From approach to avoidance: How AI agent cognitive and affective empathy elicits the uncanny valley effect

Abstract

The empathy expressed by AI agents is crucial in human-AI interactions, especially within mental health contexts. However, the mechanisms underlying users’ responses to cognitive and affective empathy from AI agents are not well understood. This study examines the theoretical mechanisms and boundary conditions that determine how AI empathy influences users’ approach-avoidance intentions. Drawing on the computers are social actors (CASA) paradigm and the uncanny valley effect (UVE), we propose a moderated dual mediation model. This model is empirically tested using two experimental studies. The results reveal distinct pathways by which cognitive and affective empathy affect approach-avoidance intentions. Specifically, cognitive empathy primarily influences approach-avoidance intentions via perceived novelty, whereas affective empathy exerts its effect through perceived warmth. Additionally, perceived warmth and eeriness together mediate the impact of affective empathy on approach-avoidance intentions. Notably, mindful AI agents strengthen the effect of affective empathy on the UVE but diminish the influence of cognitive empathy on approach-avoidance intentions, relative to mindless agents. These findings provide important insights for AI designers and companies seeking to develop empathetic, user-centered conversational agents for mental health applications.