A Drum Machine Never Wrote a Song

AI Music Composition vs. AI Production

Questions about what counts as "real" music aren't theoretical here. They're operational. We've made a deliberate choice to block AI-generated uploads and to remove anything we identify as AI-composed, whether that's the music or the lyrics. That policy forces a question a lot of platforms are still dodging: what, exactly, are we protecting?

The answer is narrower than people expect. AI has a legitimate place in music — in production. It does not have a place in composition or performance. Those are the parts that are actually art. A song can be produced by AI. It cannot, in any sense we recognize, be composed or performed by it.

That's a distinction, not a technicality, and it's worth being precise about why. Production is the process of shaping a sound that already has intent behind it — EQ, mixing, mastering, arrangement, sound design, the thousand small decisions that turn a performance into a finished recording. It's craft in service of someone else's expressive choice. Composition and performance are the expressive choice itself: what to say, how to say it, which note comes next and why, what a lyric means and what it costs the person singing it to mean it. One is technique. The other is the thing technique is for. AI tools are extraordinary at the first category. They cannot participate in the second, because there's no one home — no experience being expressed, no risk being taken, no self on the other side of the choice.

Here's the question that actually tests whether this distinction holds up: if AI replaced something like a drum machine, should we be upset, or is that just a change in technology?

It's a fair challenge, because the drum machine is exactly the precedent people reach for. Purists were upset when it showed up too. So were they upset about the synthesizer, the sampler, the Auto-Tune plug-in, the DAW that let one person do what used to take a full band and an engineer. Every one of those was, at launch, accused of replacing musicianship. Every one of those is now just an instrument sitting in the toolkit.

But look at what a drum machine actually does versus what a generative composition model does. A drum machine is silent until a person decides what pattern to program, what tempo serves the song, when to bring it in and when to drop it out, whether it should feel mechanical or human, whether it belongs at all. Roger Linn's machine didn't write a bassline for "Sexual Healing" — Marvin Gaye and his co-writers decided how it should feel and where it should sit. The drum machine executed a choice. It didn't make one. That's the whole category of tools we have no argument with: samplers, synths, pitch correction, loop pedals, DAWs, and yes, AI mastering and mixing assistance. They're production. They amplify a human decision; they don't originate one.

A generative model that composes a melody or writes a lyric on prompt is doing something categorically different. It isn't executing your compositional choice. It's making the compositional choice for you, then handing you the output. The "artist" in that transaction didn't decide what the song should say — they typed a request and accepted an answer. That's not a change in technology in the way the drum machine was. It's a change in who's doing the art. And once no one is doing the art, what's left is a product that resembles a song the way a photograph of a meal resembles dinner.

The rest of the industry hasn't settled on this answer, but it's clearly circling the same question. Warner Music settled its suit against Suno late last year in a deal that requires Suno to retire models trained on unlicensed catalogs and put real limits on free-tier commercial use; Universal and Warner cut similar terms with Udio, which now operates more like a licensed remix sandbox than an open generator. Sony refused to settle — it's still suing Udio directly, and discovery reportedly turned up tens of thousands of tracks used in training without a license. Underneath all of that, the musicians' union amended its own suit against Universal and Warner, arguing that even the labels' licensing deals with Suno and Udio cut session musicians out of compensation for recordings those deals were built on. And Spotify, having watched AI tracks get uploaded to dead or dormant artists' pages, had to ship a feature just so real artists could lock down their own profile from impersonation.

Read all of that together and it's not really a story about whether AI belongs in music. It's a story about an industry that skipped the composition-versus-production question and went straight to "can we monetize this," and is now doing expensive legal cleanup on a distinction it should have drawn on day one. The labels are negotiating licensing terms for training data while nobody has satisfactorily answered whether the output of that training deserves to be called an artist's work at all. That's the gap we're trying to close on this platform, by refusing to let AI-composed material stand in for the real thing in the first place, instead of building the policing feature after the fact.

None of this makes us anti-technology. We want producers using every tool available to make a record sound exactly like the vision in their head. We want independent artists who can't afford a mixing engineer to have access to software that gets them close. That's what production tools are for, and AI is a genuinely good one. What we don't want is a system that generates the song itself getting credit — commercial, cultural, or otherwise — for an act of composition and performance it never actually participated in, because it can't. It has no performance to give. It has no life the composition could be about.

A drum machine never wrote a song. It just kept the beat while someone else decided what the song was going to say. That's the line, and it's the one we're building Splinge around.