- Who writes history in the age of AI?
During some point in the sixth century, a monk in the great monastery complex of Anuradhapura sat holding an iron stylus in one hand and a talipot leaf in the other. The characters he wrote were actually stamped in; lampblack was rubbed into the grooves so that they would become visible.
He was transcribing a history of kings and treasures and conflicts and droughts, the ‘Mahāvaṃsa,’ which has continued to be our source of information regarding our past ever since. But as he well knew, like any scribe did, the leaf would last no more than a few hundred years before it fell into decay, and the text had to be copied yet again, and again.
In December, I saw an Artificial Intelligence (AI)-generated video about a character whom I have dedicated many years of my working life to researching: Lewke, the Kandyan chieftain whose name was bestowed upon the magnificent cannon which was recently restored to its homeland of Sri Lanka.
This video was a masterpiece; it was fluent, vivid, expertly crafted – the kind of video that captures the attention of any audience and spreads rapidly across many phones. Yet it was incorrect. Those who created it, as well as the AI that helped them to do so, clearly knew nothing about the character of Lewke.
Information about the chieftain was mixed together with information about other characters, facts about him twisted into something that sounded accurate yet was far removed from reality. Most disturbing of all, however, was the discussion below the video; thousands of people were watching this fictitious Lewke and taking it as a historical truth.
It matters more than it ever has. But to see why, we have to be honest about something uncomfortable. History has never been written by ‘the past’. It has always been written by somebody, from somewhere, for some purpose. AI does not abolish that old truth. It disguises it.
A tradition that knew it was partial
Sri Lanka has one of the longest continuous historiographical traditions in the world, and we are oddly shy about saying so.
We have the ‘Dīpavaṃsa’ and the ‘Mahāvaṃsa,’ written and rewritten at the Mahāvihāra, which trace an unbroken line of history going much further back than most other histories of comparable nations do. The ‘Cūḷavaṃsa’ took the thread forward from there through many more rulers.
Alongside the history compiled by the monastic vamsas, there was the tradition of the puskoḷa pot, where medicine, astrology, poetry, ritual, and law were taught through palm leaves in the gurukula system. This was learning handed on person to person, body to body, watched over by one who could be held accountable for it being done properly.
This is where the truth comes in. The writings were partial, for they were written by Buddhist monks in Pali, mostly concerning the kings, and mostly to emphasise the place of the island and its lineages in the middle of the dispensation. There are silences in these writings: on everyday labour, on women, on communities that would not fit into their story. No responsible historian will ever read these sources other than critically.
And yet this very partiality bears a lesson that we may not fully grasp. The writer of the text knew he was a writer. He had a name, a monastery, a lineage, and a position. This text could be challenged, it could be contradicted, it could even be reproduced with variations in the hand coming after his. There was authorship, and there was accountability in it.
That self-awareness is precisely what the algorithm strips away.
Europe, Asia and the politics of the archive
Place our tradition alongside that of Europe, and the contrast is educational rather than humbling. We have Herodotus, the curious wanderer, and Thucydides, the icy scientist – two personalities who remain in conflict within every historian until this day.
However, the most crucial step in the European tradition emerged later in the 19th century in Germany when Leopold von Ranke made the argument that history should be based on documents, which would inform us of “wie es eigentlich gewesen,” or how it really happened. This is when the reign of the archive truly began, transforming history into a profession.
That intuition came across in the holds of company ships. The Dutch maintained their thombos, their careful record-keeping of lands and households; the British superimposed their survey and ordinance reports.
When I conduct my work as a provenance researcher, reconstructing how some piece of Kandyan art came to be in a museum in Amsterdam or Berlin, I am engaging precisely these documents, and I am repeatedly brought to realise that they were created for the purposes of ruling us, not remembering us.
They documented what the colonial administration required for taxation, control, and classification. Meaning existed beyond the document. To narrate our story from these materials alone is to permit the coloniser to hold the pen forevermore.
Third way, third warning: Asia provides the third way and an even more pointed lesson. China established the most bureaucratically precise historical tradition in the world, with royal historians and special bureaux creating a steady stream of dynastic histories. Integrity was valued in the tradition to a terrifying extent.
It is said that the Grand Historian of the State of Qi, having noted in his chronicle the murder of a prince by one of his ministers, was slain for the act. His brother repeated it and was killed too. Another brother recorded the same and the minister, being unable to kill all three, allowed the truth to remain. By then, however, yet another historian was already on the road, scrolls in hand, just in case.
For its part, India long suffered from a reputation of being ‘a civilisation without history’. This myth was utterly dispelled by a book like the 12th century ‘Rājataraṅgiṇī’ of Kalhana – ‘the river of kings’ – an invaluable record of Kashmir.
There are three ways of looking at who makes history – the responsible scrivener, the collection of documents, and the historian prepared to die rather than lie about history. The common factor that binds them all together is that there was always a person involved in making history accountable to them.
The scribe with no lineage
What then are we inviting into our midst? A large language model writes the history in beautiful prose, without any lineage at all. There is no monastery, there is no name, there is no master that you can hold accountable, there is no stake in being accurate. It doesn’t know anything; it just generates the next best thing to say.
When it doesn’t have a piece of information, it doesn’t become silent as an honest historian would. It invents sources, inventions, dates, and even people – including historians who write books – just as easily and confidently as it would the truth. We have already seen fake citations used in legitimate works, as well as ‘historical photographs’ generated by AI circulating as real images online.
And yet, that is precisely why the video about Lewke may be both incorrect and compelling at the same time. Why would it not have any information about him?
This leads to the key issue. Training of such a system involves the overwhelming use of texts that are already available in English form and digitised, which means overwhelmingly, texts from the archive of the dominant North.
The man’s existence is in Sinhala documents, in Kandyan archives, in scholarly works about the repatriated cannon. And so, when the authors of the video wanted it to tell Lewke’s tale, it didn’t have much material to work with. It plugged the gap with assured creativity. They couldn’t spot the problem themselves because they did not have any knowledge of the past.
The viewers would not be able to see the mistake because the video was so convincing. Whatever I could recognise, it was due to my decades of working in archives that the machine was not even aware of.
And there lies in summary the future we are sailing into – a smooth, uninterrupted planetary scribe reproducing silences and biases of the colonial archive faster than ever possible and convincing everyone that, finally, memory has become impartial.
And that is precisely what I do not want from our future.
Holding the stylus
I do not consider myself a pessimist, however, because there is no reason why leaf and algorithm should become enemies. Used as a tool instead of being seen as the creator, AI could become the greatest ally of our threatened history in a generation.
Just think about the possibility of computer technology that is programmed to decipher and translate Sinhala, Tamil, and Pali scripts, combing through our countless palm leaf manuscripts before they fall apart. Think about colonial records from Colombo, The Hague, and London all being searched together, so that the entire story of an artefact, like that of Lewke, could finally be told instead of speculated about. Think about the potential to record and understand oral traditions before they disappear forever.
The machine that would strip away our history could also bring back what we have lost – but only if we choose who gets to wield the stylus.
Consider what is already being done. In a 2023 study in Science Advances, Dzemila Sero, working with Isabelle Garachon, Erma Hermens, and Kees Joost Batenburg across the Rijksmuseum, Leiden, and Cambridge, scanned a small terracotta, a study for a hovering putto attributed to the sculptor Laurent Delvaux, using three-dimensional micro-computed tomography.
The scan did not merely produce a tidy digital replica. It captured the fingermarks and tool marks the maker had pressed into the soft clay, not only on the visible surface but on the hidden inner walls, and characterised them quantitatively. From the breadth of the finger ridges, the team could even estimate the age and sex of the hand that shaped the figure.
In earlier work, Sero, Garachon, Hermens, Robert van Liere, and Batenburg had already mapped the wider promise of three-dimensional fingerprint recognition for cultural heritage.
This is computational imaging doing what no archive of inked words can possibly accomplish: bringing back into view the body of the craftsman, whose actual fingerprint cannot be found in the forgotten documents that failed to name him. Here, the computer doesn’t obliterate the scriber; he is brought back to life.
Think about turning this process on an artefact from Kandy made of ivory, or metal from a temple.
Think also about the ongoing efforts to digitise the ancient Hindu-Buddhist stone sculptures in Dutch museums made between the eighth and 15th centuries and to develop an intentionally inclusive database out of them featuring an AI tool prototype tested first on the statues of calm-faced Buddhas from Java to aid the research of works moved far away from their place of creation.
Who they are meant for plays a crucial role in the process. Most of them were taken from Indonesia during colonial times while some are now coming back under a repatriation programme. To digitise them, and to build the intelligent tools that interpret them, on the source country’s own terms and for its own people, is to refuse the old arrangement in which the North holds both the objects and the authority to explain them.
This is the model we in Sri Lanka should claim. We should not wait, cap in hand, for a distant machine to teach us by accident. We should build our own.
This, in my view, is the kind of path that I would like to advocate. For a start, we need to remember that AI is a tool only, and never a witness to anything, as every assertion by this artificial being should have some basis that can be verified, just as the claims of the scribe could be examined, but those of the Grand Historian, though subject to death, were not corrupted.
Moreover, we need to digitalise our own archives according to our conditions, in such a way that the data on which the machines of tomorrow will be trained shall include the views of our ancestors, not as a footnote from the databases of the North, but in their own languages and with their own custodians.
Furthermore, there should be an effort to adopt the ethics of provenance studies in the age of data and ask questions on where the information came from, who compiled it, for what purpose, and who is missing in it.
The monk in Anuradhapura knew his leaf would rot. He copied it anyway, because he understood that a civilisation survives only through the deliberate choice of each generation to carry its memory forward and to put its name to the carrying.
The algorithm makes that choice feel unnecessary. It is not. The question is no longer whether the past will be written, because the machines will see to that, endlessly and effortlessly. The question is whether we will still be its authors, or merely its subjects.
Who writes history in the age of AI? The same people who always have: those willing to take responsibility for what is remembered, and for what is allowed to be forgotten. The wrong Lewke is still out there, travelling from phone to phone. Somewhere, the true one waits in an archive no machine has read.
The stylus is still in human hands. We should be very careful before we set it down.
(The writer is an ethnographer and art historian specialising in material culture, critical museum studies, and participatory research methodologies. As a member of Provenance Research on Objects of the Colonial Era [PPROCE], the Netherlands and Dutch Research Council [NWO] committees, she engages with repatriation and decolonisation debates. She currently holds a NIAS-NIOD-KITLV Research Fellowship in the Netherlands)
(The views and opinions expressed in this article are those of the writer and do not necessarily reflect the official position of this publication)