Game-Based Platforms for Studying Virtual Agent Believability

Abstract

Believable virtual agents are central to many interactive systems, yet current methods for evaluating user trust and perception often rely on limited, decontextualized tools. My research investigates how narrative-driven, game-based environments can be used to study users’ perceptions of virtual agent believability, particularly in relation to gender presentation and role-based behavior. I am developing an interactive platform where users engage with male, female, and androgynous agents embedded in branching scenarios that simulate trust-based decision-making. The platform incorporates adaptive storytelling, real-time analytics, and machine learning to capture both behavioral and attitudinal responses. By gamifying the evaluation process, the research aims to produce richer, more ecologically valid data than traditional surveys. This work draws from human-computer interaction, affective computing, and AI ethics to explore how design choices, such as agent role, gender, and interactivity, affect user judgments of reliability and integrity. As an early-stage PhD student, I am currently refining the research questions, platform design, and experimental methods. Feedback from the AIIDE community will help sharpen the research focus, improve methodological choices, and ensure broader relevance across virtual agent applications.