Diffused Seeing

This article examines the transformation of the relationship between seeing and understanding in humans and machines by the technologies of machine learning known as ‘generative AI’. Taking Stable Diffusion as the main case study, while also looking at its competitors (DALL·E 2 and Midjourney), it starts by analysing the photographic infrastructure underpinning these generative models. The subsequent examination of ‘diffusion’ as a key concept that underpins the text-to-image generation process leads to some broader questions about the ongoing instability and dissolution of our current epistemological and political frameworks. Taking seriously the charge issued by some critics equating developments in generative AI with nihilism or even fascism, the article considers whether the current socio-technical moment can also offer some emancipatory possibilities. Images are used as part of the article not just by way of illustration but also to enact some of its argument.