HOW EMOTIONAL CONTEXT AND TRUST SHAPE COGNITIVE PROCESSES WHEN SUPERVISING AN AUTOMATED SYSTEM

As automated systems become reliable, human operators are increasingly relegated to supervisory roles. Based on electroencephalographic and eye-tracking data, we investigated how perceived control and emotional context modulate information processing and performance monitoring when supervising a perfectly reliable automated drone in a dynamic ecological task. Twenty-eight participants monitored a drone performing obstacle avoidance across two 90-min sessions. During each trial, the drone approached obstacles, indicated its avoidance trajectory, and avoided them. In the “actor” session, participants reported the direction they would take before the system’s decision. In the “supervisor session”, they only monitored the system. The drone was perfectly reliable (0% errors) to induce high trust. The emotional context was induced by unpredictable loud auditory alarms. For each phase, (1) obstacle appearance and (2) system decision, we analyzed eye fixations in phase-relevant areas of interest, event-related spectral perturbations, and evoked-potentials to assess visual exploration and decision-making (phase 1) and performance monitoring (phase 2). Semi-Markov chain modeling of EEG activity characterized the temporal organization of cognitive processes. First results from fifteen participants showed weaker stress effects when supervising the fully reliable system. During the obstacle phase, emotional context disrupted exploration only in the “actor” session, with fewer fixations on
task-relevant areas and reduced occipital theta activity. During system decision, supervision was associated with fewer fixations on the system’s choice and reduced fronto-central negativity and theta activity. Semi-Markov analyses revealed similar cognitive sequences across conditions; however, under emotion, an early occipital positive component was enhanced in the “actor” session.