“Deepfakes” and the End of the Photographic Age

Abstract

A recent innovation in artificial intelligence, known as Generative Adversarial Networks (GANs), has enabled computers to generate images that are visually indistinguishable from photographs. GAN images, sometimes called “deepfakes”\footnote{The term “deepfake” is often to refer to images (or videos) of people created for deceptive or salacious purposes. In this paper I will use the term ‘GAN image’ to refer more broadly to all photorealistic images generated by GANs, regardless of their purpose or content.}, have already been recognized to pose an epistemic threat to society by undermining the capacity of photographs to provide evidence. In this paper, I will investigate both the epistemic status and the potential aesthetic value of GAN images, as well as how the proliferation of GAN images will affect the epistemic and aesthetic value of true photographs. I will affirm the view that GAN images are a potential epistemic threat, but also argue that they are nevertheless a medium with significant potential for artistic expression. To do so, I will draw upon Dawn Wilson’s argument that photographs are ontologically dual and can be considered as both mind-independent “photo-images” and mind-dependent “photo-pictures”. I will extend that argument to GAN images to show that, while they are indeed the outputs of mind-independent computer algorithms that do not provide information about real objects in the world, they can also be skilfully generated in a way that can embody artistic intentions. Consequently, I will argue that if GAN images and photographs become indistinguishable, then photographs will come to occupy a role in society similar to that of paintings today, in that they will lose their epistemic authority but continue to be valued aesthetically.