On composition, waveform generation, and the difference the industry refuses to name.
Did you know that Michael Jackson could not read or write musical notation? It's true.
He made Billie Jean by singing each part into a tape recorder: the bassline, the strings, the beat, voiced one at a time with his mouth. He handed the recordings to session players who turned his sounds into their instruments. The bassline that became one of the most recognizable four bars in recorded music left his body before it ever touched a bass.
Nobody on earth says the session players composed Billie Jean. No one questions the genius.
Hold on tight to that thought, because the entire debate about AI music is predicated on people forgetting it.
The Conflation #
The day generative audio arrived, the industry quietly collapsed two different acts into one word. The headlines read: "The AI composes music." False in every way that matters. The AI generates waveforms, the sound. Composition and waveform generation are not the same genesis. In fact, the historical record shows that they have never been a single function. A score does not make a sound on its own. Simply put, a score is a specification: a set of decisions about form, dynamics, harmony, orchestration, and intent, handed to realizers who transform decisions into air pressure.
The abstraction level of that specification has swung wildly across centuries of legitimate composition, and nobody revoked anyone's credit over it. Baroque figured bass told the keyboardist that harmony, not which voicing. The player invented the notes every night. A jazz chart specifies the changes and a head; the jazz artist improvises the realization on the spot. John Cage composed in prose. Film composers hand sketches to orchestrators who choose the literal notes on the staves, and the composer's name stays on the cue. Brian Eno designs generative systems and is credited as the composer of what the systems render.
And then there is the version nobody argues with: the bandleader who walks into the studio and riffs. Makes the sounds with his mouth. Describes the rest to the players. The Wrecking Crew followed that formula for a decade of hits. Likewise, The Funk Brothers did it at Motown. James Jamerson invented basslines inside someone else's song, and the industry's attribution never wavered: direction owns the composition. Realization within direction is performance, even when the realization invents notes.
That is sixty years of settled practice. The memory has been lost in one product cycle.
What I Actually Hand the Machine #
What follows is what exists in my writing, before any machine renders one second of an Ex Pulvis Records track.
The form: every section, named, in ruled order. For the current single, a launch, a ride, a breakaway, and a weightless drift to a docking, each with its own instructions. The dynamics: written as structure, including silence as a named section. Four seconds of engineered nothing, placed on purpose, because the distance between quiet and loud is a compositional weapon. The tempo, flat, by ruling. The key. The orchestration: which palette exists, what enters where, and what stops completely and never returns. The motif plan: that a hook exists, what instrument carries it, what character it has, where it is born, where it returns, and which return is the biggest. Every performed word, verbatim. The lyrics have never been anyone's but mine.
That document is not vapor and it is not ephemeral. It is riffed, written, composed, and hardened. It is a score in a different notation.
The machine transmutes it the way a session band changes a bandleader's riff into gold. Next, I do the other half of the job the discourse ignores: I listen to forty renditions and bank one. The science of the process is the mathematical certainty that my instructions and large sample size executions must result in the sound I heard in my head from the beginning. Photography settled a century ago that selecting among the captures the world offers you is authorship. Forty takes, one keeper, is not a slot machine. It is an audition of acts with the same spine but notably different postures.
One concession, made precisely, because honest labels must survive hostile review: the note-level choices, the exact pitch sequence of that hook, are the machine's, made inside my constraints and subject to my selection. That is true. It is also exactly what was true of Jamerson. The industry never confused the bassist for the composer. I decline to apply a stricter standard to my session band than the last sixty years applied to the greatest one that ever lived.
The U.S. Copyright Office, for what it's worth, is circling the same conclusion from the legal side: prompts alone don't confer authorship, but human selection, arrangement, and modification can. The words, the architecture, the selection, the finishing. That is where the authorship was all along.
The Real Difference #
Am I just the bandleader with a stranger band? Not quite. There is one difference between me and the man riffing at the session players, and it is not the role.
My session band can't hear me.
The bandleader's mouth-sounds carry the actual notes: the pitch contour, the rhythmic cell, the DUN-dun-DUNNN itself. My channel to the machine is prose. Prose carries structure, character, dynamics, and intent with precision. It drops the notes on the floor. And in that gap, the machine does not compose. It interpolates, filling the missing pitches with statistically probable guesses, the way a session player who couldn't hear the hum would play something in the right key with the right attitude and hope.
That is the honest technical description of this entire era of music tools: not artificial composers, but session bands behind soundproof glass, reading the chart, unable to hear the bandleader hum.
The cost was never authorship. The gate is control. My hooks arrive approximated instead of transcribed. Every producer working this way has felt it. The track that is right in character and wrong in melody, without the producer having the words to clarify why.
The Form Has Two Boxes #
Which is why the two proffered labels, "AI-generated" and "AI-assisted," are both wrong for this type of work. The labels were born from and built for a world where one box could hold a piece of music. Film figured this problem out generations ago. Many hands were in the work, one finished piece, solved not with "check box 1 or box 2" but with a credit sheet. Director. Writer. Cinematographer. Editor. Music had the same solution in practice: composer, arranger, session musician, and engineer. The industry seems to have conveniently forgotten that it owned one.
The true credit sheet for a track from my catalog reads as follows. Conception, direction, architecture, and lyrics: human. Note-level realization and performance: machine, within human constraints, human-selected. Engineering: human. As of this month, that line has receipts: a vocal defect diagnosed by ear, located by measurement, repaired by hand, verified by number.
The compact label that falls out is not present in the binary choices on the form. It is human-composed, machine-realized. An old division of labor in recorded music, only now wearing new equipment.
The Glass Is Coming Down #
The soundproof glass is temporary. There is a microphone in transit to this studio right now, and the day it lands in my hands, the oldest workflow in pop reopens: hum the riff, record it, and hand the machine the actual notes instead of a description of them. The channel repairs. The approximation ends. And further out, we are building the version where the glass never existed at all: an engine that realizes the specification deterministically, note for note, no interpolation, no guessing.
The instruments were always going to change. Every instrument was artificial once.
The question was never whether a machine makes the sound. The question is who made the decisions the sound obeys.
Built from nothing. Made to last.
Ex Pulvis Records. From dust.