Presentation
Eyes Don’t Lie: Indifferent to AI-Generated Depression Screenings
SessionPoster Session 1
Descriptiont plays a key role in whether people will use AI-powered tools for mental health support. In this study, we explored how users respond to a depression self-diagnosis app that presents two outcomes—one labeled as "doctor-generated" and the other as "AIgenerated." The doctor-generated result was based on each participant’s actual PHQ-9 depression score, while the AI-generated result was randomly selected and sometimes matched the true diagnosis by chance. Participants (n = 47) with depressive symptoms used the app while we recorded their eye movements and pupil dilation using a Tobii Pro Spectrum eye tracker. We analyzed pupil dilation in two key moments: one second before the participant clicked to choose which diagnosis they trusted, and 0.5 seconds during the click. Participants who chose the doctor-generated result showed greater pupil dilation during this decision phase, suggesting higher cognitive engagement. We also trained an Extreme Gradient Boosting model with 20-fold cross validation to predict whether a participant would trust the AI or the doctor label based on their pupil data. The model performed well, with an AUROC of 0.871. Using physiological data like pupil size may help us design trust-sensitive mental health tools that adapt in real-time to how users feel.
Event Type
Poster
TimeTuesday, October 14th5:30pm - 6:30pm CDT
LocationRiverside East