Evaluating Demographic Bias of Virtual Humans With Gaze: Race and Sex Concordance

Research on interpersonal relationships has shown that individuals often exhibit different gaze patterns when viewing faces of other races compared to those similar to their own. In the context of Virtual Humans (VHs), participatory design processes frequently include features that reflect the designer's identity. This paper explores gaze behavior towards demographically diverse VHs in the context of emerging educational technology. Using Augmented Reality (AR), 33 participants engaged in a within-subjects study where their eye movements, learning outcomes, and perceptions were compared. Results indicate that gaze (no. of fixations and durations) was influenced by demographic concordance (when race and sex matched that of the VH). Additionally, learning outcomes positively correlated with the number of fixations, suggesting that increased visual focus may enhance learning. However, participants frequently misidentified the race of the Hispanic, Asian, and Black VH, revealing a potential perceptual misalignment likely shaped by implicit social categorization biases.