Let’s Think Together! Assessing Shared Mental Models, Performance, and Trust in Human-Agent Teams

This article reports on a mixed-methods study investigating the effects of working with one or two AI teammates in a three-person team, while also comparing to a control condition of an all-human team. These teams worked together in the NeoCITIES emergency response simulation to coordinate with one another to allocate scarce resources within a set time period. Results of the study found that human-AI teams had similar iterative development of team cognition and the importance of communication towards accelerating that development. However, action-related communication and explicitly shared goals were specifically beneficial for human-AI teams. Human-AI teams also trusted their AI teammates less, perceived less team cognition within their teams, and displayed inconsistent levels of team mental model similarity.