Let Me Explain What I Did or What I Would Have Done: An Empirical Study on the Effects of Explanations and Person-Likeness on Trust in and Understanding of Algorithms

AI-based systems become ubiquitous but stay opaque. To promote informed usage, explanations of how systems compute outcomes emerged as a promising strategy (XAI). Moreover, there is a trend of equipping technologies with human-like features (e.g., natural language, social roles) to enhance the system's person-likeness, thereby influencing users' understanding and trust. The present empirical study (N = 762) investigates the effects and interplay of explanation types and person-likeness on understanding of and trust in AI-systems. It is designed as 2 (contrastive 'why not' explanations vs. comparative 'why' explanation) x 2 (self-referencing vs. neutral formulation) between-subject study with a no explanation-baseline condition. Results demonstrate positive effects of explanations on people's understanding, whereby contrastive explanations – despite being perceived as more complex – led to significantly higher factual understanding compared to the comparative explanations. Person-likeness had no effect on trust or understanding which challenges current trends of enhancing human-like features of technologies.