Trust and empathy toward artificial agents under task-language uncertainty

Artificial agents increasingly support human decision-making, yet users often interact with such systems under conditions where task content or system reasoning is not fully comprehensible. Despite this, users must still decide whether to rely on an agent. This study investigates how trust and empathy toward an artificial agent are formed under task-language uncertainty, and how explanations shape these processes. I conducted an online quiz-based experiment ([Formula: see text]) manipulating task language (comprehensible vs. incomprehensible) and explanation (present vs. absent), and measured pre-post changes in trust and empathy. Confidence, responsibility attribution, and task accuracy were also assessed as secondary outcomes. Results showed that task comprehension significantly influenced objective performance, confidence, and responsibility attribution: participants performed better, felt more confident, and attributed more responsibility to themselves when the task was comprehensible. In contrast, trust increased primarily in the incomprehensible task-language condition, and this increase was amplified by explanations, despite explanations having no effect on accuracy. Empathy increased from before to after the task in both language conditions, with the pattern of change modulated by task language. These findings reveal a dissociation between performance-based evaluation and social evaluation of agents, highlighting that trust can increase even when independent evaluation of task content is constrained.