Is It Possible to Create a Historian with Artificial Intelligence?

Suppose we gave an artificial intelligence every archive, every book on method and every history journal. Would a historian come out of it? The short answer is no. But thinking about why not says a lot about what history writing is.

What can the machine already do?

To be honest, artificial intelligence already does many of the historian’s tasks. It scans sources, summarises them, drafts translations from old languages and searches for patterns across millions of pages. As I wrote before, these are the most time-consuming parts of research: access and processing. From this angle, artificial intelligence can be seen as a research assistant that never gets tired.

What makes a historian a historian?

I think three things make a historian a historian, and all three are hard to entrust to a machine.

Asking questions. A historian’s questions come not from the archive but from the time they live in. Why now, why this subject, why this concern? All these questions are the product of a life and a curiosity. Croce’s famous sentence says the same: all true history is contemporary history.¹ If there is no “contemporary”, there is no question either.

The instinct of suspicion. A good historian first approaches a source with suspicion. Before asking “what is this document telling me?”, they ask “what does this document want to tell me?” Artificial intelligence works in the opposite direction: it produces fluent, confident texts. Learning suspicion is much harder for it than producing information. A system that invents sources is not a bad historian; it is not a historian at all.

Putting your name to it. Defending an interpretation, answering criticism, changing your view when a new document appears… This is where the ethics of the profession lies. A historian who is wrong corrects the mistake and, if necessary, apologises. Artificial intelligence has no self to apologise for its mistakes. At most, the company that made it issues a statement.

What if we had a model write history?

If we told a model fed with every archive to “write a history”, what would we get? I think we would get an average summary of existing historiography: a flawlessly compiled anthology that asks no new questions. Original ideas are born from the friction between historians and their own time. The machine has no time to rub against.

Master and apprentice

Could artificial intelligence one day replace the historian?

Perhaps the best analogy for this relationship is that of master and apprentice. Artificial intelligence is an apprentice who reads everything and never complains. The historian is the master who asks the question and signs the work. Could the apprentice one day become a master? If history writing were only a set of skills, perhaps. But history writing is also a responsibility: an act of interpreting between the past and the present, speaking in one’s own name. For now that seat looks empty. Who will fill it is the subject of another post.

Notes

1. In the Italian original: “ogni vera storia è storia contemporanea.” Croce, Teoria e storia della storiografia (1917). ↩

References

1. Bloch, M. (2013). Tarih savunusu veya tarihçilik mesleği (A. Berktay, Trans.). İletişim Yayınları. (Original work published 1949)
2. Carr, E. H. (1961). What is history? Macmillan.
3. Croce, B. (1917). Teoria e storia della storiografia. Laterza.
4. Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, 14045. https://doi.org/10.1038/s41598-023-41032-5