Abstract

Deepfakes are manipulated digital audio-visual content that is produced to be highly photorealistic. Unsurprisingly, they can be leveraged for nefarious ends in politics. In this study, we adopted an integrative deepfake detection approach rooted in the theory of Perception of Visual Incongruity by Bruner and Postman (1949). We conducted an exploratory survey study (Malhotra and Grover, 1998) with 15 pilot subjects. Our preliminary findings point to three key factors that drive the detection of deepfake videos. First, a significant majority (64%) of cues deployed by humans to detect deepfakes are of the type tangential (both audio and visual). Second, cues that are considered to be of greater social value, such as political expectations, are perceptually accentuated. Lastly, humans deploy diverse cues to detect deepfake and authentic video.

Share

COinS