Reimagining Historical Research in the Age of Artificial Intelligence

The integration of Artificial Intelligence (AI) into the practice of historical research is no longer a hypothetical prospect. It is unfolding with growing momentum, raising pressing methodological, epistemological, and ethical questions for scholars across the humanities.

This post introduces a paper that Christian Henriot developed following the AI in Science workshop organized by the European Commission’s Joint Research Centre in May 2025. The event opened with a schematic model of the “Scientific Process,” intended to stimulate reflection on AI’s role in contemporary research. Yet, from the vantage point of the humanities—and history in particular—this model proved inadequate. It failed to capture the iterative, interpretive, and non-linear dimensions that characterize historical inquiry, as well as the intricate interplay between human judgment, computational processes, and emerging AI systems such as large language models (LLMs).

In response, the paper proposes a new conceptual and practical framework: a nine-step AI-augmented workflow designed specifically for historical research. Developed through both theoretical reflection and empirical experimentation, this model maps how LLMs can assist scholars in key research stages—ranging from discovery and analysis to writing, visualization, and translation—while emphatically preserving the central role of critical, humanistic expertise.

The AI-Augmented Historical Research workflow

Rather than a prescriptive formula, the workflow is offered as an ideal type: a scaffold for adaptation, refinement, and scholarly debate. The model is anchored in an original case study drawn from the author’s own research, illustrating how AI tools can be integrated into a rigorous, context-aware historical methodology.

A defining feature of the paper is its insistence on situating “AI” within its broader intellectual and technological genealogies. Current discourse often collapses the category of AI into conversational LLMs, risking both conceptual confusion and analytical overreach. By recovering the field’s longer arc—from symbolic logic and expert systems to today’s probabilistic generative models—the paper offers a clarifying perspective on what these technologies can (and cannot) do within the epistemic frameworks of historical scholarship.

The Evolution of Artificial Intelligence (click on link for the online version)

Conceptual imprecision, it argues, leads to methodological error. Without a grounded understanding of how LLMs operate, there is a heightened risk of either overstating their capabilities or applying them in ways that obscure rather than illuminate historical complexity.

The paper further engages with ongoing debates about digital history, particularly the shift from static textual corpora to dynamic, processable datasets. It revisits earlier protocols for applying Natural Language Processing (NLP) to historical sources and situates the uptake of LLMs within this evolving methodological landscape. While these models lower the technical threshold for computational analysis, the paper underscores the continuing value of programming literacy—not only as a means of maintaining autonomy over one’s tools, but also as a lens through which to understand how historical questions are translated into computational tasks.

Finally, A distinctive feature of the paper is the inclusion of a detailed case study drawn from the author’s own research. This section moves beyond hypothetical applications to illustrate how AI systems, especially LLMs, can be meaningfully integrated into live scholarly work. The case study focuses on a historical inquiry — A Tale of Three Merchants— involving complex multilingual corpora, fragmented historical evidence, large-scale textual analysis, and use of various methods (topic modeling, netwoirk analysis). It demonstrates how LLMs were deployed not merely for surface-level automation, but to augment deeper cognitive tasks: hypothesis refinement, interpretive pattern recognition, and multilingual contextualization.

The AI-Augmented Case Study Workflow (click on link for the online version)

Crucially, the study reflects on how human judgment and machine output were iteratively entangled—sometimes reinforcing each other, sometimes creating productive tension. The result is a rich account of what it means to “think with machines” in historical research, foregrounding epistemic responsibility, tool transparency, and domain expertise.

This paper positions itself as both an invitation and a provocation. It seeks to broaden the scholarly conversation around AI in the humanities—not by proposing quick solutions, but by offering a framework for critical experimentation. For historians and digital humanists invested in methodological innovation, the paper provides both conceptual scaffolding and practical orientation.

The full text is available on HAL-SHS or on ArXiv. Academic readers with interests in computational methods, historical research, and AI ethics are warmly encouraged to engage with the argument and contribute to the dialogue it initiates.

---

---