DALL-E in Flatland: Illusion, Space, and AI-Generated Images

The release of multimodal generative artificial intelligence (AI) models in 2022, including DALL-E, Midjourney, and Stable Diffusion, has opened up new avenues to producing images. This new paradigm has implications for ideas of vision, illusion, and representation. Computer vision research has long used photographic images as a way to enable machines to see. Generative AI, on the other hand, uses data and techniques from computer vision to create images. This is possible thanks to the use of artificial neural networks that process vast datasets of training images, many of which are photographs, and the text associated with them. This article considers the relationship between image creation, illusion, and the notion of space in multimodal models. The photographic form of AI-generated images, coupled with a lack of understanding or experience of space undergirding their creation, reveals the illusion and artifice that is part of all photography.