Back to Basics? Historical Methods and the Rise of AI

If you ask ChatGPT a medical question it will stress, repeatedly, that it is not capable of diagnosis. Although it may offer some general answers, it will ask first for further clarification, and then ultimately advise you to consult a trained professional. It appears that it has, in fact, been programmed to state explicitly that it can’t provide definitive answers to medically-related queries; lawsuits would abound were it to make a mistake. For questions about history, though, ChatGPT makes no such disclaimer. Ask it – as I did – a broad question like “What caused the British Empire’s growth?”, and it will not ask for further clarification or put caveats on its knowledge. In my case, it launched authoritatively into what it called a “clear breakdown” of factors, offering economic mercantilism, geopolitical rivalry, and the civilising mission. Yet its “breakdown” was far from complete. Alarmingly, it made no mention of the slave trade; it listed the imperial dependency on “trade and resources,” citing commodities like “spices, sugar, cotton, tea, and, later, minerals,” but quite conspicuously left out “people.”

Clearly, the stakes are higher if ChatGPT were to misdiagnose an illness, rather than misrepresent history. And yet, in recent years, certain historical subjects (particularly as they relate to issues such as slavery, empire, and race) have been co-opted by a resurgent populism, inflamed public opinion, and reshaped our political discourses. If nothing else, this should prove that there is a social and cultural imperative to investigating and discussing the past properly. At the very moment when the demand for sensitive, dispassionate, evidence-led history (the kind of history practiced by academically-trained historians) should be at its highest, AI – with its tendencies to flatten historiographic complexity, invent secondary literature, and gloss over crucial information – has emerged as a further threat to the standing of an already embattled profession.

“Tree of library classifications,” unknown artist, possibly 1900. Image via Library of Congress.

Last summer, news outlets rushed to report on the findings of a Microsoft study which identified forty occupations whose typical activities were, it claimed, most replicable by generative AI. Under media spin, the findings were contorted into two lists: jobs which are “AI-safe” and jobs which the rise of AI imperils towards obsolescence. Historians ranked second on the latter. For the authors of the report, historians are at risk of profound AI-driven transformation because they perform “knowledge work”; the collection, retention, and provision of information about the past, which AI can replicate. AI, the report conceded, is not particularly good at analysing information, only at searching for and summarising it; but, importantly, it did not consider history to be a discipline where substantive analysis is required. To make such a conclusion about historians is only possible if you take the craft of history – as those researchers did – at its most basic methodological level: facts from the past, assembled from an easily accessible trove of pre-existing (and principally digital) information, spun into a readable narrative.

A rebuttal to these ideas formed the basis of my recent article in Transactions of the Royal Historical Society, “The Historian in the Age of AI.” I connected the belief that AI can take on the job of a historian with a wider and more deep-rooted conflation by the public of academic historical research with generalized historical writing. Books by politicians, journalists, comedians and social media influencers fill the shelves of history sections in bookshops – some of these authors are consequently accorded the title of ‘historian’, even though their approach is quite explicitly to synthesise and repackage the knowledge generated by other people’s scholarly labour (and sometimes without any reference to it). The result is both to devalue the long training which academic historians undertake to investigate and argue from the past, and to obscure the importance of possessing that training and expertise in the first place. This is not to suggest that the public should be entreated to read academic monographs over more accessible historical writing; but we should recognise the disconnect that results when some forms of popular history are cast as the work of historians equivalent to the scholarly endeavour from which their information stems.

It is in this space of confusion about precisely who qualifies as a historian that AI chatbots have been able to thrive as seemingly knowledgeable providers of historical information. ChatGPT, for instance, gives the appearance of knowing far more about niche historical topics than we might expect; but when pressed on some of its claims and sources, it very quickly folds and defers to the user’s corrections. It is not able to detail accurately how it came by its knowledge – only to sound authoritative at the first point of contact. For users who are ill-informed on the methodologies of history – and, particularly, why those methodologies are essential to understanding the past – that absence of reliable provenance might seem trivial. AI can so seamlessly take on the mantle of a trusted purveyor of historical information because, perhaps, some people are already so used to viewing history only in its most watered-down form; not as the skilled investigation and analysis of the past, drawing on years of training and refined expertise, but as the readable conveyance of popular historical fact, drawing more on skills of accessible communication. Yet, when dealing with historical issues that have charged political and ideological currency, understanding this distinction between the methods of academic and popular history – and the motives of the people who wield them – should be vital.

In this respect, some of the responses to “The Historian in the Age of AI” have been illuminating. Although I stressed that the methods and skills of academic and popular history are connected, but different – and that therefore one cannot be superior to the other – for some, the solution to the rise of AI is to return to the academic elitism of earlier decades. If AI threatens the primacy of the academic historian as the trained professional charged with researching and explaining the past, then the response should be one that more clearly re-asserts the prowess and authority of the scholarly community. But such a move risks alienating an important group of historical researchers who have either no institutional affiliation or formal academic historical training; those who conduct essential scholarship as independent researchers, or autodidactically practise the methods and craft of academic historical research.

More significantly, though, I received an interesting pushback from scientists who have applied the methods of their own discipline to the study of the past. For some scientists AI is just another tool that makes their processes of, for example, data modelling, much quicker. One email even likened the rise of AI to the rise of the word processor: a tool which entrenched typewriter users resisted at first, but eventually had to adopt for its greater speed and efficiency. Because AI can extract information from documents and books at a faster rate than the human brain, for some it is a shortcut to circumvent the deeply-considered qualitative research that most historians are trained to undertake. But primary sources are not just repositories of data to be mined and tabulated; they are fragments of past attitudes and sensibilities that need careful and sensitive reading. AI – with its absence of emotional engagement and its inability to immerse itself in the fabric of the undigitized archive – makes a poor companion for those seeking to understand our human past.

Is this an opportunity, then, to go back to basics? Much of the defence of history as a subject and a profession has been, rightly, calibrated for higher education’s marketisation; to assert the employment opportunities afforded by a history degree and calculate the economic value of history departments within universities. But AI poses an altogether different threat: to become a trusted source of historical information at the expense of historians’ expertise, when it offers nothing approaching the care and rigour of a professional. Perhaps, then, we need to detail what historians can do that AI cannot: to articulate that there is such a thing as a ‘historical method’; to explain why the training which historians undertake is so essential to the recovery of the past; to detail in greater clarity how the archival labour of historians becomes, by degrees, the information contained within popular history bestsellers; and so to reveal the foundational nature of historical scholarship – wherever it is undertaken – to the collective interrogation of our shared past. For if AI threatens to strip historical research and writing back to its most basic form, perhaps that’s a challenge we should meet head on.

---

Chris Campbell is a PhD Candidate in Modern British History at the University of Cambridge. He is currently researching British soft power in the twentieth century, focusing on the British Council and its professionalisation of cultural diplomacy.

---

The views and opinions expressed in this post are solely those of the original author/s and do not necessarily represent the views of the North American Conference on British Studies. The NACBS welcomes civil and productive discussion in the comments below. Our blog represents a collegial and conversational forum, and the tone for all comments should align with this environment. Insulting or mean comments will not be tolerated and NACBS reserves the right to delete these remarks and revoke the commenter’s site membership.