From Archaeology to Artificial Intelligence: Reconstructing, Verifying, and Interpreting the Historical Past in the Digital Age

Abstract

Historical method has repeatedly widened its evidentiary base, moving from narrative and textual criticism to material archaeology, bioarchaeology, remote sensing and, most recently, artificial intelligence (AI). This paper offers a critical, conceptual synthesis of that trajectory and asks what standard of verification historians should require before accepting AI-derived claims about the past. Drawing on peer-reviewed studies of LiDAR, radar and muon imaging, handwriting and text restoration models, and large language models, it argues that AI is best understood not as a single tool but as three modes of work: recovery of evidence, generation of conjectures, and synthesis of narrative or images. Each mode has distinct failure modes. Hallucination and bias are treated as structural properties of probabilistic systems trained on uneven records rather than as occasional malfunctions. Toynbee’s Challenge and Response is retained as a heuristic, complemented by Kuhn’s account of paradigm change and Daston and Galison’s history of epistemic virtues, and its limits are acknowledged. The paper proposes a five-level verification ladder that runs from data provenance through model documentation, reported uncertainty and independent corroboration to human adjudication, and it outlines a testable design for measuring anachronism in generative image models. The argument is conceptual, and the ladder has not yet been empirically validated.