How do forces for convergence like shared mental models coexist with divergence and divisiveness in teams?

What a shared mental model actually buys a team

A shared mental model is not agreement on every idea. It is a common way of organizing knowledge about the team's purpose, the nature of the work, and how members will coordinate [1]. In space analog missions, where small crews lived in isolation for 45 or 120 days, crews with more shared understanding scored better on both a creative thinking task and a problem-solving task [1]. That held across different stages of the mission, which suggests the benefit is not a one-time warm-up effect but something that tracks performance over time.

The same review literature treats shared mental models as central to how teams think together, describing them as part of the group-level thinking processes that let members anticipate each other's needs [2]. In human-AI teams, the concept is borrowed from human teaming precisely because similar mental models of task and team have been linked to better performance [3]. So the pull toward convergence is real: a team that shares a picture of the work can act without renegotiating everything.

Why starting out disagreeing can help a team

Divergence becomes useful when it triggers adaptation. In a laboratory experiment with 66 teams, members often had different understandings of how to use their information and communication technology. The more their initial mental models diverged, the more the teams engaged in adaptation: assessing the situation, formulating a plan, and executing it [4]. Those adaptation steps, in turn, were followed by convergence of mental models and a positive change in how the teams used the technology [4].

The order matters here. Divergence did not directly produce better outcomes; it worked through the team's sequential adaptation and the convergence that followed [4]. That is a different claim from saying disagreement is good in itself. It suggests that a gap in understanding can function as a signal that something needs to be worked out, provided the team has a way to work it out.

The balance shifts with mission timing and duration

The benefit of shared understanding is not uniform across a long mission. In the 45-day isolation crews, greater sharedness was associated with generating fewer ideas in the third quarter of the mission, but also with generating more novel, original ideas [1]. In the 120-day mission, the shared mental model mattered most in every quarter except the third for problem-solving [1]. The authors read this as evidence that both when a mission stage occurs and how long the mission lasts condition the link between shared understanding and performance [1].

That nuance is worth holding onto. It means a team can be highly aligned and still show a dip, or a change in the kind of output it produces, at particular points. Convergence is not a switch that stays equally helpful throughout.

What this means for a team trying to hold both

The evidence points to a workable stance rather than a choice between unity and dissent. Shared understanding of purpose and coordination is associated with better creative and problem-solving output [1], and teams can deliberately build it. At the same time, divergent starting points can be productive when they set off situation assessment, planning, and execution that bring the team toward convergence [4]. The risk is a team that converges early and stops reassessing, or one that stays divergent without any adaptation process to convert the gap into a shared plan.

Some of the surrounding questions are still open. Reviews of shared mental models in human-AI teams note that the construct is inconsistently defined and hard to measure, which limits how confidently its effects can be extended to mixed human-machine teams [3]. Reviews of virtual teams likewise flag measurement gaps [5]. For ordinary human teams, though, the pattern is consistent enough to act on: use convergence to coordinate, and treat divergence as a prompt to adapt rather than as a failure.

Sources used in this answer

Related research questions