Autographic Crisis in Literature

In The You On AI Field Guide

Goodman's allographic classification of literature was never entirely comfortable, even in the pre-AI age. The ontology of literary works has been debated for decades: Is the work the text, the sequence of speech-acts the text encodes, the interpreted meaning, or the total aesthetic experience? Goodman's answer—the work is the text—was motivated by nominalism (he wanted to avoid commitment to abstract meanings) and by the functional need for identity-conditions that could be specified with precision. But literary practice has always treated productive history as aesthetically relevant in ways that musical practice does not. Knowing that Kafka wrote The Trial and did not want it published affects how the work is evaluated; knowing that Shakespeare collaborated on Pericles affects its canonical status. The aesthetic relevance of authorship in literature has been greater than the allographic classification permits.

AI intensifies the discrepancy to the breaking point. When a text is produced through sustained human-AI collaboration—when the human provides intentions and evaluations, the AI generates passages, and the final text emerges from iterative refinement—the text is neither single-authored (the human did not write every sentence) nor transparently collaborative (the AI's contributions are not credited). The name on the cover suggests single authorship; the productive process was distributed. The misrepresentation is not deliberate deception (the author may genuinely not know which sentences originated where), but it is a misrepresentation nonetheless—readers receive a text believing it has a productive history it does not have, and their evaluation is shaped by the belief. Whether the misrepresentation matters aesthetically is the question that defines literature's autographic crisis.

The crisis is not resolved by disclosure. Stating 'this book was written with AI assistance' does not specify which aspects of the worldmaking were human and which were machine, does not indicate whether the assistance was editorial (refining what the human already intended) or generative (contributing ideas the human had not conceived), does not allow the reader to evaluate whether the work's achievement is attributable to the human's judgment or to the machine's pattern-matching. Full disclosure would require annotating each sentence with its productive history—and even that would not capture the iterative, interleaved character of the collaboration, where each party's contribution is a response to the other's, and the final version is constituted by the entire arc of interaction. Disclosure can acknowledge that collaboration occurred. It cannot make the collaboration's structure transparent in the way that the score-performance relation is transparent in music.

Origin

The concept is this volume's synthesis, building on Goodman's autographic/allographic framework, on Jerome McGann's sociology of texts (which treats published works as collaborative products of authors, editors, publishers, printers), and on the empirical observation that AI collaboration has made literary authorship structurally indeterminate. The 'crisis' terminology is borrowed from Thomas Kuhn—a crisis in the Kuhnian sense is a period when the dominant paradigm (here, the allographic classification) encounters anomalies it cannot accommodate (texts whose productive history is opaque yet aesthetically relevant), forcing either a paradigm revision or the recognition that the field contains multiple incommensurable approaches.

Key Ideas

Allographic classification assumed transparency. Literature's identity-by-text presupposed that productive history was either single-authored or explicitly collaborative—AI produces a third category of opaque collaboration.

Readers expect authorial disclosure. The name on the cover signals that the named person wrote the sentences—AI collaboration violates the expectation without visible evidence of the violation in the published text.

Origin affects evaluation. What a text achieves as evidence of sensibility, craft, or contribution depends on productive history that the allographic framework treats as irrelevant—the framework's inadequacy is now visible.

Disclosure does not resolve. Acknowledging AI assistance does not specify which aspects of worldmaking were human—full transparency would require annotating the iterative process, which is structurally impossible to represent.