Satire, Abuse, or Disinformation? The Use of Generative AI to Depict Politicians

• In Event: Emerging Challenges to Digital Campaigning

Abstract

Recent news stories have highlighted the ways in which synthetic media is used to push falsehoods, conspiracy theories, and other disruptive media. In part, the growing concerns are related to developments in text-to-image generative AI, which has enabled the creation of quick, and easy, artificial media portraying those in public life. As opposed to Deepfakes and Photoshops, these images can generate photorealistic images of unlikely scenarios with even lower barriers to entry than before. Citizens have used these to great effect. In March 2023, images generated on the Midjourney platform of Donald Trump resisting arrest, and images of Pope Francis wearing a white puffer jacket, both went viral (Morris, 22 March, 2023; Stokel-Walker, March 27, 2023). While created as satire, the virality and number of citizens mislead raised concerns regarding the impact AI may have on democracy. This builds on prior research which addresses the impact of AI on democratic trust (Vaccari & Chadwick, 2020); it’s potential in disinformation campaigns (Goldstein et al. 2023); and ethical considerations on the use of AI (c.f. Future of Life, 2023; Ellis, 2023).

This paper will investigate the use of text-to-image generation platforms when creating synthesised images of 87 politicians from the United States, United Kingdom, and other internationally recognised political leaders. To do this, 6,188 AI generated images (and prompts) from a period of 6-months, from Midjourney’s publicly accessible Discord server are collected. These are then analysed through content analysis to evaluate a series of interrelated questions regarding the targets, tone, and framing of these images based on prior codes of satire and abuse (such as Lichtenstein & Nitsch, 2021). The analysis responds to several key research questions, including how AI-generated prompts depictions can be used to tell us about the creators’ intentions, and the primary motivations for the creation of these images. Initial findings suggest that while evidence of potentially disruptive behaviour can be seen, such as attempted fake images, abusive prompts, and even sexualised images, the majority of these creations can be viewed as a form of political expression by their creators. Often fitting within the paradigm of humorous political content and playfulness (Mortensen & Neumayer, 2021; McLoughlin & Southern, 2021). This suggests rather than treating text-to-image platforms as a threat to democracy, their use can be studied to learn more about what sections of society think about politics and politicians through new forms of expression.

Author

• Liam McLoughlin, University of Liverpool