Exploring Neurophysiological Responses to Cross-Cultural Deepfake Videos

Abstract

Deepfake videos, which use artificial intelligence techniques to create realistic but fabricated footage, have raised concerns regarding their potential to deceive and manipulate viewers. This study is one of the first of its kind that aimed to investigate the cross-cultural perception of deepfakes and uncover potential neural markers associated with their detection. Electroencephalography (EEG) data were recorded from 10 healthy participants while they viewed three categories of videos: Asian people speaking Chinese (C-C), Asian people speaking English (C-E), and Middle Eastern people speaking English (A-E). Participants were asked to determine whether each video was real or fake. Behavioral analysis revealed that participants performed better in differentiating real and deepfake videos when the provided visual stimulus was in a language they were familiar with (English) and when the actor belonged to an ethnically similar background. EEG analysis demonstrated significant differences in brain signals between the three categories, suggesting the potential use of EEG as a biomarker for deepfake classification. Machine learning models achieved accuracies of up to 84.52% in categorizing the EEG data while observing real vs. fake videos, with Support Vector Machines. These findings contribute to our understanding of deepfake perception, have implications for the development of deepfake detection methods, and highlight the importance of media literacy in the face of digital deception.