Man vs Machine: Electrodermal and Behavioural Responses to Human and AI‑Generated Faces

As technology advances and AI-generated faces increasingly resemble real human faces, it remains unclear whether human autonomic responses differ when viewing human and synthetic faces before explicit judgement. This study investigated whether electrodermal activity (EDA) can reveal pre-conscious physiological differences during the visual presentation of human versus AI-generated faces. Twenty-three adults (25–65 years) viewed static faces from three categories—human, hyper-realistic AI, and less realistic AI-generated faces—while continuous EDA was recorded during an initial passive viewing phase. Participants then completed an active categorisation task in which they classified each face as "human" or "AI", followed by a questionnaire assessing prior exposure to AI-generated content.Human faces elicited significantly higher EDA than both AI-generated categories during passive viewing, indicating stronger autonomic engagement with human faces than synthetic faces prior to explicit categorisation. In the active task, participants correctly classified 89% of human faces and 99% of less realistic AI faces, but only 78% of hyper-realistic AI faces, misclassifying 22% of them as human (60/276 trials across participants). These findings suggest that autonomic sensitivity to human social stimuli persists even when highly realistic AI faces are behaviourally difficult to distinguish from human faces, indicating that pre-conscious physiological responses do not straightforwardly translate into explicit behavioural discrimination. Self-reported familiarity with AI-generated content predicted behavioural performance but was unrelated to autonomic differentiation, further supporting a dissociation between physiological and explicit measures.Overall, the findings suggest that autonomic responses, explicit behavioural judgements, and prior experience capture complementary aspects of Human–AI face perception. Together, they indicate that increasing visual realism alone is insufficient to eliminate autonomic differentiation. These findings provide a basis for understanding how increasingly realistic AI-generated media are processed and, ultimately, for examining their broader societal implications, with relevance for the ethical development and deployment of synthetic media.Keywords: electrodermal activity, human-AI perception, Uncanny Valley, autonomic nervous system, artificial intelligence