Proceedings of the Annual Meeting of the Cognitive Science Society

As AI-driven concepts like Single-Pilot Operations challenge traditional two-pilot operations, debates among manufacturers, pilot unions, and regulators highlight the need for empirical evidence on how AI uncertainty affects human cognition and performance. This research compares human-human and human-AI cockpit teams, focusing on workload and performance when the pilot monitoring role is assumed by AI. 34 pilots completed simulated takeoff scenarios under three conditions: a human co-pilot, a reliable AI co-pilot, and a faulty AI co-pilot. Results indicate that collaboration with reliable AI can sustain workload levels and team performance comparable to human crews, whereas faulty AI behavior corresponds to increased workload and degraded team performance. Further analyses suggest that performance degradation is linked to undetected AI errors and AI limitations. Although pilots were often able to compensate for AI shortcomings, team collaboration was adversely affected, underscoring the importance of AI reliability and transparency for safe human-AI teaming.