Abstract
<p>Social biases can shape rapid judgements of others, but social desirability and limited self-awareness make these biases difficult to measure. We tested whether electroencephalographic (EEG) responses to faces could provide implicit feedback for generating visual prototypes of political leadership. Forty adults were recruited, of whom 38 provided analysable EEG data. Participants viewed 300 AI-generated faces during rapid serial visual presentation while attending to faces they would vote for, and subsequently provided explicit yes/no votes that were used to label image-locked EEG epochs. An EEGNet ensemble classified later-voted versus non-voted faces, and its predictions guided a pretrained StyleGAN3 model to generate participant-specific and multi-participant BrainVote faces. Twenty returning participants evaluated these faces and label-shuffled matched controls in a mock election two months later. Classification exceeded permutation baselines for 36 of 38 participants. BrainVote faces were preferred to matched controls, and 80% of returning participants voted for the face generated from their own EEG responses. Aggregation across participants produced increasingly homogeneous prototypes characterised by senior and masculine features, with stronger patterns among participants with larger gender-career Implicit Association Test scores. EEG-informed generative AI can therefore visualise sample-level implicit leadership associations while revealing how aggregation may amplify demographic stereotypes.</p>