Infinite Barnacle: The AI Image and Imagination in GANs from Personal Snapshots

Abstract Today’s artificial intelligence image generation tools create images from datasets. These training sets are typically images sourced from the World Wide Web. However, artists may produce their own datasets from photographs. This essay explores one such process. In it, the artist discusses training a generative adversarial network (GAN) from images of personal memories. These images are shared here not as public artworks, but as personal photographs: snapshots reproduced and newly imagined by a machine. The essay explores the distortion that AI image generation introduces to memory and imagination, connecting ideas of photography to cybernetics to expose new ways of theorizing the image in the current stage of AI. It concludes that a theory of A imagery may borrow from theories of traditional photography but must examine its distinctions.