Related Experiment Video
Updated: Jul 2, 2026
07:34
Perceptual and Category Processing of the Uncanny Valley Hypothesis' Dimension of Human Likeness: Some Methodological Issues
Published on: June 3, 2013
Neural signatures of hyper-realistic AI-generated faces: dissociating behavioral indistinguishability from implicit
Alice Mado Proverbio1,2, Mariia Dosaikina3,4
1Cognitive Electrophysiology Laboratory, Department of Psychology, University of Milano-Bicocca, Piazza dell'Ateneo nuovo 1, 20162, Milan, Italy. alice.proverbio@unimi.it.
Scientific Reports
|July 1, 2026
Summary
Even though people can't easily tell AI faces from real ones, the brain detects the difference. Neural signals show AI faces trigger stronger familiarity and aesthetic responses.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Cognitive Science
Background:
- Generative AI creates hyper-realistic synthetic faces, challenging human perception.
- Behavioral studies show AI faces are often misclassified as real and rated as more familiar/attractive.
Purpose of the Study:
- To investigate neural processing differences between real and AI-generated faces.
- To determine if the brain can distinguish AI faces despite behavioral limitations.
Main Methods:
- Electroencephalography (EEG) recorded from 30 participants viewing 440 real and GAN-generated faces.
- Behavioral validation of face perception, familiarity, and aesthetic judgments.
- Source localization analysis using standardized low-resolution brain electromagnetic tomography (swLORETA).
Main Results:
- AI faces were poorly identified as artificial and rated higher in familiarity and aesthetics.
- Face-evoked event-related potentials (ERPs) showed AI faces elicited enhanced N250, P300, PN400, and late positivity.
- swLORETA revealed reduced engagement of specific brain networks for real faces, particularly male ones.
Conclusions:
- The human brain possesses neural markers sensitive to the artificial origin of hyper-realistic AI faces.
- Enhanced neural responses to AI faces suggest heightened familiarity and aesthetic appeal due to algorithmic processing.
- A dissociation exists between explicit recognition and implicit neural evaluation of synthetic versus real faces.
