Auction houses throw a wrench into the works by allowing the market as a whole to weigh in. If an artwork gets sold at auction for significantly more or less than where it "should" be selling at, the whole system suffers. So the galleries representing the artists will bid on the artworks themselves and auctioneers will pretend that the piece was sold, to someone in the back, if it doesn't meet the reserve.
The takeaway I got from it was that value of art is not and can never be intrinsic, somebody is acting as a taste-maker, but that somebody never has complete control over the market.
If you think about it, it's not a whole lot different from other kinds of tradable good, (currency comes to mind) it's just that there are far fewer external indicators of value, making it more difficult to operate rationally. There's stability within the system, but the system is artificial and can seem rather arbitrary.
I don't mean that I have to directly identify with whatever they're creating but when someone is singing or playing a song, on some level you like to think about what they were feeling when they wrote or performed it. Likewise, something that always made me feel affinity for a band (especially as a teenager who was learning to play music) was the feeling that this was something I could do myself if I spent the time or developed the skills. I was drawn to a lot of punk bands and bedroom experimenters because I heard things that sounded like something I would do if I was more talented or experienced. It's hard to explain but it made things feel more personal. I think machine-generated music could certainly exist and be programmed to be indistinguishable from human-generated music but as soon as I knew it was created solely by algorithms, I'd lose much of the intangible (and irrational) connection I might have shared with the actual art.
On a purely musical level, there are already some artists that I really enjoy, who I suspect use algorithmic composition heavily in their music. I was wondering about this one the other day, since it's sort of similar to algorithmically mashed-up sample pieces that I've made (but better):