Bandcamp has published a policy called “Keeping Bandcamp Human,” and I want to argue with its title.
I should clarify where I stand, because I am not a neutral party. I help my fellow musicians record and arrange, and some of those songs get released. I have a stake in what platforms do with the music I touch. What I believe about that music is precisely what this policy cannot accommodate: a song, once made, can be worth as much as its creator, sometimes more. That is why the work matters.
The policy asks me to believe something else. It asks me to sort every track into one of two categories, human-created or AI-generated, and to accept that the first is where the value is and the second is where it is not. My objection is not that Bandcamp will sort badly. It is to the sorting itself.
Bandcamp states:
“Music and audio that is generated wholly or in substantial part by AI is not permitted on Bandcamp.”
What is “substantial part”?
Where in the chain from idea to master does the disqualifying substance enter? The quantized drums, the pitch-corrected vocal, the amp modeled in software, the reverb with no acoustic original, the sample from a library someone else recorded, the algorithmically generated rhythm, the “AI-based” instrument that is really a synthesizer, the patch generated by a model the composer trained: is one of these somehow more human than another? How many decisions must be delegated before the result crosses the line?
These are ordinary questions about how recorded music is made.
Music has been inseparable from tools for a long time. They capture performances, correct them, repeat them, transform them, synthesize sounds that never existed in nature, and allow one person to construct performances that no group of people ever played together. Software was already shaping how music is made long before the advent of “AI.”
This does not mean that every use of AI is identical to every older technology. It is not. Pressing a button that produces a finished song is different from adjusting a compressor. Training a model on millions of recordings raises questions that an analog delay pedal does not. A tool can affect authorship, labour, copyright, attribution and compensation in ways another tool does not.
But none of that gives us the boundary Bandcamp’s policy requires.
“AI” names techniques, not components detectable in a song. “Substantial” names a quantity without telling us what is being measured. If a model creates a chord progression but a musician performs every instrument, is that substantial? If every note is written by hand but the singer’s voice is transformed by a model, is that substantial? If a model produces twenty ideas and the musician rejects nineteen, rewrites the twentieth and records it from scratch, where is the AI in the creation?
The line does not exist in the music itself. It has to be imposed, and imposition is not discovery.
Bandcamp also states:
“If you encounter music or audio that appears to be made entirely or with heavy reliance on generative AI, please use our reporting tools to flag the content for review by our team. We reserve the right to remove any music on suspicion of being AI-generated.”
There are three different standards in those two sentences: “appears,” “heavy reliance,” and “suspicion.” None is defined.
Bandcamp has every right to operate its platform this way. My objection is not that it lacks the authority to do so, but that the standards it has chosen cannot be precisely located in the music itself. Once removal can follow from suspicion, the practical burden shifts. A musician whose work happens to sound “AI-generated” may find that making the music is no longer enough; they may also have to establish an acceptable history of how it was made. The work is no longer judged only as a work. Its legitimacy depends on its provenance.
My difficulty with this distinction is not only theoretical.
Several years ago, I watched a major television drama. I enjoyed the show itself, but I remember being repeatedly distracted by the music. Some of it felt generic and strangely careless. What bothered me was not that every cue failed, or that I could have written something better. It was the sense that, somewhere in the production that had taken so much care with its actors, sets and images, the music had occasionally been allowed to become merely sufficient.
At the time, my reaction was simple. I thought the music was poor, wished someone had done a better job, and moved on.
Years later, systems such as Suno have made that memory harder for me to dismiss.
They can now produce, in minutes and at negligible cost, music polished enough to compete with much of the functional music already being made for television, advertising, games and online media. I do not mean that they can reliably make great music. I mean something more uncomfortable: the economic justification for paying a human being to produce mediocre music is becoming difficult to defend.
This will cost musicians work. I see little reason to pretend otherwise, and even less reason to believe that musicians have the power to prevent it. Generative AI will continue to enter music whether musicians approve of it or not. Professions have disappeared before when technology made parts of their labour cheaper, faster or unnecessary, and I see no reason to assume that musicians are exempt from the same history.
That inevitability is not, by itself, an argument that the change is good. Something can be unavoidable and still destroy things worth preserving. But preserving a profession cannot, by itself, be the principle by which we decide what music deserves to exist.
Protecting musicians and protecting music are not identical projects. If a human musician produces indifferent work and an AI system produces something better suited to the context, what exactly are we being asked to protect?
I am not entitled to demand that the world preserve my role as a musician. Nor do I think any profession is entitled to survive simply because people have built their lives around it. What I can hope is that we preserve fair compensation and consent, along with the conditions in which beautiful music can still be made.
That seems to me a more reasonable thing to protect than human provenance itself.
If “keeping Bandcamp human” means protecting musicians from exploitation, taking consent and compensation seriously, and sustaining the conditions in which people can continue to make remarkable music, I have little quarrel with it.
But if it means deciding that a work deserves to remain because of what made it rather than what it became, then I am less convinced.
I do not know what makes music human. I am not sure anyone does. More importantly, I am no longer sure it is the right question.
A song does not owe us a biography. It does not become better because someone suffered over it, spent years learning how to make it, or depended on it for a living. Nor does it become worse because some part of its existence was delegated to something non-human. Those things may matter when we talk about labour, consent, authorship or compensation. They do not tell us what the song is worth once it reaches our ears.
Perhaps music only needs to be worth listening to.
If we cannot tell the difference without first asking who made it, I am not sure the problem is with the music.