Why Latin needs AI more than AI needs Latin

The historian James Hankins once described the untranslated Latin corpus as a continent — a vast landmass of human thought that most of us will never set foot on, not because it's hidden, but because it's written in a language fewer and fewer people can read fluently.

When I built my first Latin translator as a freshman, I didn't have the vocabulary for any of this. I didn't know what a loss function was. I just knew that I might be taking my last Latin class ever, and that felt like a small tragedy I could do something about.

The translator was, and remains, nothing special. But the friction it removed mattered. A historian who can get a rough, fast read on a 16th-century treatise is a historian who can ask better questions of it.

That's the argument John Martin and I make in a forthcoming piece: the bottleneck on a huge amount of historical scholarship is not insight or access — it's reading speed in a dead language. And reading is exactly the thing these models do tirelessly.

I think this is one of the quietly important things AI can do for the humanities. Not replace the historian. Just hand them the continent.