From my viewpoint there is a common denominator to sound that makes it music, but music tastes don't last forever. What AI can't produce is novelty of expression. And also novelty in performance, like those small timing and volume variances created when someone is performing (jamming) in a group. But suppose AI has enough of a data set to even mimic that, people will seek out a new sound at some point. I could even imagine some weird sci fi techno dystopia where music is no longer recorded because AI can mimic it almost instantly so people flock to in person performances.

My point is that expressions of emotions have a certain bar of excellence. What make clair de lune so classic? I don't think you can quantify that. Nor can you fully recreate Jimi Hendrix's playing style to the ear of a guitarist who's been playing for years. Maybe you can fool a lot of people but it will always feel like a cheap imitation and music tastes will have moved on by then.

Off topic Side note: This is one of those rare internet posts that hits the bullseye for me. Everything from the discussion of the half-life game engine to making music in the early 2000s, to the thoughts on the music industry and LAN parties.

OP you and I are bredren. This is what the internet used to be like, finding out that your own experience is shared. All your thoughts and opinions mirror a lot of my own. I gave up the chase of being a guitar for hire a couple years ago. All the best man.