A literary story does not need to change for readers to begin reading it differently. In 2026, Jamir Nazir’s “The Serpent in the Grove” won the Commonwealth Short Story Prize. Controversy followed when speculation arose that one or more of the shortlisted stories may have been partially generated by artificial intelligence. The authors strongly rejected the allegations, and no such claim should be treated as established. Yet the uncertainty alone produced material consequences. Granta kept the stories on its website but announced that it would no longer enter external publishing partnerships without editorial control. The work itself remained the same. What changed was the story attached to their creation. Audiences now had to ask themselves; if a machine contributes to the writing, who becomes the author?
Generative AI can participate meaningfully in literary production and produce texts that readers consume and enjoy. However, present language models should not be recognised as autonomous authors.
When the product passes for poetry
The first temptation in debates about AI is to begin and end with the product. Brian Porter and Edouard Machery found that non-expert readers could not reliably distinguish poems generated by ChatGPT from poems written by established poets. Participants sometimes rated the generated poems more highly because their meanings and emotional qualities appeared more immediately accessible.
Roland Barthes makes dismissing this experience difficult. In “The Death of the Author,” he rejects the belief that a text’s meaning originates entirely in the author’s intentions. Meaning emerges through reading rather than arriving as a message delivered intact from an authorial mind. An AI-generated poem can mean something even if the system does not understand its own words. Therefore, meaning and authorship are not interchangeable. A sentence produced by accident can acquire meaning when placed in a poem, but the accident does not receive the byline. Evidence that readers can value AI-generated language establishes the literary potential of the product. It does not settle who is responsible for it.
Creativity was never a solitary act
Rejecting autonomous AI authorship does not require reducing AI to a passive instrument. It proposes language, redirects ideas and produces possibilities that its user may not have anticipated. Mark Coeckelbergh therefore questions the assumption that creativity must belong exclusively to either the human or the machine. Creative participation does not automatically produce equal authorship. Editors, translators and performers can transform a work without necessarily being named as authors. The more useful question is not whether the machine contributed; but what kind of contribution it made and who converted that contribution into a purposeful public work.
The author may be dead, but the author function remains
Michel Foucault’s “What Is an Author?” helps explain why authorship continues to matter even after Barthes separates meaning from intention; the author function organises discourse. An author’s name groups texts, shapes interpretation and determines how institutions value and circulate writing. Copyright, prizes, contracts and reputations all rely on this function. A language model cannot currently accept a publishing contract, approve a final manuscript or explain why one output was retained while hundreds of alternatives were discarded. More importantly, it cannot accept responsibility when a published text plagiarises, defames or harms. Aaron Hertzmann argues that art is fundamentally a social act through which agents communicate with one another. Computers can facilitate artistic production without becoming artists, just as we attribute a photograph not to the camera but to the person who uses it to make and communicate creative choices . In AI-assisted literature, the human writer prompts the model, assesses its suggestions and chooses what survives. Even when a model supplies an unexpected image or elegant sentence, the human gives that language its public context by selecting, revising and publishing it. Authorship belongs to the participant capable of making the work answerable to others.
Distributed creativity cannot mean distributed irresponsibility
Accountability becomes more important when generated language carries the biases of its training data. Bender and her co-authors warn that large collections of internet text amplify dominant perspectives. This problem operates below obvious stereotypes. In two minimally prompted ChatGPT stories generated by Wentworth, the male protagonist becomes a structural engineer associated with bridges, calculation and professional achievement, while the female protagonist inherits a garden, cries in the dirt and finds fulfilment through domestic inheritance, nature and car . Neither narrative openly announces a theory of gender, instead, gendered conventions organise occupation, setting, metaphor and emotional attention. Michele Elam calls the model’s disrupted relationship with history “algorithmic ahistoricity”: training data are removed from their original chronological and social settings, allowing past conventions to return as present defaults If creative agency is distributed across a human-machine system, responsibility must not disappear into that system. The model cannot investigate its defaults or decide that a culturally familiar pattern should be resisted.
The human author can.
Participation without personhood
The controversy surrounding “The Serpent in the Grove” demonstrates that readers and literary institutions have not stopped caring about authorship. A byline does more than explain where words originated. It identifies who made the relevant decisions and who stands behind the finished work. AI’s capacity to surprise its user and shape a text justifies describing AI-assisted writing as distributed or collaborative creation. It does not yet justify describing the model as an autonomous author. The distinction is ultimately one of responsibility rather than human exceptionalism. We do not need to deny the machine’s creative contribution to preserve authorship. We need a vocabulary that can recognise participation without surrendering responsibility.