In January 2024, the news caused a stir. The winner of the Akutagawa Prize, Japan’s highest literary award, had just admitted, in complete relaxation, that at least 5% of her book had been written using ChatGPT.
With this 5% of “cheating”, one would be tempted to say, at least a departure from what is supposed to be the fruit of a human creative process, coupled with the confession of the crime by its author, there were enough to raise eyebrows in the international editorial community.
However, according to Eat the Beethe 33-year-old novelist from Saitama, there is nothing scandalous about it. His novel, Tokyo-to Dojo-to(“Tokyo’s Tower of Compassion”), takes place in a futuristic Tokyo where artificial intelligence plays a central role. For the sake of realism, Rie Kudan placed the words of a real AI, ChatGPT, in the mouth of this AI. Shocking? Yes and no. The novelist explained that she uses artificial intelligence to stimulate her own imagination, which remains the main author of the content.
The following month, in February, it was at home that the debate was reignited. Jonathan Werberthe son of the author of Ants Bernard Werberpublished his third novel, The best writer in the worldat Robert Laffont. Its main character is an AI, V39, invented by Thomas, a young and brilliant engineer. He, to test her, asks her to write “the perfect detective novel”. A challenge, in which the AI turns out to be… useless.
And there, it’s no longer fiction. Jonathan Werber, questioned to know if he had, like the Japanese Rie Kudan, called on an AI to make the AI of his novel speak, admits that he ” try “and that he would even have “well-intentioned”. “I inserted short stories into my novel supposedly written by my AI, but by generating them by ChatGPT, the result was so mediocre that even to serve my purpose, which was to show that the AI is not conclusive in In this case, it was not usable”, he confides.
The Web is full of perfectly artificial successes
Too bad to be a writer, AI? However, the Web is full of perfectly artificial successes. Allan Trevorthe author of 1,491 books written in just two years, has made a small fortune on Amazon. The giant removed it from its catalog when the web caught fire at the revelation of the deception, but Allan Trevor has not yet made his final point. Forty new works were published in August alone. The titles Marijuana and orgasm, hypnosis brutal, or How to masturbate your partner promise hours of unforgettable reading on the Fnac website. Even his portrait, with the venerable white beard, was generated by the Midjourney AI. And what about Fire and Fury, by Doctor Miles Stone?
Behind a title that evokes faulknerthis essay subtitled The implication of global climate change in the Maui firerose to the top of American sales in the heart of summer 2023, while a mega fire ravaged the small Hawaiian island. Le livre sonnait si juste en plein traumatisme, presque en temps réel et ponctué de rapports officiels, que d’aucuns ont même supposé un moment que son auteur puisse être le pyromane criminel à l’origine de l’incendie… Là encore, a pure algorithmic creation. Pour calmer les foules, Amazon a retiré le titre, encore, et limite désormais le dépôt de nouveaux livres à 3 par jours sur sa plateforme d’autoédition. Would more be too much?
Algorithms are today incapable of writing valid “narratives”
Jonathan Werber, to avoid the platitudes of AI in his novel, therefore began writing “in the style of”, to show what seemed interesting to him: “Writing defects”.
The best writer in the worldwhich can be read as a vade mecum of generative AI as well as AI for Dummies from First Editions, offers a nice overview of the joys and limits of exercise. Jonathan Werber considère que les algorithmes sont aujourd’hui incapables d’écrire des « narratifs » valables, sauf à être extrêmement nourrie de détails. “The AI is blocked, because it cannot take risks. Its products must be clean, not shocking, and this makes the story not bad, but totally uninterestinghealthy »
Another gap in long texts is that artificial intelligence would be incapable of doing what English speakers call setup and payoffwhich can be translated as “preparation and regulation”. “Basically, the novelist brings in an element that will resurfacegir 80 or 120 pages later, adds the writer. The machine does not know how to produce this type of echo. »
“AI is a raw material. It’s like electricity, you can make an electric chair or Christmas garlands from it” (Alexandre Jardin)
However, AI has not yet written its last word. “AI is a raw material. It’s like electricity, you can make an electric chair or Christmas lights out of it.”bounces the facetious Alexandre Jardin. For him, having a strong opinion on the subject is “as stupid as being for or against electricity”. And if the debate turns around the axis “with or without AI, like the pure and the impure”this is because we only consider ChatGPT type AI.
But asking ChatGPT to write your book is “completely stupid, even the opposite of literature”. “North American writing tools come from a history other than ours. Our mental paths are not the same. We know that literature is the space of the greatest freedom, while they categorize it into pure and impure. »
Le drôle de zèbre a donc eu l’idée de concevoir une IA qui, à l’inverse des IA génératives, vous pose des questions plutôt que de vouloir y répondre. Its tool is called yourscrib.ai Launched last spring, this software is designed to dialogue with yourself. “It’s a mirror effect, to speak with your unconscious, to find the major subjects that you carry within yourself, if you wish to write,” he declared in a video posted on Facebook to explain the shutdown of his software at the very moment of its launch. “All the servers blew up!” “, he laughs.
Alexandre Jardin had planned everything except the highly addictive nature of this AI. « On s’est rendu compte que les utilisateurs restaient la nuit entière dans la phase exploratoire, c’est-à-dire l’étape de découverte de leurs sujets les plus profonds. » Yourscrib, the AI that literally “prompts” its human, is available in three stages. The first, Discuss, promises to extract “your real topics, emotions and opinions”. The second, Design, allows “to explore different scenarios and plots in your story without fear”. The third, finally taking action, Writing, to direct and write “your story with the help of artificial intelligence”.
“If it’s free, we’re the product”
And as we know that “if it’s free, we’re the product”, the site opens the cash drawer above all. THE pricingsince we’re talking a little globish, announces the price in AI units, that is to say in tokens, which is equivalent to a certain number of words in a text. “No more than 100 euros for the first stage”assures us its inventor and owner. But there is also the Françoise button, like Françoise, like Verny, her first editor. Françoise is, it seems, very greedy for tokens. “This is a very important step, which tells you if you are moving away from, or if you are getting closer to, your real subject. »
Alexandre Jardin has also calibrated Françoise on one of his own books, an old one, published in 2011 by Grasset, Very good people. “It’s a book about my collaborator grandfather. Au bout d’un moment, Françoise m’a dit : “vous avez écrit ce livre fondamentalement pour libérer vos enfants de cette mémoire familiale, mais vous l’avez terminé sans en avoir dit un mot. Vous vous êtes égaré. Vous avez été fascinated by your grandfather and the questions that his life raises in you, you missed your children, you missed the reason for your book“. »
AI slips into gaps
Cruel, Françoise, but fair. Alexandre Jardin sait que c’est vrai, et s’il avait eu cette information au moment de l’écriture, il se serait sans doute rapproché de son intention première. “But the worst part is that the tool didn’t give me its point of view, it gave me mine! “, he laughs.
A software capable of stimulating creativity and then guiding the novelist, we worry that Yourscrib could then replace the work of the editor. But the novelist wants to be reassuring and thinks that, on the contrary, this will lead “a democratic revolution”. “When you don’t have access to the city center, where power is shared, you don’t have access to this type of advice. So that doesn’t mean that my AI replaces the job of editor, it means that it slips into the gap. I don’t believe that things replace each other. It’s something else that’s being created. » L
Human creation: the organic publishing label
The book, with or without generative AI? Or rather, how can we distinguish the wheat from the chaff? Since this year, the French self-publishing platform Librinova has been marketing a solution for authors wishing to claim the organicity of their creation: an AI-free label, called “Human Creation”, affixed to 100% human books. Nicolas Gorse, general manager of Dott (scooters), came up with the idea in the face of a void on the subject in the publishing industry. Librinova made him her partner. AI tracking in a book is carried out by combining three methods. The Transformers, for starters. A deep learning model that allows you to analyze long texts. Next comes “the mathematical detector”, explains Gorse, in which words are transformed into tokens whose homogeneity is checked. “Unlike human language, an algorithm repeats itself. » En cas de suspicion d’utilisation de l’IA, la troisième phase, un entretien humain sur le processus de création du livre, permet de percer immanquablement les faussaires, nous -assure-t-on. The authors alone decide whether or not to take the label. The co-founder of Librinova, Charlotte Allibert, is keen: “The use of AI is not for us a moral consideration, it is a call for transparency. » Will Human Creation establish itself as the organic publishing label? “Not if publishers don’t take up the subject, warns Charlotte Allibert. Technology is scary, and it’s normal for things to take a little time to settle down. But on the reader’s side, I think that being able to identify the origin of a book will become fundamental on commercial sites. »
Related News :