Generic Pastness. AI Image Synthesis and the Virtualization of the Archive

In current debates, generative AI appears as a technology of the future. At its core, however, is our relationship with the past. By turning vast digital archives of past images into training data, AI image synthesis models such as Dall-E, Midjourney, and Stable Diffusion have transformed the history of visual culture into an exploitable resource. Each seemingly new image they produce is an interpolation of pre-existing images, a latent possibility within a virtual archive of potential images. At the same time, these models are fundamentally a-historical as they substitute any historical specificity with the generic »pastness« (Jameson) of recognizable visual patterns. The challenges these developments raise for historical image archives will be the seminar's focus.

Roland Meyer is a media and visual culture scholar specializing in the history and theory of networked image cultures. His most recent book, Gesichtserkennung (Wagenbach 2021), explores the cultural and social implications of automated facial recognition. His current research focuses on AI image synthesis, virtual image archives, navigable images, and the media archaeology of augmented spaces. Currently, he is a researcher in the CRC 1567 Virtual Lifeworlds at Ruhr University Bochum.

Participation online is possible through our VIMEO CHANNEL: https://vimeo.com/event/3726617

Scientific organization: Johannes Röll