While I don't think generative music AI is going to go away, the legal questions are likely to get ugly pretty soon (they already are, to be honest). What will likely happen is a system for managing and brokering "legit" training data deals with labels/publishers/rights holders and AI companies, along with a system for attribution. Having a background in generative AI myself, I know that datasets are the bread-and-butter of the industry—no data, no gen AI. So rights holders should technically be in charge. But regulation has to get behind them.
That said, I also agree with all the statements to the effect that the human element is still essential to music. These models are necessarily "behind", in the sense that they can do nothing without human data, and their "job" is only to replicate the underlying probabilities of that data. Sure, they can interpolate in that space, but interpolation is a far cry from the creativity of a human composer. We'll be in the driver's seat for a very long time, imho, because we're still the ones defining what to be in charge of—that is, what music is in the first place. Of course, as others have said, cookie-cutter music gigs will go away. But although those may be "a living", they're no way to live. 🙂
I also tend to think the prompt-to-create music generation stuff will become very boring for people very soon. It's impressive, and can be pretty funny at times (in the case of models like Suno), but since there's no cultural connection to a living creator, no "bio", no performer, no tours, no broader social relevance/zeitgeist, there's really very little to keep peoples' interest, long-term. My take is that ultimately people care about music because it's human, not because of what it sounds like.