**
The Politics of AI Music
Generative AI Music is not a tool that comes from the creative ether
I recently went to an event introducing “The Call”, an installation by artists Matt Dryhurst and Holly Herndon at the Serpentine Gallery in London. Herndon and Dryhurst trained an AI model using performances from 15 choirs, and used it to produce a number of works - physical, sonic and otherwise. The artists encouraged the audience to reframe how and where art happens. This is summarized well by a mantra that Dryhurst often repeats: “Don’t confuse the media for the art.” In this view, art is not outputs created by the model, but the surrounding processes by which art is created. Dryhurst and Herndon argue that AI is a new mode of art creation, not a replacement for it.
This installation comes concurrent with the release of their new book All Media Is Training Data. In this book, Dryhurst and Herndon again frame humans at the center of creation, leveraging new AI tools to make novel pieces. These are both political stances - in a world that is removing humans from art production, Dryhurst and Herndon are foregrounding humans in art creation. While we can try to reclaim the humanity in AI art, it is difficult to do so without reckoning with the adverse effects these technologies have on people, and the power dynamics which they introduce. AI may be another instrument for creating art, but it is clearly not like previous technological advances.
Generative AI Models Have Politics
Langdon Winner may call generative AI a political technology. In Winner’s view generative AI models might have inherent politics because of the process by which they are trained, requiring the unauthorized use (theft) of work. Alternatively, they might be political for their capacity to replace human artists, like workers replaced by pneumatic pumps in 19th century Chicago. Given these dynamics, does an AI-generated franken-song replace or abuse human artists? I think the answer is both, but maybe not for reasons we’d expect.
Replacing Artists By Hijacking Our Creative Imaginary: Is This New?
While I’m not sure we replace humans by using AI to create, using a generative model does save the creator from needing to listen to other artists to derive inspiration from them - a different type of replacement altogether. It doesn’t take much to see a world in which AI slowly replaces artists in our larger creative social imaginary, inspiring itself over and over again. While we used to consume art to derive inspiration for new creation, the inspiration is now built into the model. No consumption (or humans) required. Where this might lead us as listeners is murky, but where that leaves artists is crystal clear: out of a (paying) job and any future royalties. Models are certainly replacing humans, and I contend this is a bad thing, but is this a totally new phenomenon?
We could point to the invention of player-pianos and how this eliminated the jobs of many musicians who relied on live performance for income - I think this is an excellent comparison. Our society became less musical once we didn’t have to know how to play songs from sheet music, and the same happened when pianos started playing themselves. With each jump, human artists were devalued economically, but somebody still had to write the music. On this jump, we are further mechanising our consumption of music, and potentially entirely eliminating humanity from it.
Dataset Is Law: Stealing Inherent in Inspiration
As I listen to AI generated music, like that at “The Call” or generated by AI models like Suno, it’s obviously built on the work of many other artists. You can clearly hear the melody and intonations that are ever-present in pop music like a faint ghost of Billie Eilish. Sure, the output appropriates work of other artists by virtue of the model, but does the song replace humans as artists? In this sense I agree with Dryhurst and Herndon - in using AI tools we’re not being replaced as a creator, just creating new outputs. Unfortunately, how we derive these outputs matters.
The catch with AI generation is that our creative process and impact is totally shifted from appropriation to theft. Models are using Billie Eilish, Bob Dylan, Cher, The Baha Men, Mozart, and every other artist that might be available on the internet to train their models. I’ll set aside the obvious copyright implications for now, to focus on the impact on the creative process. In using Suno, there is now another creative cook in the kitchen: the software engineers who develop Suno. There are political choices embedded in their product that are now changing how (and what) we create - and we have little control over what and how they design their product, and how they license their training sets historically and in the future.
It is easy to point out that almost any instrument you can think of was a technology at some point. The piano was human-designed and human choices impact how we all make music with it. What’s different here is that a piano is built out of steel, wood and finely calibrated strings, while an AI model is built using songs that were created by a human and posted on the internet. I cannot control what Suno includes in their dataset like I can control what’s included in my music catalogue - and that’s a political problem that we must account for.
In Defence of Human Processes: Art is Still Human
As an artist, I wholeheartedly believe in the kind of hand-wavey value that many folks (Dryhurst and Herndon included) place on music being created by humans - the beauty of imperfection, the humanness of a pocket, the “feel” of a live room. I think these things are certainly lost when using AI to make music, but I’m not sure that is most concerning to me.
The scarier truth of the present moment is that human artists will once again lose even more of the little power they have. There will be many artists pushed out of their craft because of this technology. More painfully, the artifacts of their many hours of work are being stolen and used to train their replacements without compensation (Lauren Berlant’s Cruel Optimism comes to mind). This is a power dynamic which can solidify big tech companies' control of our culture and imaginations - things we value as inherently human. I’m not sure any amount of reclamation of humanity within AI music can overcome this damage, and I don’t think normalizing our media as training data will either.
**