How situational risk shapes pilots’ decision-making in response to human-versus AI-labeled recommendations

As AI-based decision support becomes increasingly integrated into aviation, this study examined whether situational risk moderated pilots’ landing decision-making when identical recommendations were attributed to human versus AI sources. Thirty-two pilots completed a controlled landing decision task using a primary flight display with a 2 × 4 × 2 repeated-measures design: source label (human, AI), risk level (safe, moderate risk, high risk, extremely high risk), and decision stage (initial, final). Behavioral measures, drift–diffusion modeling, and eye-tracking were used to assess decision processes and visual attention. Results showed that situational risk systematically reduced landing rates. Under extremely high risk, the AI condition showed lower landing rates, higher recommendation concordance, and a greater probability of switching initially discordant decisions toward the recommendation than the human condition. Drift-diffusion modeling revealed a significant three-way interaction in drift-rate magnitude: under extremely high risk, absolute drift rates were greater in the AI condition than in the human condition at both decision stages. Decision thresholds were lower in the AI condition during the initial decision stage under safe and moderate risk. Eye-tracking results showed that, under extremely high risk, the AI condition was associated with lower fixation counts on the glide slope and higher fixation counts on the localizer. These findings support treating source labeling as a design consideration in aviation decision-support systems, particularly under high risk.