Can an AI Detector Decide Who Wrote a Book? The Goncourt Controversy Raises a Bigger Question

5 hours ago

4 min read

A literary prize can change a writer's career. This week, artificial intelligence may have changed one for a very different reason.

On 25 September, the Académie Goncourt removed Haitian-Canadian novelist Thélyson Orélien's debut C'était ça ou mourir from its longlist after allegations that the novel had been largely produced using generative AI. Orélien denies using AI to write the book. The academy said its decision was intended to protect the integrity of a prize whose role is to honour literature written by human beings. The crucial complication is that the AI allegation has not been conclusively established: different detection tools have produced different results.

AFP's reporting on the controversy notes that the dispute now sits alongside separate plagiarism allegations concerning some of Orélien's earlier work. Those are distinct issues and should not be treated as proof of the AI claim about his novel.

The accusation is becoming almost as powerful as the evidence

This is where the story becomes important for every working author. We have spent the last few years asking writers to be transparent about AI. That is reasonable. If a machine has generated substantial passages of a novel, readers, publishers and prize organisers are entitled to care about it.

But the other side of that bargain is becoming harder to ignore. If somebody accuses a writer of using AI, how does that writer prove the negative?

Recent research in AI & Ethics makes an important distinction. AI-text detectors observe statistical properties of a finished text; they do not directly observe provenance. In a documented multi-tool audit, the same human-authored manuscript received materially different classifications from different commercial detectors. The paper argues that detector output should not, by itself, be treated as proof of authorship or misconduct.

That does not mean detection tools are useless. Some are becoming considerably better. It means the consequences attached to a detector score need to match what the technology can actually establish.

AI-assisted is not the same as AI-authored

I think writers need to be much clearer here, because treating every use of AI as equivalent helps nobody.

I write my books. The characters, worlds, dialogue, story decisions and final prose are mine. I can still use AI around that work: to research a subject, visualise a location or character, organise material, explore marketing ideas, produce supporting imagery or help run the creative business surrounding the books. Those uses can raise their own questions about accuracy, copyright and disclosure, but they are not automatically the same thing as asking a model to write the novel.

The distinction is creative responsibility. If my name is on the cover, I should be able to stand behind the creative decisions inside it.

The current controversy demonstrates why publishing needs vocabulary more sophisticated than simply 'AI' or 'not AI'. A writer who asks a tool to generate chapters is in a very different position from one who uses AI to build a launch plan or create a visual reference for a fictional starship.

Writers may need a creative audit trail

There is a practical lesson here that I suspect will become increasingly important: keep your process.

Keep early drafts. Keep revision histories. Keep notebooks, research, structural outlines, deleted scenes and dated manuscript files. Not because writers should be forced to prove their innocence every time somebody runs a paragraph through a detector, but because provenance is becoming valuable.

The irony is obvious. For years, writers have worried that AI companies could ingest their work without a transparent record of where it came from. Now authors may themselves need better records demonstrating where their own work came from.

Canadian literary organisations are already confronting the problem. Reporting from the Canadian Press says major prizes including the Giller Prize and Writers' Trust have policies barring AI-written works and largely rely on publishers to know how a book was produced.

Publishing needs process, not panic

There is a legitimate reason for publishers and literary prizes to protect human authorship. AI-generated fiction can now be produced at industrial speed, and readers deserve clarity about what they are buying. But protecting human writing also means protecting human writers from conclusions that outrun the evidence.

A sensible system would combine clear disclosure rules, publisher knowledge of the editorial process, manuscript provenance and where appropriate. Detection tools used as one piece of evidence rather than an automatic verdict.

That may be less dramatic than announcing that an algorithm has caught somebody. It is also much closer to how serious publishing decisions should be made.

The question for writers

If somebody accused you tomorrow of using AI to write your novel, could you prove that you didn't, and should you have to?

About Rob Frankson

Rob Frankson is the author behind 121 Minutes and the Near Galaxy Saga, writing science fiction while exploring the practical intersection of creativity, technology and publishing. Read more about Rob Frankson.

Join 121 Minutes

Get new AI & The Author articles, HMSS Broadside updates and news from the Near Galaxy universe. Join or subscribe to 121 Minutes.

Join / Subscribe to 121 Minutes

AI & Editorial Transparency

AI & The Author is edited and published by Rob Frankson. Artificial intelligence is used to assist with news research, initial drafting, content organisation and supporting imagery. All articles are reviewed and, where necessary, edited by Rob Frankson before publication. The opinions, editorial position and final decision to publish remain the author's.

Can an AI Detector Decide Who Wrote a Book? The Goncourt Controversy Raises a Bigger Question

In this article