Capitalism is precisely the problem, because if the end product were never sold nor used in any commercial capacity, the case for "fair use" would be almost impossible to challenge. They're betting on judges siding with them in extending a very specific interpretation of fair use that has been successfully applied to digital copying of content for archival and distribution as in e.g. Google Books or the Internet Archive, which is also not air-tight, just precedent.

Even fair uses of media may not respect the dignity of the creators of works used to create "media synthesizers". In other words, even if a computer science grad student does a bunch of scraping for their machine learning dissertation, unless they ask and get permission from the creators, their research isn't upholding the principle of data dignity, which current law doesn't address at all, but is obviously the real issue upsetting people about "Generative AI".