How AI is unlocking ancient texts — and could rewrite history

[rcjordan]

https://www.nature.com/articles/d41586-024-04161-z

[ergophobe]

I follow this vaguely and it is amazing that it is finally starting to happen. It's still very much "human in the loop" and too expensive to roll out generally, but it's coming.

Of course, in 1998 when I first started teaching paleography, I told my students they would be the last generation to need to study this because I expected it to be automated by 2020. They laughed. They thought it was absurd. I turned out to be way off.

Like the self-driving car people, I looked at how fast we went from unable to do OCR on type to reasonably accurate handwriting recognition and just failed to understand the degree of difficulty in deciphering an old handwriting in an old or dead language. And as with self-driving cars, it seems so obviously naive in retrospect. Obviously I should have known better.

In 2021 when I told my students the above story they laughed, but when I said *they* might be the last generation who needs to learn this, they did not laugh. They were more, "Hello? Tell us something we haven't thought of."

It might be one of those things like nuclear fusion that always feels 10-20 years away, but a lot of these projects show promise that we are less than 20 years away from scanning a lot of text and having some AI SAAS return the transcribed text, especially if it has some basic information to start from (language, time period, location).

Automatic transcription as a commodity by 2044?