A book for sale is copyright. I can’t make a duplicate of it and sell it. But I can read it and learn, and use that knowledge for life.
Why can’t an AI model be fed a paid book and learn from it? Say feed it an entire library of books.
You cannot “steal” information. Me attaining that knowledge does not deprive others of it. You could make an argument for copyright infringement, but honestly, who cares? Me asking an AI does not deprive others of say a book sale, because inherently I am unlikely to be interested in a book on that topic if I just want the magic black box to give me a possibly hallucinated answer instead.