Interactions with artificial agents in virtual reality: A study of behaviors in game-based tasks

This study investigates how the appearance and behavior of virtual agents influence human reactions during immersive interactions in virtual reality. Participants engaged in two interactive games—a word-matching task and an iterated Prisoner’s Dilemma—with virtual partners that varied in their degree of humanlikeness and dominance-related behavior. Behavioral responses, encompassing cooperation patterns and posture changes (based on video recordings), as well as self-reported mind attribution (agency and experience) and emotional data were examined. Results show that hybrid-looking (uncanny) agents triggered greater eeriness and postural contraction among participants, indicating discomfort, though not increased threat or reduced likability. Mind attribution—particularly agency and experience—was positively associated with likability, challenging earlier theoretical claims. In the Prisoner’s Dilemma, women showed higher cooperation and forgiveness rates, especially toward robot-looking agents. Together, the findings suggest that agent appearance and attributed mental capacities influence how participants evaluate and respond to virtual partners.