This one isn't so much about the future, as what is happening now.
There's an AI music generator called Suno. (There's also one called Udio, but I haven't been able to generate anything tolerable with that, so I'm really talking about Suno.) It generates half-decent songs: some are bad, some are on par with uninspired stuff you hear on the radio.
What do you hear on the radio? Anything recent at all? I don't. I had to look up what the top 100 songs were. They were country and rap, neither of which I listen to. Some rap-country crossovers. Um. So what Suno is generating for me is, to my tastes, quite a bit better than what I'd hear on the radio. But even if you confine songs to ones somewhat to my taste, Suno does half-decent. Not inspired, but definitely on par with the low end.
I've got an album of reasonable stuff I've generated on Suno. It's actually a half-album, so far but I'll fill it out to a full 60-minute album as I find more successes.
Suno is a good, extremely well-versed, but very idiosyncratic musician. Since I have a free account, Suno owns the copyrights to the music not me. That's OK with me: the music truly is generated, with minimal direction from me. It's very good at doing variations on its melody. It's not good at understanding a melody I give it. The overall musical structure isn't good: if you skip from one bit of a song to another it's often the same, not louder or faster or in a different key or playing with a different theme or anything. Humans know to have a fractal structure to songs but Suno doesn't yet. Some of the harmonica music it generated, I tried playing it, and at least at first it was physically possible on a real harmonica.
Its generated lyrics are cheesy. It keeps rhythm and rhyme while being semantically fluff. I often edit or completely replace the lyrics. I've been writing stories, so sometimes I paraphrase the stories as lyrics, or even paste the stories in as lyrics verbatim (it's rather humorous what it does with that). I tried generating songs based on other people's prompts on Reddit rWritingPrompts, but Reddit disallowed AI anything in that subreddit and deleted my attempts to link to songs based on their prompts. Song lyrics need to be much shorter and simpler than even short stories, but still they should have more semantic content than a single sentence. They should introduce a few ideas and memorable phrases and anecdotes. Suno's got a big stock of cliche phrases and idioms. There have been several times where I liked Suno's chorus or some snippets of verses more than what I could come up with. But Suno's not going to fill out the plot for you, you have to do that yourself.
You can "extend" existing songs, and those extensions diverge in random directions. Especially instrumental ones. Some of the songs I've liked best have started at one place, then through extensions have meandered intototally different styles. That's not a normal human structure, but hey, you have to encourage an artist's strengths.
Is AI music going to push out human music? I'm surprised there isn't already AI music in the top 100. Actually, I'm not sure there isn't. I wasn't able to stomache listening to the top 100, I only got through the top 3. I could see generating a song, deciding you like the general structure, then workshopping the lyrics and instrumentals and doing live human performances of it. I suspect this is already being done. And as long as I don't listen to popular music anyhow, it doesn't really matter if it's AI-generated or not. AI generated music is certainly going to be a tool for low-budget movies and video games.
I have always written music myself, by hand. My best effort is already quite short of Suno's (I gave it only the prompt, acoustic melodic guitar ballad, and that it should be a song about wanting to get back to Lori Lee, but I can't right now). Collaborations are currently better than either alone, but standalone, in my case, AI has already won.