Presentation
Human-Centered Explainable Artificial Intelligence in Partially Automated Vehicles: Expected and Unexpected Vehicle Actions
SessionThursday Poster Session
DescriptionAutonomous vehicles (AVs) leverage Artificial Intelligence (AI) to sense the driving environment and make driving-related decisions. Currently, the most advanced AVs available to consumers require a human driver to oversee their automated features, as AVs are susceptible to errors. Explainable AI (XAI) communicates the AI’s intentions and the AV’s actions in a user-friendly way, keeping the driver informed. The present study investigated the effects of explanation timing, driving scenario type, and expectation type on perceived usefulness and information sufficiency in the explanation, using a within-subjects factorial design. Participants were shown different AI explanations in scenarios that were either expected (followed the law) or unexpected (violated the law), and rated the usefulness and the sufficiency of the information in the explanation. Results from 34 participants showed that they found the explanation more useful and the information more sufficient when it was provided before or during the AV’s action, rather than after. Participants also rated the explanations as more useful and sufficient when they were for expected actions than for unexpected ones. Overall, explanations in AVs should include more information when the vehicle acts unexpectedly, and be provided earlier rather than later to promote safer driver-AV interaction and the adoption of AVs.
Alternate Presenter
Event Type
Poster
TimeThursday, October 22nd5:30pm - 6:30pm PDT
Location
Similar Presentations