How LLM-Awareness and Explanation Format Shape Driver Experience During AV Anomalies

The integration of Large Language Models (LLMs) into autonomous vehicles (AVs) aims to enhance human-AV interaction through natural language explanations. However, the extent to which driver awareness of LLM involvement, and explanation format itself, influences driver experience during unexpected AV behavior remains largely unexplored. In this study, we investigated the impact of drivers’ beliefs about the explanation source (i.e., LLM-generated vs. human experts) across three explanation formats: causal, counterfactual, and contrastive. Results showed a tendency toward greater trust in human expert-derived explanations, particularly in counterfactual scenarios. Furthermore, contrastive explanations significantly reduced mental workload compared to causal and counterfactual formats, demonstrating a distinct advantage in cognitive processing efficiency. These findings indicate that strategically tailoring the selection of explanation source and format to the time-sensitivity and complexity of specific driving scenarios could improve driver trust and performance.