Evaluating the Potential of Interactivity in Explanations for User-Adaptive In-Vehicle Systems – Insights from a Real-World Driving Study

Due to advances in artificial intelligence (AI), humans are increasingly facing algorithm-generated content in everyday applications. To avoid threads to the system’s transparency and trustworthiness, the approach of explainable AI (XAI) will play an important role when designing these systems tied to the needs and characteristics of their end-users. Our work investigates explanation strategies for AI-based adaptive in-vehicle systems from a human-centered point of view. We present two explanation concepts: one interactive and one text-based approach. The concepts were evaluated and compared in a real-world driving study with 36 participants. The aim is to assess whether interactive engagement with explanations fosters the system’s understandability and the user’s mental model. Our results did not show significant differences between the concepts. Both groups performed well when assessing their mental model after experiencing the explanation concept. However, we found significant decreases in the mental model when measuring it again after participants experienced the prototypical adaptations of the system during the test drive.