Should AI Teammates Give All the Answers? Examining the Role of Different AI Information-Sharing Techniques on Team Cognition in Human-AI Teams

Abstract

This study investigated how the information-sharing traits of an AI teammate, such as augmenting team memory (ATM) and changes in intra- and extra-team information (IET), can enhance situation awareness (SA) and responses to unexpected events in human-AI teams. Thirty-one teams of two participants and one AI flew a simulated UAV task with the AI's information-sharing attribute set to ATM, IET, or control. Results showed that IET teams, whose AI provided direct solutions to task disruptions, were the most likely to overcome them. The ATM teams, whose AI helped teammates find the solution themselves, also outperformed the control group, but also exhibited more action communication, a better perceived shared mental model with the AI, and a more positive perception of SA compared to the IET condition. These results highlight that while providing direct, immediate AI solutions can improve responses to task disruptions, there can be a cost to team development.