Vision as Collapse: A Typology of Recursive Forms from the Avant-Gardes to Generative AI

Violaine Boutet de Monvel, Ph.D.

Since 2023, AI developers have warned against model collapse – the contamination of training sets with synthetic content that progressively degrades model performance. Exemplifying a positive-feedback-driven failure, it produces effects such as semantic repetition or pixel noise. Yet collapse is not merely a breakdown: it also acts as a recursive mirror, raising the question of what happens when a system sees itself. Algorithmic vision no longer transmits the world (as in tele-vision) but increasingly generates worlds from within.
My project uses this dynamic as a conceptual lens to rethink the lineage of modern and contemporary art. Drawing on media archaeology and the digital humanities, I examine how recursion – both a computational and aesthetic principle – has shaped a distinct typology of collapsing forms, from the avant-gardes to generative AI. I argue that the latter – often framed as radically new – belongs to a longer history in which vision is not only extended but folded inward.
Marshall McLuhan once noted that Cubism bypassed linear perspective to encompass the total field – a pictorial approach that now finds a statistical analogue in the latent spaces of diffusion models. From the golden ratio and Fibonacci spirals to modernist abstraction, video feedback, and glitch, iterative patterns have structured artistic practice across centuries. By reactivating these histories through computer vision, the project thus seeks to situate generative AI within a broader genealogy of recursive composition.