Abstract This essay by Antonio Somaini develops a critical theory of “latent spaces” — the compressed, vectorized mathematical substrates at the heart of contemporary AI models — and their impact on the ways in which the past is processed and interpreted. Opening with the Trump administration's AI-assisted purge of words and images from federal documents, alongside PragerU's production of deepfake videos for the White House's Founders Museum, Somaini argues that AI technologies are introducing a new regime of mediation that determines what can be seen, said, and known by encoding historical materials as vectors in abstract computational spaces. The article provides a technical and philosophical account of latent spaces as a new kind of Foucauldian historical a priori, comparing them with different understandings of the “archive” in its technical, institutional, and epistemological dimensions. Somaini then surveys a wide range of contemporary artistic practices — including works by Nouf Aljowaysir, Erik Bullot, Grégory Chatonsky, Taller Estampa, Alexander Kluge, Kevin B. Lee, Trevor Paglen, Gwenola Wagon, and Pierre Cassou-Noguès — that embody what he terms a “meta-archival impulse”: a critical desire to probe, expose, and resist the algorithmic processes through which AI models organize, transform, and sometimes erase historical materials. The essay concludes by arguing that the political challenge of our moment is twofold: developing critical methodologies for examining latent spaces as new infrastructures of memory and resisting the privatization of these meta-archives by a handful of hegemonic tech corporations.
Antonio Somaini (Thu,) studied this question.