Artificial Intelligence offers powerful new opportunities for Digital History, particularly in the analysis of large, structured datasets. At the same time, it risks obscuring what historical data fundamentally consists of: uncertainty, context dependence, and interpretation. This post introduces an iceberg model that distinguishes between visible, formalised data (names, places, events) and the largely invisible layers of historical meaning beneath the surface. Using the prosopographical research database REPAC as an example, it shows how AI-supported methods, especially Retrieval-Augmented Generation (RAG) in nodegoat, can help to address uncertainty without flattening it. AI thus becomes not a substitute for historical interpretation, but a tool that more tightly connects data, context, and research questions.
Kaspar Gubler (2026) studied this question.