If most people listen to music that isn't at the top of the complexity scale (can't prove that, but it sure seems likely) and 97% of people can't tell the difference.. does it matter at all that there are some minor limitations?
The only issue I can see with AI asset generation is that the models were trained by taking material the trainers didn't own. As soon as that is a settled thing (socially, by convention, law, Butlerian Jihad, I mean whatever) -- it's over folks.