Remember me
Artificial intelligence isn’t inherently good or bad. Like the printing press or the written word itself, AI is a force – and whether it enhances or undermines historical practice will depend on who uses it, how, and why.
AI’s most obvious advantage is that it can process vast amounts of data much faster than your regular historian, which turns it into an incredible asset. The trick here is to use it as it was originally meant to be used – not to replace human interpretation but to identify patterns that humans might not notice and, thus, to enable us to come up with better, more nuanced questions.
But it goes beyond that. AI is forcing historians to rethink what constitutes evidence, authorship and interpretation. In doing so, it is not just supporting existing historical methods, it’s helping evolve them.
Though I am not worried that historians will be replaced by AI any time soon, there is a risk that indiscriminate use might flatten out the practice. Good history requires doubt, contradiction, ambivalence, even uncertainty, but AI models are designed to produce smooth answers even when the data is messy or the past is unknowable.
Moreover, history isn’t just knowledge – it’s a method of knowing. If AI shortcuts the method, it threatens the discipline’s integrity. There is little doubt that AI will be part of historical research, but it is crucial that historians shape its use to enhance rather than undermine our craft. We should resist the urge to scale for scale’s sake, and instead use AI to deepen, not dilute, historical understanding. If we do that, and embed AI within the discipline’s values – critical thinking, reflexivity, contextual awareness – then historians will become central in answering the question of whether AI is good or bad for history.
We must experiment thoughtfully, push back critically, and make the case that slow, humanistic thinking still matters, even in an age of rapid computation. I am confident that the tables will turn, and we’ll soon talk about the ways in which history is good for AI.
Delfi Nieto-Isabel is a lecturer in medieval and digital history, and co-director of the Digital Lives Programme at Queen Mary University of London
“The best results come when humans work alongside machine-generated intuitions”In his classic Foundation series, sci-fi author Isaac Asimov explored the idea of statistically based ‘psycho-history’, by which the future could be predicted through the mathematical analysis of data on the past and present. With the rise of generative AI, it may feel as if a dystopian version of Asimov’s future is edging towards reality, but there are grounds to be a little more positive.
A great deal of what historians do is attempting to understand past societies by observing patterns in the evidence they left behind. Machines are spectacularly good at this, and the large language models of contemporary AI take this to a new level.
A bilingual funerary inscription in Latin and Greek – one of countless texts from the ancient Roman world that AI can search and compare (Image by Getty Images)
To take an example from my own area of work, texts inscribed on stone in a variety of ancient languages survive from the Roman world in their hundreds of thousands. These are often fragmentary and difficult to interpret. To make sense of them, both individually and en masse, as evidence for ancient societies and languages, historians need to compare these texts with one another. But to do so requires familiarity with them all, and the ability to find and compare parallels or identify patterns. Large language models specially trained on this material can now offer suggestions for missing words, dates and origins of this material, precisely because they see patterns across the material that a solitary human might find only after days or weeks of library work – or not at all.
The crucial point, however, is that these are suggestions – and the models can offer multiple suggestions, ranked by probability and exposing some of the underlying reasoning. The best results come when humans work alongside machine-generated intuitions. Correlation is not the same as causation, and history is a fundamentally human ‘science’. Nuance, context and, above all, humanity is fundamental to understanding our past, present and future. If that is surrendered to a machine then all of it is lost, and we are no longer writing our own history.
Every generation writes its own history, it’s said. Powerful AI can stimulate that process – but it certainly cannot write it for us.
Jonathan Prag is professor of ancient history at the University of Oxford, and a co-author of the Aeneas AI model for contextualising ancient inscriptions (predictingthepast.com)
“Fake ‘historical’ images and texts are already circulating and, with ever-greater sophistication, they will become harder to spot”Historians have been using AI for many years as an important tool for exploring and analysing the vast digitised and born-digital materials that increasingly constitute our primary sources. For instance, using AI it becomes possible to extract the names of all members of parliament mentioned in historical Hansard, identify the themes and topics on which they have spoken, and trace changes of emphasis over time. Handwritten text recognition technologies enable the accurate transcription of manuscript collections at scale. There are many other examples.
Hansard, the official record of British parliamentary debates compiled by reporters is another vast repository of information that AI can help interrogate (Image by Alamy)
This has been unhelpfully elided by the use of ‘AI’ as synonymous with generative AI. Machine learning, however, has not captured the public imagination in the same way as GenAI, with its easy-to-use chatbot interfaces. The speed with which GenAI has become near-ubiquitous, allowing no time to consider all of the ramifications, seems bewildering. Familiar software packages and media platforms are increasingly cluttered with invitations to ask an AI assistant for help. The AI-generated is entangled with the human-created in ways that undoubtedly pose real difficulties for the integrity of the historical record. Fake ‘historical’ images and texts are already circulating and, with ever-greater sophistication, they will become harder to spot.
This is a huge challenge, but it is also an opportunity for historians to apply their knowledge and expertise to a problem that requires particular critical skills. A 2025 study sponsored by Microsoft sought to identify those occupations most at risk from GenAI. Historians came second, behind interpreters and translators. Dredge operators, last on the list, can apparently be more confident about their future role.
But historians can help navigate this new knowledge landscape. Insightful source criticism, deep understanding of context, awareness of archival absence and loss, and a commitment to centring the human experience in the technological are key elements of historical research and practice. Arguably, they have never been more useful.
Jane Winters is professor of digital humanities at the School of Advanced Study, University of London
“Reading is thinking. Writing is thinking. These core skills cannot be outsourced without losing something fundamental”We seek to understand ourselves by exploring our pasts – and AI will not change this fundamental urge. The digital revolution has already transformed the historian’s work considerably, without fundamentally altering our purpose. Much change has been positive. Barriers of geography, cost and materiality have collapsed, broadening access to history as never before.
Less profitably, instant access and keyword searching deprive us of context, and we are less likely to understand the systems that sort, sift and deliver our sources. Always, we shape and are shaped by our tools in a relationship that must be carefully negotiated.
Utopians and naysayers are united in their confidence that AI will change everything. We must recognise such predictions as products of a moment of frenzied development and investment. Once the hype has died down, historians will adapt – just as we did to that other over-hyped but transformational technology, the internet.
The Maughan Library, King’s College London: “Students learn that critical reading and precision writing are the essential tools for engaging with the complex human past,” says Chris Sparks (Image by Getty Images)
We will use AI where it can help us – to automate repetitive tasks, to ‘distantly read’ sources identifying areas for closer inspection, or to hone our prose. But we will remain wary of its dangers – not least its tendency to hide its sources and make things up. Hopefully, new tools will become available that reduce our dependence on centrally controlled systems, allowing more thoughtful, bespoke and informed use.
We must think carefully about how we use AI in the classroom, too. Students learn that critical reading and precision writing are the essential tools for engaging with the complex human past. Reading is thinking. Writing is thinking. These core skills cannot be outsourced without losing something fundamental. They can be honed only with practice, and we must protect space for that in our teaching. Yet we cannot simply ignore AI, because we have begun to encounter its output routinely in our daily lives.
History is a project, not a product. If AI ushers in an era of fast, cheap, mass-produced text, then the skills that history teaches will only become more valuable.
Chris Sparks is a senior lecturer in digital history at Queen Mary University of London
“AI offers the chance to explore archives at scale, ask new questions and uncover new patterns"There are plenty of anxieties about AI. This should not surprise us: in the past, new technologies such as the railway resulted in moral panics. Rather than a threat, AI can be used – just as other technologies have been.
It is important to draw a distinction between functions sometimes described as ‘generative’ and ‘hermeneutic’. The former is what first comes to mind: it can be used to write text, including historical arguments. The latter is using AI to ‘distant read’ large bodies of texts, including archives.
Much criticism levelled at AI tends to focus on generative AI – for example, teachers’ fears that students will ‘cheat’ on essays, simply entering the assigned title and word length as a prompt, then leaving AI to produce a passable composition within seconds.
A Jewish woman holds a copy of fake papers enabling her to flee the Netherlands after the Nazi invasion: AI helps researchers search huge archives of testimonies by Holocaust survivors (Image by Getty Images)
Where AI has greater value for historians is as a hermeneutic tool. It is helpful to remember a point that Todd Presner makes to scholars in his 2024 book Ethics of the Algorithm: Digital Humanities and Holocaust Memory – that computers ‘read’ differently from humans, but neither better nor worse than us. The challenge facing historians is how to use this technology to ask new questions, particularly to interrogate vast archives that cannot be comprehensively searched through more traditional methods.
One area in which I, and other interdisciplinary teams, have been experimenting is using digital methods to access stories of Holocaust survivors in oral history collections. These are typically too large for any one person to watch in their life. The USC Shoah Foundation Visual History Archive, for example, contains more than 55,000 interviews, each 2–3 hours long.
Rather than replacing close reading – or watching and listening – of these interviews, AI offers the chance to explore archives at scale, ask new questions and uncover new patterns that invite close reading. For example, what pronouns – I or we – do men and women use to tell what is variously a solo or shared story? Where and why do silences occur in oral history archives as the past becomes too hard to retell? Critical use offers opportunities to do history differently.
Tim Cole is professor of social history at the University of Bristol
“AI promises to make history faster, but also shallower”AI is often hailed as the next great tool for historians. It can scan entire archives in seconds, collate references across databases, and generate polished prose at the click of a button. On the surface, this seems like liberation from drudgery. But, in speeding up our work, AI also risks hollowing it out.
History is not simply a matter of data extraction. The working historian does more than harvest facts: we interrogate silences in sources, wrestle with contradictions in established narratives, and read against the grain of the archive. AI machine tools excel at spotting patterns but fail at understanding nuance. They can tell you how often a phrase appears in a colonial newspaper, but not why irony mattered to the writer, nor how absence itself can be evidence. Outsourcing too much to algorithms risks confusing fluency with insight.
The Cullavagga manuscript written in Sinhalese Pali: AI assists historians mapping connections between texts in various languages (Image by Alamy)
The bigger danger, however, lies in how AI will shape the dissemination of history. Museums, publishers and even classrooms may come to rely on AI-generated summaries to teach the past. This could democratise access, but it could also flatten differences and sanitise conflict. History risks being repackaged into uniform, bite-sized ‘lessons’ optimised for clarity rather than complexity. That should alarm any historian who believes that the discipline’s value lies in argument and debate.
AI will certainly change the rhythms of our work. It helps me, as a historian of modern subcontinental Buddhism, to map far-flung connections between texts in English, Hindi and Sinhalese. But if I do not remain critical, it is all too easy to mistake the machine’s quick outputs for deeper interpretation. And if the wider public comes to accept AI’s versions of the past as authoritative, the historian’s role as interpreter and critic could be diminished.
The challenge for us is clear. AI will not replace historians, but it could deskill us if we let it. Indeed, it promises to make history faster, but also shallower. Our task is to ensure that the speed it offers does not come at the cost of depth – because history done fast is rarely history done well.
Bhadrajee Hewage recently completed a DPhil in history from Trinity College Oxford, researching trends in subcontinental Buddhism during the late colonial and early postcolonial periods
This article was first published in the December 2025 issue of BBC History Magazine