PUB - Publikationen an der Universität Bielefeld

A healthy balance between trust and distrust in AI advisors is of tantamount importance in critical situations such as healthcare scenarios. Krügel et al. have shown that, for moral decisions, trust in AI advisors does not vary as initially expected when those advisors are presented as typically untrustworthy. Building on Krügel et al., this study constructs a more immersive setting with a visual novel and the speech-capable embodied virtual agent Floka for audiovisual advisory. The resulting data shows that transparent information about (un-)trustworthy sources of training data does indeed not influence advice-taking behavior and neither does perception of the AI agent as measured via the Godspeed questionnaire. However, Perceived Likability and Intelligence significantly correlate with how often AI advice was taken.