From Black Box to Transparency: The Impact of Multi-Level Visualization on User Trust in Autonomous Driving

Autonomous systems' "black-box" nature impedes user trust and adoption. To investigate explainable visualizations' impact on trust and cognitive states, we conducted a within-subjects study with 29 participants performing high-fidelity driving tasks across three transparency conditions: black-box, standard, and enhanced visualization. Multimodal data analysis revealed that enhanced visualization significantly increased perceived usefulness by 28.5% (p < 0.001), improved functional trust, and decreased average pupil diameter by 15.3% (p < 0.05), indicating lower cognitive load. The black-box condition elicited minimal visual exploration, lowest subjective ratings, and "out-of-the-loop" behaviors. Fixation duration showed no significant difference between standard and enhanced conditions. These findings demonstrate that well-designed visualizations enable balanced trust calibration and cognitive efficiency, advocating "meaningful transparency" as a core design principle for effective human-machine collaboration in autonomous vehicle interfaces. This study provides empirical evidence that transparency enhances user experience and system performance.