The alternative for many of these un(der)appreciated books is that they will get unceremoniously dumped in the future anyway. The publishing industry and libraries etc dispose off lots and lots of books.
So at least with the AI companies they are scanning them and preserving them digitally. Not just in the trained weights, but also as raw training data for future runs.
P.S. I'm not sure why you need to make fun of your own ignorance? Just look up the word you don't know and don't mention it?
The AI companies’ working assumption is that if someone found it worth printing, it has enough information content to help train a model. That assumption might be invalid with some of the more rambling self-published books, however.