Sometimes right, sometimes wrong: Drivers' responses to inconsistently accurate automated vehicle system confidence...

Automated vehicles (AVs) are becoming increasingly equipped with intelligent functions that support drivers' decision-making. Human-machine interfaces (HMIs) that communicate an AV's confidence in its ability to navigate challenges in the driving environment are expected to become a pervasive feature. While this type of confidence display can enhance drivers' situation awareness, information presented to drivers may not always reflect accurate, real-world conditions, which can misguide perceptions and contribute to poor decision-making. Also, repeated exposure to inconsistently accurate information can reinforce negative biases. This study investigates how initial exposure to a series of both accurate and inaccurate information affects AV drivers' perceptions, behavior, and physiological responses in later interactions. Using a visual HMI displaying an AV's self-assessed confidence in avoiding a roadway obstacle, in a first phase, thirty participants were (unknowingly) assigned to two groups: one initially exposed to accurate confidence information, and the other to inaccurate confidence information. In the second phase, participants experienced the reversed information accuracy condition. The vehicle was highly reliable, but the AV confidence information was manipulated to either be aligned or misaligned with the system reliability. Across 12 takeover scenarios, drivers decided whether to take control of the vehicle, and their takeover decisions, trust levels, and physiological responses were collected. Overall, participants who were initially exposed to accurate information demonstrated heightened attention, faster voluntary takeovers, higher trust, and increased reliance on system information. In contrast, those initially exposed to inaccurate information spent more time monitoring the driving environment. Also, participants initially exposed to accurate information displayed higher cognitive workload (measured physiologically) and unchanged trust levels. This observation was also true when inaccurate information was presented later. The number of voluntary takeovers did not differ between the two groups. These findings highlight the role of initial information presentation in shaping drivers' perception and behavior, offering insights for designing AV systems that support effective human-AV interactions.