This study aims to discover the impact of the interaction between learning performance (as behavioral realism) and realistic appearance (as form realism) of AI-powered software robots on user trust. The study also aims to reveal how and why the interaction happens, especially from the dual processing perspective of affective and cognitive human responses.
We adopted an experimental research methodology in a movie recommendation system environment where machine learning-based recommendations are widely used. We conducted a 3 × 2 factorial design experimentation based on the three levels of learning performance (low, mid and high) and two levels of realistic appearance (caricature avatar and digital human). We used ANCOVA and the PROCESS Macro to analyze our models.
Our results confirm that learning performance (intelligence) is a critical factor influencing user trust in software robots, and this intelligence–trust relationship is influenced by their realistic appearance. Our results further reveal that there are two significant intermediating mechanisms, i.e. affective and cognitive user responses, and that the intelligence–appearance interaction effect on trust is explained especially by the affective response mechanism.
This study provides valuable implications for creating optimal learning performance and realistic appearance that can lead to trust in various settings where AI-powered software robots are utilized.
This study has sveral contributions to the literature. First, in addition to the well-recognized factor of anthropomorphic characteristics, this study investigates another critical behavioral factor of AI-powered robots (learning performance as intelligence characteristics) and the intriguing interaction between the two realism factors. Second, drawing upon the mediated moderation perspective, the study proposes a novice perspective on how and why the two realism factors can build user trust (the underlying mechanisms).
