Nurse, Scalpel!

"Guillaume Allary, founder of the publisher Allary Editions, said forbidding authors from using A.I. was like barring them from ordinary search engines."

— The New York Times, Charges of A.I. Writing in a Lauded Novel, and a Fight Over the Literary Future, September 25, 2026

Earlier this fall, the Académie Goncourt removed Thélyson Orélien's debut novel "C'était ça ou mourir" ("It Was Either That or Die") from the list of contenders for its literary award after an anonymous tweet claimed that an AI detector measured its chances of being human-written at zero. The novel documents a migrant's dangerous journey from Haiti to Canada, partially inspired by the author's own road to Montreal after his family home in Port-au-Prince collapsed in the 2010 earthquake that killed about 160,000 people. The account behind the accusation was partly run by a self-described "anti-woke" columnist at Le Figaro, also employed at a venture capital firm investing in at least six AI-powered businesses. The tool used for AI detection is Pangram, which outperformed two main competitors in a 2025 independent study, though how it does so remains unclear even to Pangram's founder, who called its algorithm a black box. Other novels nominated for the Goncourt Prize passed their Pangram tests.

This piece is not about the multiple political dimensions creating this scandal. In response to both valid critique and racist harassment that followed, Orélien denied ever turning to generative AI for text creation and admitted to using research tools for help with idioms in unfamiliar languages; the main character's route lies across Spanish- and Portuguese-speaking places in the Americas. Differentiating between researching and generating sounds sensible enough, and yet a certain resemblance between the two led Guillaume Allary to form a simile.

Both are fairly recent inventions in the grand scheme of human writing tools. Still, the main idea that makes Allary's comparison possible is that ordinary search engines are not what they used to be even a couple of years ago, before 42% of American adults used generative AI chatbots specifically to search for information. If the most ordinary engine is Google, then its 90% share of the industry means that, for a great many people looking things up online, the first thing they see is not the results most matching the query, but a built-in AI Overview of sources it aggregated. In its own words, "Google does not offer a permanent, official 'off' switch to completely disable Google AI overviews or core AI features"; it does not allow me to copy and paste this sentence, and I type it by hand like a medieval scribe. There is nobody to attribute this quote to, however, because no one has ever said these words in that order. The text the AI Overview linked as a source of information used slightly different wording; it came from one Google Chrome Help forum user responding to another, who asked how to disable the feature.

Allary's point will most likely resonate with authors who rely on so-called assistive AI as opposed to mindless text generation, but my point is they are not different. Much of the assistive AI used for grammar and spelling correction or summarizing research (helpful, ethical, difficult to avoid) relies technologically on generative AI (soulless, derivative, plagiarizing). To once again cite the machine, "Yes, Google's AI Overview (formerly known as Search Generative Experience, or SGE) is a generative artificial intelligence feature." The division is designed to self-soothe and signal virtue: there are authors who define their use of AI to generate paragraphs based on prompts as assistance, because it technically does assist them in text creation.

My point is that, despite the growing omnipresence of performance-worsening AI in search engines, translators, synonym dictionaries and other tools used by authors, forbidding them from using AI assistance is not the same as banning online research altogether, and the difference does not lie in the good assistive AI and its evil generative doppelganger. We cannot separate those. It is in what AI as a tool can do, which is something that lead pencils and IBM ThinkPads cannot: actually creating the text for us. Before generative AI, there was no magic quill or keyboard that could come up with plot twists, descriptions and finales upon request. This kind of help for authors goes beyond any technologies designed to assist people physically unable to hold a pen or use voice-to-text, and attempts to solve the struggle of writing on the cognitive level.

Struggling to produce a piece of writing often comes down to not being sure what you want to say, which is the same thing as not knowing how to say it. Both also mean not having a clear-cut idea of what you think. Joan Didion was not talking only about herself when she said, "I write entirely to find out what I'm thinking, what I'm looking at, what I see and what it means." AI is a tool assisting writers with thinking.

We should not use tools to assist us with thinking precisely because there is no way to cleanly separate assistance and doing it for us, in our stead. Turning to a language model for inspiration and ideas comes dangerously close to asking somebody else to take over your cognitive process for you. Coming up with ideas yourself and then appointing AI to choose the exact wording to convey them is the same desire to share the intellectual effort, delegate the task of cognition. We delegate tasks when they are so complicated as to demand professional expertise, like that of a lawyer or electrician, and when they are simple and tiresome enough to be accomplished by some sort of intern whose work we only need to assess and review afterwards. Thinking cannot become either of these two things, because cognition and ideation should be something one trusts oneself to carry out. Trusting AI over oneself means relying on the competence of nobody in particular, standing on the shoulders of everybody ever. In a generated text, the collective responsibility for authorship gets diluted among everyone from Enheduanna to the most downvoted commenters on Reddit and Yahoo Answers.

We understand that freedom and responsibility exist in a quantum entanglement, where one cannot be relinquished without the other. The third thing tied to them as tightly is self-respect, and since a computer must never make managerial decisions as it can never be held accountable, then whoever has little enough trust in their cognition to surrender their freedom of thought and speech to a computer must also let go of self-respect, and with that the respect from others. Many conversations about AI are about the politics of its creators and users in relation to others; it is undeniably political as a technology and as a cultural phenomenon. Naturally, book publishers and literary prize committees will establish any participation rules they consider fair, and in deciding how much AI assistance is allowed there surely will be politics at play. For all that, trusting nobody, including oneself, is always a personal decision.