The [YOU] on AI cycle describes collaboration between human and AI as a spectrum, from editorial refinement at one end to generative surprise at the other. Distributed authorship names what happens at the far end of that spectrum: the moments when the machine’s contribution is not a performance within the human’s score but an alteration of the score itself. These moments are not aberrations to be corrected; they are the constitutive feature of the most productive human-AI creative collaborations, and they demand a theory of authorship that the existing categories cannot supply.
The cycle’s claim that AI is the most powerful amplifier ever built acquires a specific meaning in this context. An amplifier that can only render what it is given amplifies without authoring. An amplifier that can alter the purposes of its operator through its outputs is doing something closer to co-authorship—and the question of whether the human maintains sufficient worldmaking authority to constitute the work as genuinely theirs, rather than genuinely shared, is the question distributed authorship forces into view.
The concept emerges from the intersection of Goodman’s philosophy of art with the practical experience of human-AI creative collaboration. Goodman’s score-performance model provided the clearest prior account of distributed creative labor: the composer specifies identity through notation, the performer fills what notation leaves open. But the model requires a formal boundary between specification and interpretation, a boundary that natural language—the medium of human-AI collaboration—cannot enforce. When the specification is a natural-language prompt and the interpretation is a natural-language response, and when each response can reshape the next specification, the boundary between score and performance dissolves into a continuous process of mutual revision.
The resulting category has no precise precedent. Ghost-writing involves one agent producing what another will claim; translation involves one agent rendering what another specified in a different medium; editorial collaboration involves one agent improving what another originated. None of these captures the specific feature of human-AI collaboration in which the machine’s contribution can be constitutive of the work’s worldmaking project rather than merely facilitative of it.
The scheme-content relation under collaboration. Goodman’s symbol system framework requires that a scheme-content relation be established: the worldmaker must intend the symbols to refer in particular ways, selecting this referential function over alternatives. In distributed authorship, the scheme-content relation is established incrementally, through the back-and-forth of human specification and machine response. Neither party establishes it alone; it emerges from the dynamic of their interaction.
Process as score. In distributed authorship, the work’s identity is determined not by a fixed text or notation but by the process of its construction—the specific trajectory of worldmaking decisions, human and machine interleaved, that produced it. The process is the score; the interaction is the notation. This means that two texts with identical word sequences could be different works if their authorial structures differ, and that the standard allographic criterion—the text determines the work—is insufficient for evaluation.
Revised purposes and worldmaking authority. The human in a distributed authorship context maintains worldmaking authority when they hold the purposes that determine the criteria of rightness by which the collaboration is evaluated—when they can accept, reject, and redirect the machine’s contributions against standards they, not the machine, set. When the machine’s contributions revise the human’s purposes, the revision is the human’s act even if the material that prompted it was the machine’s; and the resulting work is authored by revised purposes that emerged from the dynamic of collaboration.