Predicting a Malicious Stop Sign: Knowledge, Exposure, Trust in AI
Event Type
Technical Groups
Surface Transportation
TimeTuesday, October 5th4:00pm - 4:15pm EDT
DescriptionAutomated vehicles are dependent on successful Artificial Intelligence (AI) for safe roadway operations. AI employs computer vision techniques for detecting roadway signs. We investigated how human drivers perceive AI's capabilities in an adversarial driving context where a stop sign is compromised. We found that participants with higher knowledge of AI tended to trust AI more, and those who reported more experience with AI had a greater understanding of AI. Participants correctly deduced that a maliciously manipulated stop sign would be more difficult for AI to identify. Nevertheless, participants still overestimated the AI’s ability to recognize the malicious stop sign. Our findings suggest that participants might not have an accurate understanding of specific AI systems leading them to overtrust the AI in certain conditions.