Human-AI Communication in First Aid Response: Effects on Team Performance, Trust, and Teamwork Appraisal

Abstract

This study examines human-AI communication in simulated first aid scenarios, exploring its impact on team performance, AI teammate trust, and perceived workload and self-efficacy. Twenty-five participants without first aid training collaborated with an autonomous 9-1-1 call taker (ChatGPT-3) to administer bleeding control actions on a specialized mannequin. During the simulations, participants followed inconsistent and, at times, inaccurate AI instructions while passively responding to AI commands and questions. However, participants perceived the AI's instructions to be accurate, helpful, and easy to follow and, irrespective of first aid performance, participants consistently reported high self-efficacy and trust in the AI teammate. The study highlights the importance of team roles (i.e., leader versus follower) and role-based communication in fostering trust and perceived self-efficacy within human-AI teams, and outlines design implications for human-AI teaming in safety-critical contexts.