Examining the Impact of Varying Levels of AI Teammate Influence on Human-AI Teams

Abstract

The implementation of AI teammates is creating a wealth of research that examines how AI teammates impact human-AI teams. However, AI teammates themselves are not static, and their roles and responsibilities in human-AI teams are likely to change as technologies advance in the coming years. As a result of this advancement, AI teammates will gain influence in teams, which refers to their ability to change and manipulate a team's shared resources. This study uses a mixed-methods experiment to examine how the amount of influence AI teammates have on a team's shared resources can impact the team outcomes of human teammate performance, teammate perceptions, and whole-team perception. Results indicate that AI teammates that increase their influence on shared resources over time can stagnate the improvement of human performance, but AI teammates that decrease their influence on shared resources can actually encourage humans to improve their own performance. Additionally, AI teammates that are highly influential on shared resources can make humans perceive a greater cognitive workload. However, qualitative results indicate that these impacts on human performance and perception do not consistently impact the acceptance humans form for AI teammates. Rather, humans form acceptance for AI teammates if said AI use its influence to manipulate resources to benefit the personal goals of human teammates. These results have critical implications for human-AI teaming as it shows that the influence AI teammates have on shared resources can be designed in a way that improves human performance. However, future research is going to need to focus more critically on how the personal goals humans have, which may not align with a team's overall goals, are going to mediate the effectiveness of the AI teammate influence.