Noam M. Elcott: The Mean and the Manifold

Genealogies of the AI Image

The guest lecture by art historian Noam M. Elcott examines a central paradox of contemporary AI image generation: The growing visual realism of AI-generated images does not necessarily correspond to greater informational richness. Elcott investigates this divergence, addressing fundamental questions: What notions of images, reality and seeing inform AI image production?
Over the last two years, outputs from advanced AI models have improved at an astonishing rate. At the same time, the default mode for AI image generation has rightly been dismissed as “slop.” Worse yet, as visual realism increases, data realism has decreased: AI images are ever more “realistic” and useless (at least as further training data). How can we explain the degeneration of AI image generation? The answers lie not only in technical details but also in ways of seeing — both human and machine.

Noam M. Elcott is a professor for the history of modern art at Columbia University. He is the author of Artificial Darkness: An Obscure History of Modern Art and Media (University of Chicago Press, 2016), as well as essays on art, media, technology, the law, and other subjects from the nineteenth century to the present, published in leading journals, anthologies, and exhibition catalogues. He co-directed The August Sander Project (MoMA/Columbia, 2016-2021), co-edited a special issue of Grey Room on “Art Beyond Copyright” (2024), and was a principal investigator of the Data Science Institute grant on “Art Images and AI: Latent Space Interpretability, Art History, and the Law.” His current book projects are August Sander’s People: Photography and Identity in Weimar Germany and Art TM: A History of Modern Art, Authenticity, and Trademarks.

The lecture will be held in English.
It is Organized by the Department of Art Theory.