Author ORCID Identifier

Date of Award

Spring 5-14-2024

Document Type

Thesis (Master's)

Department or Program

Computer Science

First Advisor

Lorie Loeb

Second Advisor

James Mahoney

Third Advisor

John Bell


Recent advancements in Passthrough technology and Transformer based Large Lan- guage Models present a unique opportunity to delve into the intricate dynamics of hu- man social behavior. This study investigates the effects of human-controlled avatars and AI-controlled avatars on individuals with social anxiety within a Passthrough environment. A total of N=19 participants engaged in interactions with a human- controlled avatar via a head mounted display, followed by an AI representation, both within the Passthrough setting. Conversations were structured to elicit moments of stress or challenge in participants’ lives. Parameters such as heart rate, sincerity, stress levels, feelings of social pressure, and comfort were measured to gauge influence. A comparative analysis was conducted with earlier studies that utilized less advanced technology, including algorithm-based AI and human-operated avatars controlled via keyboard and mouse. Results indicate a positive impact of these technological ad- vancements, with participants perceiving the AI agent as more lifelike and relatable than systems explained in prior research. Additionally, interacting with the human- controlled avatar in the Passthrough environment appears to alleviate perceived stress and social pressure while maintaining a high degree of social presence compared to face-to-face encounters (in vivo). Notably, the human-controlled avatar demonstrates a more pronounced positive effect on social anxiety compared to its AI-controlled counterpart.