Seeing inside the black box: Effects of algorithmic knowledge on system trustworthiness in aviation

Kurzfassung

Automation has shaped aircraft design for the past seven decades, moving from fully manual control to highly automated systems. Despite this, pilots retain final authority for safety-critical decisions, which requires continuous trust calibration. This study investigated how algorithmic transparency influences pilots’ learned trust in an AI-based cockpit assistant. Twenty-one commercial pilots were randomly assigned to an early transparency condition (detailed explanations of the AI’s algorithms and performance provided before the tutorial) or a late transparency condition (the same information delivered after the experiment). Each pilot used the assistant during a tutorial and a realistic simulated flight. Mixed-effects ordinal models showed higher trust ratings for the early compared to the late transparency group during both tutorial and flight, and a larger trust increase between phases. In conclusion, providing pilots with timely, clear information about AI processes modestly improves trust and supports calibrated reliance.