Welcome to CogSci Unpacked, an exciting blog series dedicated to summarizing academic papers from the Cognitive Science, a CSS Journal. Our goal is to bridge the gap between academia and the broader public, fostering a better understanding of cognitive science and making it accessible and relatable to all. If you’re curious to dive even deeper, we invite you to explore the full academic paper.
Imagine a soccer coach with a group of players and a season of training ahead. The coach could simply place each player in the positions where they are strongest today, or could think further ahead, spending the season training one player to become a defender, another to become a forward, and so on. This kind of prospective thinking recognizes that dividing labor—allowing each player to specialize in a particular role—benefits the team not only by matching team members to what they are best at, but by allowing them to improve “on-the-job” and prepare for the roles they will eventually be best suited for. Indeed, classical economic theories recognize that the improvement individuals make through practice is one of the key advantages of dividing labor (Smith, 1937; Yang & Ng, 1998). However, this prospective dimension has been largely overlooked in existing research on division of labor, which typically explains division of labor as a problem of matching static competences to team goals (e.g., Curioni, 2022; Goldstone, Andrade-Lotero, Hawkins, & Roberts, 2024; Xiang, Vélez, & Gershman, 2023).
Do people train collaborators to optimize long-term collaborative goals?
Our recent study (Tian, Gershman & Xiang, 2026) took a first step toward filling the gap. Across three pre-registered experiments (N=600), participants trained two military defense teams to counter two types of attacks (land and air). The long-term goals varied, such as whether it was feasible to deploy only one team for both attacks, or whether both teams needed to be deployed. The teams’ competences also varied, such as whether one team was better at both land and air defense or whether each team was better at one type of defense. Participants made a series of training decisions before assigning teams to roles in the final battle (see Figure 1).
Overall, participants trained collaborators by taking the long view—accounting for how each team’s competence would develop, and whether they could rise to meet the long-term goal—rather than simply training the team that was currently more competent, lagging behind, less versatile, or could improve the most. Consider a case where the Red team was more competent than the Blue team at both land and air defense. When the two attacks occurred at different times, one team could handle both, so participants trained and deployed the already-stronger Red team for both roles—there was no reason to divide labor. But when the two attacks occurred simultaneously, each team had to cover one attack, which meant the weaker Blue team would inevitably be responsible for one of them. Here, participants shifted their training toward Blue, building it up for the role it would eventually be forced to play. Crucially, this shift happened during training, before any role had been assigned—participants were not reacting to a division of labor already in place; they were anticipating one and preparing their collaborators for it.
This anticipatory reasoning even reshaped how labor was divided in the first place. In some scenarios, two different divisions of labor looked equally good based on the teams’ starting competences alone—yet participants reliably favored one over the other, because they were reasoning about how training would tip the balance. In other words, the expected payoff of training didn’t just follow from the division of labor; it helped determine what that division would be.
In summary, these findings show that division of labor involves a form of prospective social reasoning. People do not simply match existing competences to tasks; rather, they invest in collaborators’ growth to optimize long-term collective outcomes. This perspective provides a starting point for understanding how training supports effective division of labor, and how humans cultivate and ultimately benefit from one another’s evolving competences.
References
Curioni, A. (2022). What makes us act together? On the cognitive models supporting humans’ decisions for joint action. Frontiers in Integrative Neuroscience, 16, 900527.
Goldstone, R. L., Andrade-Lotero, E. J., Hawkins, R. D., & Roberts, M. E. (2024). The emergence of specialized roles within groups. Topics in Cognitive Science, 16(2), 257–281.
Smith, A. (1937). The wealth of nations (pp. 3–16). New York: Modern Library.
Tian, F., Gershman, S.J., & Xiang, Y. (2026). Training collaborators for effective division of labor. Cognitive Science, 50, e70247.
Xiang, Y., Vélez, N., & Gershman, S. J. (2023). Collaborative decision making is grounded in representations of other people’s competence and effort. Journal of Experimental Psychology: General, 152(6), 1565.
Yang, X., & Ng, S. (1998). Specialization and division of labour: A survey. In K. J. Arrow, Y.-K. Ng, & X. Yang (Eds.), Increasing Returns and Economic Analysis (pp. 3–63). Springer.