Will AI Replace Historians?

Just Give Me the "Dirty Purple"

When I was teaching world history (a very challenging subject to teach!), there was always a lesson and subject that took me out of my comfort zone. For example, what do I really know about the Byzantine Empire? It’s not a subject I’ve studied in depth. So we instructors would joke: just give me the “dirty purple,” a reference to old mimeographed notes, a standard summary of the basics of a subject, with no depth behind them.

AI is a bit like those old “dirty purples.” It’ll give you the basics on a subject. Of course, AI is also more flexible and powerful than that. But AI draws from what’s already out there, and its responses are based on, shaped by, the queries you put to it. AI, as powerful as it is, is a limited and fallible tool.

There’s been talk of AI replacing historians, and I suppose that’s based on a misunderstanding of what historians do. It’s that old image of Professor Dryasdust, the one who teaches history like he’s reading from an encyclopedia. Lots of names, dates, facts. Memorize this, regurgitate that. And I suppose an AI program could do that as well as Dryasdust. Perhaps better.

Sea biscuit?

But historians aren’t walking talking encyclopedias. Sure, as a historian, you have to start with the facts. But history is really about judgment, discernment, sensitivity that’s developed over time. It’s about immersion, research, grunt work, effort. And, if you’re teaching history, it’s about enthusiasm—conveying your interest, your excitement, your knowledge of the subject to students in a way that resonates with them, perhaps even inspires them. I can’t see how AI can do that.

History can also mean getting down and dirty with objects. With things. History can be tactile, thrilling. Want to know about primitive weaponry? You can read books, of course, but why not visit a museum, handle the weaponry, perhaps even fire them. Alternatively, history is enriched by travel and a sense of place. I’ve learned a lot about the Manhattan Project and the atomic bomb, but my knowledge became more meaningful to me when I visited Los Alamos and walked “ground zero” at the Trinity test site in Alamogordo, New Mexico.

I worry that the very skills we are most proud of as historians are the ones the bean-counters and “patriots” seek to replace. So, for example, historians are supposed to be fact- and evidence-driven, not driven by agendas, biases, etc. We are not supposed to glorify America, explain away slavery, and support a narrative of American exceptionalism and greatness.

Historians are also all about uncovering new facts, developing novel interpretations, and that’s not AI. AI is good at summarizing quickly what is already out there. AI doesn’t create new knowledge. AI doesn’t travel into dusty archives to uncover neglected manuscripts; it doesn’t travel to far-flung places to interview witnesses to historical personages and events.

Good historians follow the facts and ask new and provocative questions. That is not what AI does—but what AI does do well is perhaps what some administrators, politicians, etc. want from historians. Quick summaries and “lessons,” very conventional, in response to limited queries.

In short, too many people just want the dirty purple. They want the basics, devoid of nuance, subtlety, and controversy. A dirty purple that supports preselected agendas. For sure, AI can provide that: too bad it’s not in any sense real history.

This is not to say AI is useless; as it matures, it will doubtless become more flexible and powerful. But tools like AI must be used cautiously. We’ve probably all had the experience of using a power tool that bit us because of our inexperience. Think of AI in that way, as a tool to be used with considerable care, one that rewards judicious use by skilled operators.