Abstract

This study explores the impact of embodiment on trust and user experience by comparing an embodied AI (Pepper) and a disembodied AI (Alexa) in a competitive rock-paper-scissors game. Using a mixed-methods ap-proach, 71 participants interacted with both systems, revealing that Pepper scored significantly higher in trust (M = 4.18, SD = 0.67) and engagement (M = 4.30, SD = 0.72) compared to Alexa (trust: M = 3.87, SD = 0.79; engagement: M = 3.65, SD = 0.81). Participants noted Pepper’s physical gestures and expressions enhanced social presence, fairness, and emotional connection, while Alexa’s pre-dictability and efficiency were valued for task-oriented interactions but perceived as less engaging and occasionally manipulative in competitive settings. Privacy concerns were more prominent with Alexa due to its disembodied, cloud-based nature. The findings extend the Capability-Benevolence-Integrity model of trust, demonstrating how embodiment enhances integrity and emotional engagement. Practical implications suggest embodied systems like Pepper are better for social and emotional contexts (e.g., education, therapy), while disembodied systems like Alexa excel in efficiency-driven tasks. This study addresses gaps in trust dynamics in competitive human-AI interactions and highlights trade-offs be-tween embodiment and efficiency. Limitations, such as the controlled setting and modest sample size, call for future longitudinal and cross-cultural research to fur-ther explore these dynamics.