Wednesday seminar: The Meta-Archive of AI: Latent Commons & The Future of Collective Memory

This seminar is part of the Wednesday seminar series this autumn, where we explore a wide range of emerging methods and practices in artistic research. Here you can learn more about the series and discover upcoming seminars!

Together we will explore how art can help us approach AI not only as a technology, but as a space where culture, power, and imagination are stored, mirrored, and at times lost. Can art open new ways of engaging with AI as an archive – not just as a technical system, but as a site where collective memories and futures are negotiated?

Generative AI models, such as OpenAI’s GPT series, Stable Diffusion, platforms like Midjourney, and Google’s Gemini, are deep neural networks that approximate complex probability distributions over their training datasets to generate novel content. I argue that these models capture what I now call Latent Archives: a potential second-order archive that stores the distributions and correlations derived mostly from large-scale online datasets, enabling their novel re-generation. By encoding these structures in latent space, generative AI functions as a compressed repository of the underlying forms of cultural memory—a meta-archive that stores not the original data, as conventional archives do, but the patterns that condition its possible reappearance. While inevitably incomplete and shaped by the structural biases of their training data, these serve as reservoirs in which human culture is currently being mirrored, stored, and potentially reconstituted. Yet the main platforms offering storage and access to these GenAI models are corporations and private firms in the pursue of capital. What are the implications of this on collective, digitized memory when seen from the lens of a potential technofeudalist era?

Participant: Paola Torres Nuñez Del Prado in conversation with Robert Brecevic
Host: Petra Bauer
Profile Area: Art, Technology and Materiality