Who are the targets of non-deceptive political deepfakes?

Synthetic media, commonly known as deepfakes, is video or audio material doctored by the use of some form of artificial intelligence. Recent technological advances have made the production of deepfakes increasingly accessible to the public, including several apps and free software. Deepfakes have been much debated in both academic and popular press, where focus has been placed almost exclusively on deception and other epistemic threats to democracy. While deception is an important angle, it is only one among many ways in which deepfakes are currently used for political communication. Most saliently, there is a rich production of political satire on platforms such as YouTube, Instagram and TikTok. However, in spite of their seemingly wide spread, little is known within the academic community about non-deceptive political deepfakes. How big is their audience? What kind of actors are sharing/producing such content? And what is the demographic and political composition of the targets of the videos (i.e., whose faces or voices have been superimposed)? The goal of the present paper is to answer such questions. In pursuing it, we set up a twofold methodological procedure. In the first step, we conduct an in-depth qualitative coding of 200 deepfake videos uploaded to YouTube, Instagram and two pornographic video sharing sites that allow deepfakes, collected over a period of three months. Based on the qualitative coding, we perform a larger statistical analysis on the demographic composition of a corpus of 5000 videos collected over the course of three months. Our findings indicate that some demographics correlate with certain kinds of satire, such as pornographic deepfakes almost exclusively using the faces of female politicians. In discussing the findings, we call attention to the dual nature of deepfakes. On the one hand, they clearly reinforce existing power asymmetries, and on the other, they are opening up for novel forms of satire and resistance. We conclude by outlining some future avenues for both conceptual and empirical research to be undertaken on non-deceptive deepfakes.