Automating history? Conceptual pressures at the foundations of historical practice

The use of Artificial intelligence (AI) in historical research, especially in the form of Large Language Models (LLMs), is on the rise. LLMs are now used for a range of purposes, such as enhancing OCR text, transcribing oral history interviews, or exploring vast corpora of historical sources. Further uses are also being experimented with, from improving existing digital methods (e.g. argument mining) to corpus compilation. As these examples show, multiple stages of the historian’s process are increasingly affected by these famously opaque technologies, from initial research through to dissemination.

Nevertheless, some suggest that AI is merely the newest technological development available to historians, presenting familiar methodological challenges in new forms. In this paper, I argue that this characterisation is misleading, as the ways in which AI not only facilitates but automates research tasks presents us with unique and novel challenges. In particular, I contend that the automation of interpretation- and analysis-related tasks places pressure on fundamental concepts we use to understand the production of historical knowledge, including the central role of ‘historians’ themselves. Indeed, the rise of AI has already led some scholars to suggest that machines can undertake such tasks and may themselves qualify as ‘historians.’ I counter these claims, arguing that they focus too heavily on the final products of historical work, rather than on the processes through which they are created. More productively, I suggest that these conceptual pressures force us to articulate precisely what the role and activities of historians involve in order to determine which of these we can and should seek to automate.