From Sources to Models: AI, Data-Centered Practice, and Historical Knowledge in the Anthropocene

In recent years, archivists have grown used to being asked: "Have you digitized everything yet?" This deceptively simple question signals a shift in the epistemic environment of historical research: cultural heritage is increasingly expected to exist as machine-actionable corpora and, simultaneously, as potential training material for AI systems. This article argues that the most consequential transformation associated with digitization and AI is not a change of medium but a change of methodological form: the modeling layer of historical practice—data structures, extraction workflows, standards, and infrastructures—has become newly visible and therefore newly contestable. Conceptually, the article reconceives historical data as structured abstractions derived from sources rather than "givens", and treats data modeling as a primary site of critique where selection, typification, and uncertainty are operationalized. Methodologically, it challenges the assumption that only edition-grade outputs are worth producing and proposes graded "good-enough" intermediate representations, provided they are standardized, documented, and reusable. Against both technological enthusiasm and blanket refusal, the article positions AI chiefly as useful infrastructural labor while insisting that models themselves become objects of historical critique and governance. Finally, it connects these claims to the Anthropocene: scalable yet criticizable historical data infrastructures are conditions for producing public-good evidence relevant to ecological governance and to the defense of democracy, without collapsing historical knowledge into technocratic prediction. Narrative and interpretation remain indispensable for translating such representations into orientation under uncertainty.