Pilot Study of an Anxiety Induction Method Applying Generative Artificial Intelligence and Mixed Reality

Worry is theorized to be a predominantly verbal, abstract thinking style that suppresses vivid mental imagery and blunts affect, thus serving avoidance. In order to test a new way to circumvent this avoidance, we examined whether personalized, AI-generated depictions of worry scenarios can elicit anxiety. These images were presented in a mixed reality worry gallery, arranged spatially and embedded in the physical environment using a mixed reality head-mounted display. Applying a within-subject repeated-measures design, we assessed subjective state anxiety and affect, as well as physiological responses (heart rate and skin conductance level) of 34 healthy adults (70.6% female, 24.47 years on average). Additional measures included representation accuracy and realism of worry depictions. Multivariate analyses of variance revealed higher, but only medium-sized anxiety ratings for AI-generated worry versus neutral images. Negative affect increased and positive affect decreased after worry-image presentation. Higher perceived realism and alignment were associated with greater anxiety. Physiological indices did not show significant increases. Our results suggest that AI-generated worry imagery can induce subjective anxiety and modulate state affect in mixed reality settings. Further work should improve stimulus reliability, include clinical samples, and benchmark mixed reality against other presentation modalities such as imaginal worry exposure to clarify equivalence and incremental value. • AI-generated worry images induced anxiety and negative affect in healthy subjects • Higher depiction accuracy and realism correlated with stronger anxiety responses • Higher trait anxiety and worry were related to stronger subjective anxiety • Intra- and inter-individual variance in responses warrant procedure refinement