Trust in artificial intelligence: understanding and calibrating trust and distrust in the human-AI interaction

Trust has emerged as a key measure in human–AI interaction in recent years. Trust in AI is recognized not only in academic research but also in industry and politics, where it is often considered a remedy for issues related to fairness, accountability, and transparency in AI systems. This importance of trust necessitates a thorough understanding of it. In particular, research into explainable AI (XAI) has suggested that explanations and other forms of transparency can increase trust in AI. However, the empirical evidence for this assumption is inconclusive. This dissertation explores potential reasons for this ambiguity and aims to contribute to a better understanding of end-users’ trust in AI. As such, manuscript 1 investigates post-hoc explanations and highlights the distinction between trust and behavioral measures of reliance, while also emphasizing the importance of human-related factors in AI-assisted decision-making. Manuscript 2 presents the first comprehensive validation study of trust questionnaires in the context of AI, advocating to consider both trust and distrust. Manuscript 3 explores certification labels as an alternative approach beyond traditional XAI methods to increase end-users’ warranted trust, demonstrating their potential. Overall, this dissertation seeks to provide a more holistic understanding of trust in AI by illustrating how to increase calibrated trust and distrust to a level warranted by the AI’s trustworthiness.