People occasionally ask whether (or assume that) computers can currently read the kind of manuscripts I study.
The answer at present is: not well enough for the purposes of my field.
If you've a big (and I mean big) Latin work in early print or in a steady textura, it can be time-efficient to train a computer on it. This will give you a transcription that's error-free enough for [sniff] historians.
The more chaotic world of the cursive and cursive-derived script models used for many (often short) Middle English works is another matter.
That's not to say computers can't do this and won't one day do it. But—and here's a factor people often forget—who will pay for it? Can formidable pattern-recognition learn to handle a really gnarly anglicana hand? Very possibly. Does anyone want to put cash down for that? Well.
The field awaits the benevolent billionaire willing to ready AI for a tongue with more than 500 attested ways of spelling the word through, in order to extract money from a field of study where we struggle to get money for pencils.
