On Emulation

As I was going through my old article compilations from the early 2000s, I stumbled on an article about Tom Waits when he released his Blood Money album, and he had been working with the granddaughter of Leon Theremin (playing theremin). I love this line: “She was a true Russian, the musical equivalent of the good butcher who uses every part of the cow." He also said, "The best comment on a piece of music is usually made through another piece of music," "Songwriters don't go to school; they sit around and listen to other people's records. You stick your ear in the speaker, and try to figure out how somebody did something. It's like studying footprints or trails or ancient maps. Usually what you hear is the inaccurate attempt to emulate someone else, and then you try to emulate that…in the process, something else happens. Because what intrigues me is not the emulation, but the possibility of surprise."

In AI music there's no emulation, unless you make a concerted effort to reverse engineer it. AI certainly does butcher lyrics, or it's a combination of both you and the algorithms doing the butchering. There definitely has to be a lot of cutting to make AI music work with lyrics. One thing AI music won't do is create a song based on a chord or chord change; you can't specify to start a song on a C minor 6 chord (although that's a possibility in the future I would presume): You'd give it the option to be strictly diatonic or chromatic, and then it would extrapolate the chord changes based on that “seed chord”. Both ways of generating music is a process of discovery, whether it be starting with a chord, or 2 or 3 spins on a set of lyrics. The primary difference is that you aren't being generative based on the things that you did before. Over the course of time you are learning things, but learning in a different way than a machine. Once you add minor 6 chords to your harmonic vocabulary, you're not going to software to make more songs with minor 6 chords. That machines will be organically creative is probably a pipe dream. Once someone has been doing things manually for decades, AI music is merely an interesting toy. But I'll always want to use it. (It’s my only way to do country music).

Again, the problem with AI music is that the harmony is too simple. If someone were to cover these songs, or recreate them from scratch, they could expand the harmonic possibilities and would make it more interesting. AI music is also a form of portrait photography in a tangential way, as you are giving it a set of instructions (or coaching it) via lyrics so you get the results that you want.