DEEPFAKE IN FACE PERCEPTION RESEARCH

Abstract

Presents the state-of-the-art Deepfake face replacement image collage method, an artificial intelligence (AI) product that can be used to create high-quality, realistic videos with a fake or replaced face, with no obvious signs of manipulation. Based on the DeepFaceLab (DFL) application, the process of creating video images of an “impossible face” is described step by step. The results of the experiments of studying the perception patterns of the moving “impossible face” and their differences in statics and dynamics are presented. The stimuli were two DFL-generated models of virtual sitters with impossible faces: a video image of a chimerical face, in which the right and left sides belong to different people, and a Tatchered face with the eyes and mouth areas rotated by 180°. It was shown that the phenomena of perception of the “impossible face”, registered earlier under static conditions (integrity of perception of the split image, distraction and inversion effect), are preserved and acquire a new content when dynamic models are exposed. In contrast to the collaged images, the original faces in statics and motion, regardless of egocentric orientation, are evaluated positively at the level of high values. Under all tested conditions the gender of the virtual sitter is determined adequately, the perceived age is overestimated. Estimates of the virtual sitter’s emotions from his video images are differentiated into basic (stable) and additional (changing) states, the ratio of which depends on the content of a particular episode. Deepfake image synthesis technology significantly expands the possibilities of psychological research of interpersonal perception. The use of digital technologies simplifies the creation of “impossible face” stimulus models necessary for in-depth study of representations of the human inner world, and creates a need for new experimental-psychological procedures corresponding to a higher level of ecological and social validity.