Daddy's Car credits an AI draft and a human arranger separately

What happened

In September 2016, Sony Computer Science Laboratories released 'Daddy's Car,' a pop song written in the style of the Beatles using a research system called Flow Machines. The lab's own press archive page states the division of labour in one sentence: 'the Flow Machines software generated the composition,' and 'once the Flow Machines software generated the composition, it was arranged and produced by Benoît Carré.' Flow Machines itself, according to the lab's own overview page, was 'a research project funded by the European Research Council (ERC) and coordinated by François Pachet,' built around modelling musical style rather than producing a finished, releasable track.

What the documents say

The overview page describes the project's ambition in terms of style: treating the recognisable manner of a composer, what makes a Beatles song sound like a Beatles song, as something a system could learn from a corpus of recordings and apply to new material. That is a materially narrower claim than 'the AI wrote a song,' and the lab's own credit line for 'Daddy's Car' reflects that narrowness directly: the system is credited with generating the composition, the underlying melodic and harmonic material, while a named human, Benoît Carré, is credited with arranging and producing it, turning it into a finished, performable recording with lyrics and vocal performance. The lab's press archive does not claim the system alone produced a releasable song, and the credit itself requires a human stage.

Why it matters for makers

The mechanism worth separating is composition-as-generation from composition-as-finished-work. A system trained to imitate a style can propose melodic and harmonic material consistent with that style; on these documents it does not handle arrangement, lyric writing, vocal performance or production decisions, all of which 'Daddy's Car' credits to a human. For a songwriter evaluating a generative tool today, the useful question is not whether a system can produce style-consistent material, several since have, but which stage of the songwriting and production process it replaces, and which stages require a credited human contributor, as this 2016 release did.

What to check before you use it

Before describing any generated output as 'AI-composed' in full, a maker can check, against whatever documentation the tool provides, which stage of composition, arrangement or production the generation covers, and credit the humans who did the remaining stages, following the precedent this 2016 release set. This is an editorial reading of what the lab's own materials establish, not a general rule for how every generative tool must be credited.

• Does my chosen tool generate raw compositional material, a finished arrangement, or something in between?
• Who is doing the arrangement, lyric writing and production work the tool does not do, and are they credited?
• If I describe a track as 'AI-composed,' does that claim match what the tool's own documentation says it does?

'Daddy's Car' is often cited as the first AI pop song; the lab's own record of it describes something more precise and more durable as a template, a generated draft finished by a credited human musician, with each stage documented separately rather than collapsed into a single unexplained claim.

Continue in Listening & research, or Now and Then used separation, not generation, to finish a demo.