Aftermarket Music

 

Music does require a kind of "fantasy", at least initially, for the motivation to play it. That disappears when the instrument doesn't feel right, and the effort you've put in is uninspiring. People dream of playing the piano, but you might only have a small out-of-tune spinet. Your dream was that it was a grand.

What we seem to value collectively is essentially a romantic one supported by a story, like making an album like Joni Mitchell's Blue or Springsteen's Nebraska, with just a notebook, a guitar, and a 4-track, alone in a room on the heels of some traumatic event. This is probably the optimal spiritual experience for a writer, but I've found I can get a similar feel or mood with AI, though it's anathema to the romantic approach, even if we can tolerate the rough-hewn nature of a solo guitar-and-voice performance. We want "masterpieces" we think must have been created by a tortured artist.

With AI music, the idea-to-product process is faster. It's rapid prototyping in which the prototype can stand as the finished product. Attempting to take it back in your hands and play it yourself is back to manual approaches. But your hands are ultimately more interesting. For example, I can't get a good swing or 12/8 feel in AI, or mixed-meter, or a polyrhythm if I want it, and I can't shape the harmony as I would on a guitar or piano, and certainly not use alternate tunings. The vocal phrasing and syllabic stresses can often be clunky.

AI is good at recording, mixing, and mastering, things that I sometimes find tedious because of the repetitive nature of the process. With AI, it snaps together easily, and I can move on to the next lyric idea without mixing or mastering.

What is sorely missing with AI is a production style: So far, I haven't liked working with any of the "producers". They aren't a Rick Rubin or a Brian Eno. The musicians are great, especially the singers, who sing whatever I want. But they're still unsung. When they can finally be identified (or when it's easier to make your own training set), I'll stop using them and use my own voice for the vocals. But it will still be "aftermarket", with the stems all your own, but it won't be a Blue or a Nebraska.

Music as a genie has never been in a bottle, so it's futile to think recorded music will ever be. But AI music can't be anything other than "aftermarket" art. It's not that it's bad; it just can't be genuine. What is genuine for me are the lyrics and how language shapes the result. They are LLMs after all.