Validating Trust in Human-Robot Interaction through Virtual Reality: Comparing Embodied and "Behind-the-Screen" Interactions

Abstract

Human-agent interaction is commonplace in our daily lives, manifesting in forms ranging from virtual assistants on websites to embodied agents like robots that we might encounter in a physical setting. Previous research has largely been focused on “behind-the-screen” interactions, but these might not fully encapsulate the nuanced responses humans exhibit towards physically embodied agents. To address this gap, we use virtual reality to examine how simulated physical embodiment and the reliability of an agent (automated robotic crane) influence trust and performance in a task simulating a quality assurance role and compare it to a “behind-the-screen” interaction. Out of 119 participants, the data revealed there is a marked behavioral shift observed when reliability hits a 91% threshold, with no influence from embodiment. Furthermore, participants displayed a tendency to trust and defer to the decisions of embodied agents more, especially when these agents were not infallible. This study accentuates the need for transparency about an agent's capabilities and emphasizes the significance of ensuring that the agent's representation is congruent with the nature of the interaction. Our findings pave the way for a deeper understanding of human-agent interactions, suggesting a future where these interactions might seamlessly blend the virtual and physical realms.

Keywords: Human-robot interaction, embodied agents, signal detection theory, virtual reality, trust calibration, decision support systems

DOI: 10.54941/ahfe1004408