Presentation
Do AI-Decision Support Systems Land? Investigating Framed Accuracy of Artificial Intelligence on Pilot Performance
SessionWednesday Poster Session
DescriptionThe integration of AI into aviation decision support systems (DSS) offers opportunities to improve pilot performance, trust, workload, and situational awareness. However, effective human-AI interaction requires careful calibration of trust, transparency, and reliance, especially for student pilots. These systems provide data-driven recommendations, but pilots must make final decisions, requiring shared situational awareness. While such tools can reduce workload and improve efficiency, they also introduce risks like automation bias and overreliance, where pilots may follow incorrect recommendations. This study examined how disclosing AI accuracy affects decision-making during simulated mechanical failure tasks. This study recruited student pilots with 70-130 flight hours. Participants choose flight paths in simulated mechanical failure scenarios using DSS with either high (90%) or low (60%) accuracy. Two groups are compared: one informed of system accuracy and one not. Measures included decision accuracy, trust, workload, and situation awareness. Results showed positive effects of accuracy disclosure, specifically through increased decision accuracy and decreased workload. Situation awareness was approximately equal across all groups. Trust was higher in the group without disclosed accuracy, even for scenarios in which the recommendation was objectively incorrect, suggesting that opaque systems contribute to overtrust in AI.
Event Type
Poster
TimeWednesday, October 21st5:30pm - 6:30pm PDT
LocationSummit Pavilion