New paper accepted to the 2026 ACM Conference on Fairness, Accountability, and Transparency. Cedric Faas will present our work at Facct’26 in July in Montréal, Canada.
AI Support in Safety-Critical Tasks: How Perceived Autonomy and Meaningfulness Shape Human-AI Joint Performance
Cedric Faas, Richard Uth, Sarah Sterz, Markus Langer, Anna Maria Feit
Abstract:
In safety-critical and time-sensitive tasks, AI support can enhance decision accuracy and safety. However, it may also undermine users’ psychological needs, which generally affect job satisfaction and performance. In this paper, we empirically and theoretically examine how perceived autonomy, meaningfulness, and satisfaction relate to task performance in safety-critical AI-supported tasks. Specifically, we conduct a simulated human oversight experiment, where participants (N=274) monitored a drone that faced ten critical situations, choosing from six possible actions to resolve each situation. To examine the effects of autonomy, participants had full-human control or an AI system constrained choices to four, two, or only one option (between-subject study). Surprisingly, only extreme restriction to one selectable action improved performance but significantly reduced perceived autonomy and meaningfulness, with these effects intensifying over time. At the same time, when given multiple action choices, participants with higher perceived autonomy performed better. We situate these findings in the work design literature and advocate a shift towards system design that supports psychological processes.
Be First to Comment