Latent Deep Space: Generative Adversarial Networks (GANs) in the Sciences

The recent spectacular success of machine learning in the sciences points to the emergence of a new artificial intelligence trading zone. The epistemological implications of this trading zone, however, have so far not been studied in depth. Critical research on machine learning systems, in media studies, visual studies, and “critical AI studies,” in the past five years, has focused almost exclusively on the social use of machine learning, producing an almost insurmountable backlog of deeply flawed technical reality. Among this backlog, one machine learning technique warrants particular attention from the perspective of media studies and visual studies: the generative adversarial network (GAN), a type of deep convolutional neural network that operates primarily on image data. In this paper, I argue that GANs are not only technically but also epistemically opaque systems: where GANs seem to enhance our view of an object under investigation, they actually present us with a technically and historically predetermined space of visual possibilities. I discuss this hypothesis in relation to established theories of images in the sciences and recent applications of GANs to problems in astronomy and medicine. I conclude by proposing that contemporary artistic uses of GANs point to their true potential as engines of scientific speculation.