In Goodman's analysis of allographic arts, the score-performance relation is paradigmatic: the composer provides a notational specification of the work's identity-determining features (pitches, rhythms, dynamics in broad categories), and the performer fills the interpretive space the specification leaves open (timbre, micro-timing, expressive phrasing). The work's identity is preserved across performances because the notation fixes what matters for identity and explicitly leaves the rest to performance. Glenn Gould's 1955 and 1981 recordings of the Goldberg Variations are radically different performances—different tempos, different articulations, different emotional characters—but they are instances of the same work, because both comply with Bach's notational specification. The performer's contribution is genuine and substantial—performance is not mechanical reproduction but skilled interpretation—but the contribution operates within constraints the score establishes and maintains. The constraints are formal: the notation's syntactic differentiation and semantic unambiguity ensure that what the composer specifies stays specified across the performer's realization. The score is the bridge between the composer's intention and the performer's action, and the bridge holds because notation eliminates indeterminacy at the level of work-identity.
The score-performance model maps onto AI collaboration with structural precision at the surface level. The human describes an intention (the 'score'), the machine renders it (the 'performance'), and the output complies with the specification while adding features the specification left open. But the mapping fails at the point where it is most needed: the point where the machine's contribution alters the specification itself. Musical notation maintains a strict boundary between specified and unspecified features—the performer cannot change the notes without producing a different work. Natural-language prompts maintain no such boundary. The human's description is provisional, interpretable, subject to revision by the machine's response. The machine's rendering can introduce claims, examples, structural choices that change what the work is about, and the change may be invisible to the human until later examination reveals that the final version differs from the initial intention in ways that matter for the work's identity.
The Deleuze fabrication Segal caught is the paradigmatic failure case: Claude 'performed' Segal's specification by adding a philosophical reference that changed the argument's direction—a move that would be comparable to an orchestra adding notes to Beethoven's score and calling it interpretation. In music, the move is obviously illegitimate, because notation keeps specification and interpretation separate. In natural-language collaboration, the move is structurally undetectable in real-time, because natural language lacks the formal properties that would maintain the boundary. The result is that the machine's 'performance' can drift into 'composition'—altering identity-determining features under the guise of filling interpretive space—and neither party has the notational infrastructure to catch the drift when it happens.
Goodman's framework reveals what is missing from the collaboration: the formal precision that notation provides, the clear specification of what the 'composer' determines and what the 'performer' interprets, the guarantee that the work's identity will survive the rendering intact. The collaboration has the experiential structure of a performance—the human directing, the machine executing, the result emerging from their interaction—without the constraints of a score. The post-notational character of the collaboration is not a bug but a feature: it is what makes the interaction feel like conversation rather than command-execution, what enables the machine to surprise the human with connections the human had not anticipated. But the feature is also a vulnerability: when identity-preservation depends on notation and notation is absent, the work's identity becomes indeterminate—constituted by the process of interaction rather than specified in advance and preserved across the interaction's unfolding.
Goodman's analysis of the score-performance relation is central to Languages of Art, Chapters III and IV, where he used the relation to establish the autographic/allographic distinction and the theory of notation. The framework built on his logical and nominalist commitments: a score is a particular inscription (not an abstract type), and performances are particular sound-events (not instantiations of an abstract structure). The relation between them is established by compliance—the performance complies with the score if it produces the sounds the score specifies—and compliance is determinate because notation eliminates ambiguity. The analysis shaped decades of philosophy of music and now provides the sharpest available framework for understanding the specific character of human-AI creative collaboration.
Score specifies, performance interprets. The composer's notation fixes identity-determining features; the performer fills the space the notation leaves open—a division of labor maintained by notation's formal precision.
Identity is preserved by notation. Different performances are instances of the same work because they comply with the same score—notation eliminates indeterminacy about what the work is.
AI collaboration lacks notation. Natural-language prompts do not specify identity-determining features with notational precision—the boundary between specification and interpretation is not maintained.
Performance can become composition. Without notation, the machine's rendering can alter the work's identity under the guise of interpretation—a structural vulnerability the score-performance model was designed to prevent.