Synthetic Pasts: The Generative Historicity of Artificial Intelligence
Generative AI models trained on massive amounts of historical data do not merely imitate the past — they articulate it. By synthesizing text and images conditioned on pre-existing cultural material, these models produce what we call “synthetic pasts”: outputs that, while entirely artificial, retain an uncanny historicity derived from the epistemic and aesthetic structures embedded in their training data. Synthetic Pasts deploys an experimental approach where we train and interact with open-source pre-trained models (Stable Diffusion, Mistral, Llama), while posing questions of critical and historical nature concerning machine-generated historic texts and images. The project provides a controlled research environment meticulously designed to observe and document the training process of generative AI, with training data ranging across historical literary fiction, art and comics, scientific images and texts from the 19th and 20th century. Drawing on posthumanist media theory, media archaeology, and Hayden White’s concept of the practical past, the project treats the outputs of these models as epistemic machines: probes capable of revealing patterns, biases, and probability distributions latent in cultural data that would otherwise remain invisible across a longue durée timespan. Through an iterative methodology integrating computational experimentation with humanistic analysis, the research develops what we call “synthetic hermeneutics” — a co-interpretive framework at the interface of human and machine cognition. Synthetic Pasts is a collaboration between Uppsala University (Department of ALM / Centre for Digital Humanities Uppsala), Linnaeus University, and Echo Chamber ASBL (Brussels), bringing together an interdisciplinary team: Anna Foka (PI), Matts Lindström (Co-PI), Ilan Manouach (Co-investigator), and Per Israelson (Co-investigator) — working across media theory, media archaeology, literature, computational creativity, and artistic research.