The Coming Revolution Against AI Music

The greater the proliferation of AI music, the greater the revolt against it

AI music is the best thing to happen since prog rock.

Ted Gioia has been documenting and prognosticating on the current and future state of AI in the music industry for a while now. My ears perked up most recently when he cited a claim from an AI company that its application had created over 100 million new songs.

Having both a foot in technology and a foot in music, I’m intrigued by all of this. A few years ago, I had a conversation with a Google data scientist where we talked about ways to test AI capabilities and I suggested trying to get the AI to play jazz. I was laughed out of the room. Alas, fast-forward and the new Turing test isn’t whether you think you are talking to a human when in fact you are talking to a computer, it’s whether you think you are hearing and seeing human creativity when in fact you are getting output from a black box.

Will a computer ever play jazz? Well, to start with, we need to more closely define what we mean.

We could mean: “Will the computer be a jazz player? Will it be able to competently (at minimum) improvise in concert with live human musicians? Perhaps using idioms that ground that improvisation within the context of jazz?”

I’m actually not all that interested in those questions. I think that line of questioning is a bit solipsistic.

The more interesting, and valuable, question is: “Does a listener think they are listening to jazz?”

And that’s an interesting question, not just from the clinical Turing-esque point of view of distinguishing if you are engaging with an AI. It’s a question that has to deal with education. And culture. And perception. And value.

It’s a question that fundamentally starts at the level of: “Do you (human) know what jazz is?”

And “How do you define it?”

And quickly goes down a rabbit hole into the fundamental question of “What is music?”

I should note that when I’m talking about AI, I’m talking here about the currently in-vogue generative AI that apparently we all love. I’m not talking about whatever past, current, or future AI capabilities may be used to help musicians design new synthetic instruments or map out complex arrangements or act as experimental interlocutors in live performance. The range of AI use cases is far greater than what gets reported above the fold, so for the sake of simplicity, this diatribe is wholly focused on generative AI.

That said, I could care less about what we are currently calling AI. Underneath the perceptive noses of most people, the underlying technology will change and evolve before they’ve had a chance to understand what the current technology is. But I’m very invested in what the perception of what AI is means for music and music listeners.

And I’ve come to two very rudimentary conclusions.

First: Once technology goes mainstream, it becomes frozen in time. Windows 95 “looks” like the late 90s. Geocities, MySpace, LiveJournal all “look” like their era. And so it will go for Twitter and the social media boom of the teens. All technologies end up looking like bellbottoms and big brown ties. The current generation of AI is no different. While the underlying capabilities — and more importantly, the implications — of generative AI will evolve and continue to have a behind-the-scenes influence on future capabilities, the current application of typing into a box and being amazed by the return of a semi-plausible college entrance essay or lines of half-correct code will go the way of Wacky Wall Walkers. Conclusion: none of us know how the current generation of AI will influence later generations of AI, but it is safe to say that the majority of “music” created by AIs of the current generation will sound woefully dated before long.

And it has nothing to do with the genre in which the AI may be producing the music. It has to do with the fact that the AI is necessarily trained to mimic things that already exist. When humans do this, they quickly end up producing music that feels cramped, unoriginal, and out-of-date. The current state of AI doesn’t deal well with creativity in the sense of values and aesthetics that evolve over time — especially as relate not to something as (relatively) quantitatively measurable such as musical formality, but rather to human experience and the influence of that experience on music making.

Second: AI may end up being the best thing to happen to music since half-hour long electric bass solos and the introduction of exploratory synthesizers in rock and roll. Those solos and explorations provided a big fat target at which musical revolutionaries could aim their riffs and fury. So, in the same way that punk and NWOBHM rose up to overtake the bloat of progressive rock and arena shows, so too could a new generation of (human) music/musicians rebel against and destroy AI music. I’m a human musician, so in my estimation I think this would be a good thing. (And yes, I realize that punk and NWOBHM gave rise to a new generation of arena pandering, but let’s take on one thing at a time here).

While I fully appreciate the fact that it seems like crap music always rises to the top, I dare say that a general revolt against AI music would help bring a lot of genius underground musicians to the attention of a wider audience… and this could have lasting effects.

Because AI has a key weakness when it comes to music that I haven’t mentioned above. Here’s the deal: developers of AI are driven to improve their models. The intent is to produce “better” results. Ultimately, AI text generation (for example) will meet a level of perfection (that’s someone’s ideal). So too, foreseeably, with AI music. Perfect sound forever, Right?

But that’s the fatal flaw. Because great music — the kind of music that moves people, compels people, and dare I say changes people’s lives — isn’t perfect music. It’s human and humane music. When we fall in love with music, we are also falling in love with the people who create it and our peers in their audience who adore it.

No one would want to listen to music created by a perfect person. It would be intolerable. And no one would listen for long to music that did not have an evolving human story behind it. It would bore us to tears.

Without The Beatles, the music of The Beatles is often meh. Without John Coltrane, the music of John Coltrane does not exist. Without the story of who Daniel Johnston was, the songs are nursery rhymes. But with the story, they were an ongoing portrayal of tragedy and triumph. My opinion. Your milage may vary. But this is the way I see it.

And in the same way that The Damned and The Clash stomped on the keyboard operas of progressive rock’s worst tendencies, a new generation (and, I’m sure many members of the current crop of musicians both legendary and unknown) will stomp on AI music. And the introduction — even the proliferation of AI music — will be the catalyst for wide change in the way that music is appreciated. At least by those who can still recognize the value in humanity.