What's missing in the output of the music generation systems is any sort of true human variation or individual spark. Even if you get it to emulate Bob Dylan, it won't be able to add any quirk or defect that a real artist would inevitably produce either on purpose or by accident. It suffers from being too perfect and as more of this material is generated, it's flat, uniform nature will form noticeable patterns. A reasonably seasoned listener will just go "oh, that's AI" and quickly dismiss it as its value is no different to a plumbing flyer stuffed in your letter box.
Counterpoint to that: I do not find that problem (lack of variation, unifortmity) with AI-generated music. It doesn't learn music theory, recording best practices, then Bob Dylan and produce a sterile Bob Dylan clone. It just learns the waveforms and the frequency and time components. Teach it with off-key singing here and there, it's going to produce off-key singing here and there.
But more realistically, it get taught with a huge variety of music, with with reinforcement to know how to classify it and figure out what traits make a certain style and what what part is singing and what classifies a male and female singer, and such, what tempo is, etc. I've had it—without me specifying—generate sections with accelerating tempo, then dropping back, and using that scheme as an element in the music. So, it will know what you mean by Bob Dylan (though you typically need to describe a style that is like Bob Dylan, and it's blocked from calling or recognizing it as Bob Dylan). Also, I've given it my own completed music and had it reimagine it in made different results. Unexpected things like having the song seemingly end, in full force...then quietly come back in and develop a variation on a theme...
Counter to being too perfect, a lot of the output can be a little rough, like a group of capable musicians working out a idea on the fly, that a person might want to refine and edit before it's a completed song.
Again, my interest is more along the "assistance" use of AI, but ultimately I'd love to have it be a producer that could make suggestions of perhaps several different directions for development. Most of the biggest artists rely on that, and have for a long time. Michael Jackson would come in beat-boxing and idea vocally, and others would take it from there. But there are also capable musicians that benefit from the right producer. Billy Joel struggled, even with hits, until he worked with Phil Ramone, who was able to convey more of the artist in the recordings. And of course the Beatles and George Martin were a perfect match—they probably would not have had the breadth of style without him.
Finally, most of the music people hear are borrowed or copied from someone else, have been for a long time. Surfin' USA is a Chuck Berry song (Sweet Little Sixteen) with the lyrics changed. Dazed and Confused was just a stolen song entirely (Jake Holmes). Bach's Air on a G String is a vital component of A Whiter Shade of Pale. Sowing the Seeds of Love by Tears For Fears doesn't copy a particular song, but tried to get as close to the Beatles in general as it can.
But I agree in general that it doesn't replace true inspiration, it just tries to sound like true inspiration by listening to true inspiration (at some point it will be difficult to differentiate, though). It's also fed with a lot of routine boring music, because people buy routine, boring music. Younger listeners and those less sophisticated older one want to hear songs that. essentially, they've heard before. They don't want to be challenged.
By "challenged", I mean...I grew up playing classical piano, and was eventually drawn to progressive rock. Offhand, a single that I liked from the first verse was Selling England by the Pound, and it was by very first listen to Genesis—a friend had recommended the album. But when I bought The Power and the Glory, by Gentle giant, I had their albums and knew what to expect, and when I played the first song, Proclamation, I knew right away it would take several listens before it sounded good. It was like eating a new food that's kind of bitter tasting, but knowing that it will be one of your favorites when you get used to it (and in fact, it's one of the most infectious songs of ALL TIME).
But some people like Big Macs, and are only happy with Big Macs. I think AI will keep them happy. I think it can also help humans with true inspiration realize their ideas, but at the same time I can see it replacing grocery store music with ease. And even background jazz while dining. And more.