systems need to help users manage interruptions and task switching and monitor their allocation of attention (Roda, 2010). Enabling systems to manage all those attentional processes will require users to give up a significant amount of personal agency and requires a high level of trust because any attempt to monitor the system’s decisions related to attention management would be distracting and defeat the purpose of deploying it (Glikson & Woolley, 2020). Additional research is needed to understand how to design those systems to make the right judgments when managing attention and to develop and maintain the trust necessary to be effective.
Third, the underpinnings of TRS are an integration of well-established research in behavioral economics, decision science, negotiations, and collaborative problem-solving, including consideration of the implicit and transactive nature of the collective goal dynamics underpinning collective reasoning (Fitzsimons et al., 2015). However, tools to support individual reasoning about goals and priorities are relatively underdeveloped compared with systems to enhance memory and attention. Moreover, although there has been some work on decision-support systems to scaffold collective deliberation and decision-making, there is very little to facilitate the gathering, integration, and prioritization necessary to facilitate collective reasoning—creating many opportunities for future work.
Fourth, several important open questions related to how the three transactive systems mutually regulate and adapt to changing environmental conditions remain. Because of the rich complexity of the systems individually and in combination, no one method or set of studies can fully capture all of the potential influences and related outcomes. For developing theory and testing predictions using frameworks such as the TSM-CI, it will be necessary to combine empirical studies with computational models and experiments. Although a few studies have made initial progress at this intersection (Gupta, 2022), there are many opportunities for future research to explore this fertile area for additional research.
Across these areas of research, articulating the TSM-CI into computational terms can open up even more opportunities to enhance the integration of machine-based “collaborators” into human teams. Some of the recent studies discussed demonstrated how collaborative-process indicators could serve as real-time signals of the functioning of the transactive systems to enable intelligent “agents” to intervene to improve CI in real-time (Zhao et al., 2023). Future research should continue developing machine-readable cues to convey social information to agent-based teammates to facilitate collaboration with their human counterparts. Such approaches will be essential to the emerging COHUMAIN domain (Gupta et al., 2023), which draws heavily on the psychology of collectives and aspires to enable the integration of AI into human collaboration in ways that will facilitate truly integrated CI.
