I Feel Safe with the Prediction: The Effect of Prediction Accuracy on Trust

Abstract

Mental model building and trust are important topics in the interaction with Artificial Intelligence (AI)-based systems, particularly in domains involving high risk to user safety. The presented paper describes an upcoming study on applying methods from Human-Centered Explainable AI (HCXAI) to investigate the influence of AI accuracy on user trust. For the interaction with an AI-based system, we use an algorithm designed to support drivers at intersections by predicting turning maneuvers. We will investigate how drivers subjectively feel towards different presented and implemented accuracy levels. Participants will be asked to rate their respective levels of trust and acceptance. Moreover, we will investigate, whether trust and acceptance are depending on personality traits.