The AI is reading text on unopenable scrolls based on miniscule differences in texture on areas that were inked. It was trained on openable scrolls where they knew what text was on the other side of them. The whole point of the article was to talk about how this makes the contents of the unopenable scrolls at all accessible, unlike prior because humans cannot read them. It feels like you didn't read the article.
Slapping an edit here: this comment was inaccurate. Compared to using AI for this, humans working out what the scrolls say is infeasible given how long it would take, but I don't think it's fair to say impossible