What Did My Car Say? Impact of Autonomous Vehicle Explanation Errors and Driving Context On Comfort, Reliance, Satisfaction, and Driving Confidence

TL;DR

This study investigates how autonomous-vehicle explanation errors affect passenger comfort, reliance, satisfaction, and driving confidence, across driving contexts and user traits. Using a within-subject, online simulation with 232 participants and three explanation accuracy levels (accurate, low 'what'/'why' error, high 'what'/'why' error), the authors show that explanations contribute as much to trust and acceptance as actual driving performance; errors degrade all four outcomes, with larger harms for more severe errors. Contextual factors (perceived harm and driving difficulty) and individual differences in prior trust and AV expertise further modulate these effects, indicating the need for context-aware, personalized explanation strategies. The findings have practical implications for designing trustworthy AV explainability systems, informing UI design, deployment guidelines, and future research on adaptive explanations that maintain safety, user satisfaction, and adoption. They also highlight that accuracy of the 'what' description may be particularly crucial for user acceptance, suggesting a prioritized starting point for explanation design in safety-critical AI systems.

Abstract

Explanations for autonomous vehicle (AV) decisions may build trust, however, explanations can contain errors. In a simulated driving study (n = 232), we tested how AV explanation errors, driving context characteristics (perceived harm and driving difficulty), and personal traits (prior trust and expertise) affected a passenger's comfort in relying on an AV, preference for control, confidence in the AV's ability, and explanation satisfaction. Errors negatively affected all outcomes. Surprisingly, despite identical driving, explanation errors reduced ratings of the AV's driving ability. Severity and potential harm amplified the negative impact of errors. Contextual harm and driving difficulty directly impacted outcome ratings and influenced the relationship between errors and outcomes. Prior trust and expertise were positively associated with outcome ratings. Results emphasize the need for accurate, contextually adaptive, and personalized AV explanations to foster trust, reliance, satisfaction, and confidence. We conclude with design, research, and deployment recommendations for trustworthy AV explanation systems.

Figures (4)

• Figure 1: Example images from four driving scenarios: (1) the AV slows for a pedestrian crossing the road on a foggy day in the city, (2) the AV is cut off by a vehicle turning right from the left lane, (3) the AV merges around a slow lead vehicle in a forested area, (4) the AV moves past a cyclist on a suburban street. Written AV explanations appear on the vehicle's dashboard display.
• Figure 2: Mean Outcome by Error Condition Across All Scenarios. Plot A shows mean Comfort Relying on AV, Plot B shows mean Reliance Preference (Binary, scaled to 1-10), Plot C shows mean Satisfaction with AV Explanation, and Plot D shows mean Confidence in AV Driving Ability. Error bars are standard error. Results consistently demonstrate Accurate > Low > High.
• Figure 3: Mean Harm and Difficulty Across All Scenarios. Results show harm and difficulty are generally correlated, but differ depending on the specific scenario. We find that these contextual characteristics directly impacted outcome ratings as well as influenced the relationship between errors and outcomes.
• Figure 4: Relative Importance of Factors on Reliance Decisions. This shows the mean self-reported importance rating of each factor with respect to a person's reliance decision (errors bars are standard error). Results show the accuracy of each type of information, perceived contextual harm and difficulty, and the driving ability of the AV were all important factors.