How AI changed its approach to musicIt remembers, some time relatively recently, it was in a friend's car, traveling home from something or other. One track on their car playlist, a jumpstyle track with horny furry lyrics, stood out. Horny furry jumpstyle? Fuck yeah! When it got home, it looked up the track and found out that it was, in fact, entirely AI generated.
It felt a little disgusted that it could ever have liked it, but with the effect of novelty, a car stereo, a playlist of otherwise non-AI music and the law of truly large numbers, it makes sense that one AI track would have caught it's (and its friend's) attention. It now had first hand evidence that AI could produce convincing enough musical output it was interested enough to actually look up a track. The creator responsible for the track was prolific - releasing an entire, polished-sounding album every few weeks, densely enough to border on spam - and it made it consider the fact it could easily improve its musical output by orders of magnitude by embracing AI.
So it stopped.
Took some time to reflect.
And ran in the opposite direction.
What generative AI does to music
So this post isn't really about generative AI tech, but you probably gathered that. It has nothing new to say on the subject that hasn't already been repeated ad nauseam in tech skeptic spaces. But the story from the intro really did cause it to reconsider how it approached making music.
Closer listening to the AI piece revealed a whole bunch of issues, but the overall impression of it was that it was a polished product, well produced (albeit with some obvious artifacts of the model, especially in the vocals), but also a sense that the piece lacked focus, lacked drive, lacked the kind of energy that would have come out of a musician, sincerely, putting in the effort to compose and produce a jumpstyle track about yiff. It felt... flat. Lifeless.
It's not the only one to pick this up. In his video on AI in music, Adam Neely goes as far as to say that he doesn't consider AI generated works to even be music, because they lack any connection to the history, creative process, and act of making music. The technology is, in his words, "anti-musical".
As somebody who has been making music in one form or another for most of its life, it deeply feels this point. It knows the amount of joy that would come out of making a track like that AI one. If it sat down to write some horny furry jumpstyle, it would have an absolute blast. The result might be janky, it might not flow super well, it might not have the surface-level polish, but it would be fun, it would be joyous, it would be real.
If you will permit it to be poetic, the music would have actually lived, in its bedroom studio. A recording is, as the name would suggest, a record of the music that existed in the room. The music itself, in its view, does not exist on the recording, it existed in the room, as a creative act, a process. Music happens in the room, a recording is just... a recording of it.
What generative AI has done to music is make polished, "finished" works, simulacra of recordings of music, into something that can be made without any music ever actually occurring. The music never lived. So its new approach to making music is to let the music live.
Letting the music live
This one has been involved, at some point or another, with all kinds of music. As a kid it played brass in a wind band, performing classical and film music. During its teenage years it picked up production, and made truly awful dance tracks with LMMS on a laptop. As an adult it has explored more electronic music and livecoding, it plays bass guitar, and is learning the Linnstrument. Of this entire "career", over a decade of music-making, it has only ever released one EP, Experiments 1, a collection of live electronic jams.
Others, especially in the electronic space, may consider a career that only resulted in one EP, despite making music for over a decade, to be a failure. This one used to think that way too. Berating itself for being somebody who "doesn't actually make music", who is "just playing around".
But what AI recordings have made it realize it is that the process of music is the point. Its musical life is one that has bought it immense joy and fulfillment, and bought joy to others too. Everybody from the audiences of junior wind band performances, the friends it made shitty rock songs with in college, those who have listened to its EP or any musical sketches it shared in group chats. Anybody who has tuned into process streams, anybody who has ever enjoyed spending time with it jamming even if it lead nowhere in the end. It's a life of music, the music being just as alive as it is. Who cares that only a small percentage of that music was recorded and made it onto a formal release. It doesn't matter.
When it first released Experiments 1, the imperfections bugged it. It only released the pieces in that state because it had told itself that it would release some kind of music that month, and didn't have time to polish them in the production stage. The title of the EP was picked as a way to try and placate its perfectionism, making it explicit that these were unfinished works.
In the months since, Experiments 1 has grown on it, specifically for those same reasons it wasn't super happy when it released it in the first place. It's a record of living music. Music that came from ideas it had, that is never finished, that was changing even as the recordings were made. It's a performance of music that had space to live.
In a world where AI models can produce music that sounds "polished" and "finished", the distinctly real quality that an EP of jams has is something it values more than if it were to have polished up the tracks and given them a production pass or two. The recording is not the music. The music exists in the room, it lives, it has space to breathe, space to change. The recording is just one performance of music that may never be performed again, or may continue to evolve and change in the future. Real music is never truly finished. It's a process, an act, a living thing.
AI has changed its perspective on making music for the better, by providing it a perfect example of what it doesn't want its music to be. Well produced, but dead. Let your recordings be messy and your ideas be unfinished. Let your music change over time as you change as an artist. Let the music live.