Submission 372
Perceiving People Who Aren’T Real: Neural and Behavioral Responses to Presumed AI-Generated Faces
SymposiumTalk-04
Presented by: Martin Maier
As generative AI becomes more widely available, computer-generated “people” that appear deceptively real—so-called deepfakes—are entering our daily lives. The social impact of these stimuli depends not only on their visual quality but crucially on the psychology of the perceiver: how beliefs, expectations, and affective context shape perception and evaluation. This talk presents a series of studies examining how presuming faces to be “real” or “AI-generated” influences neural processing and emotional judgments—first using identical real faces paired with explicit authenticity labels, and later, comparing real and AI-generated faces while measuring participants’ own authenticity judgments. Using EEG and behavioral measures, we investigated how authenticity beliefs modulate early perceptual, semantic, and affective processing of neutral, happy, and angry faces. Complementary behavioral studies extended these findings across all basic emotions, showing that negative compared to positive emotional valence can render social judgments relatively immune against labeling images as fake. Together, these studies illustrate how beliefs about authenticity shape emotional processing beyond the visual input itself, and how emotional content influences the impact of AI-generated social stimuli. These findings connect to the symposium’s broader focus on how emotions, beliefs, and expectations guide perception in technologically mediated social environments, advancing the understanding of how we can navigate contemporary challenges to social cognition.