Since this Monday, Thélyson Orélien’s novel C’était ça ou mourir, described as a “phenomenon of the season,” winner of numerous awards and still in the running for the Goncourt or the Renaudot, is at the heart of a major controversy: was it written by an AI? Before addressing this matter in detail, perhaps we should take a step back, as you do in your essay Une langue à soi (Stock). To what extent is the comparison between the advent of printing and that of AI, or ChatGPT, still relevant?
The appearance of generative AI is altering our relationship to writing, which itself was codified by Gutenberg’s invention nearly six centuries ago.
What is often forgotten is that what we take to be eternal truths about language — notions we use routinely: author, printer, plagiarism, quotation, copyright… — are in fact modes of our relation to language in the “Gutenberg era.”
In Antiquity, the transmission of knowledge and speech was governed by doxography, a topic Barbara Cassin has written about extensively: we say that x said y, without y being considered the exact words of x. Similarly, up until the printing press, there had never existed two identical copies of the same book — particularly in a literary world where lexical spelling had not yet stabilized.
For the first time, with the advent of the printed book, one could find exactly the same edition in Rome, Manila, and Veracruz — and more prosaically in Sens and Orléans. The notion of author changed, and with it the legal complex author-editor-printer that guarantees the authorship of texts.
What changes with AI?
Well, these vertiginous changes that we have somewhat forgotten and that have become norms for us are once again being upended. For the first time in history, language can be produced by no one, by something other than a person, by a vast chain of other languages, itself algorithmically surpassed by a wager on the credibility of a next element n+1 in the chain. Many old notions expire. To put it shortly with an example: in the Gutenberg era, there was no doubt that a text, even if anonymous, had an author; in the ChatGPT era, it is legitimate to doubt that a text, even signed, has one.
Even though this does not change everything overnight — I continue to provide my own responses to this interview, which can then be read as we are used to doing — I believe it signals a deep paradigm shift whose contours we have not yet finished mapping.
Is this how you understand the Thélyson Orélien affair?
We live in an algorithmically accelerated “era of suspicion.” As often, here is a concept of late modernity — one that Nathalie Sarraute coined to describe the crisis of fiction in the wake of World War II — that now takes on a new turn in the age of AI.
In short, can we not rely on algorithms to distinguish texts written by algorithms?
Can we trust algorithms to separate human language from the language of algorithms? As with COVID, we now speak of “false negatives” and “false positives.” I sense that some AIs tend toward the former, like ChatGPT, and others toward the latter, like Pangram.
This is a question that must be taken seriously. Personally, I have already experienced erroneous detections on texts I wrote, and failures to detect texts written by ChatGPT that I had barely touched.
I understand the statistical arguments, but I see a structural problem in the position of all those who defend the validity of AI detection in the Orélien case.
Ultimately, they are trying, by technical means, to put genius back in the bottle by restoring a definitive truth. I think the challenge we face is far greater. We must learn to live in a world where doubt is omnipresent.
In your book, you propose to find the key to this learning in an episode from the Quart Livre of Rabelais, one of the masterpieces of the French Renaissance. Could you explain it to us?
It is one of the most beautiful moments in the Quart Livre of Rabelais, in chapters 55 and 56. Pantagruel and his companions, still on their way toward the oracle of the Dive Bottle, sail at the edge of a frozen sea when Pantagruel arises in the midst of the ocean and cries: “Compaignons, oyez vous rien? Me semble, que ie oy quelques gens parlans en l’air, ie n’y voy toutesfoys personne. Escoutez.”
The crew panics; indeed, voices of men, neighing, cannon shots begin to be heard, but no one appears.
Panurge, as usual, yields to panic and it takes Pantagruel’s humanist wisdom to restore calm: he reviews several learned explanations drawn from the Ancients, before the pilot yields the key to the mystery. The previous winter, a bloody battle had taken place at that very spot and the cries, the clash of arms and the sounds of battle froze in the air with the cold. When spring came, they thawed and could be heard by people who were not there. Pantagruel even catches handfuls of them on the deck: the frozen words look like multicolored dragées and only regain sound when they melt in the hands.
What is the link between this tale and generative AI?
The first link, as we just discussed, is an analogy between Rabelais’s era relative to pre-Gutenberg times and our era relative to the pre-AI era. Rabelais, born in the 1480s, is, with respect to printing, what we would call a digital native. But the changes were then somewhat slower. The hypothesis I propose is that the radical strangeness of the episode of frozen words for a reader today would have been less striking for the contemporaries of Rabelais.
In what sense?
The tale, which we can only call fantastical given our relationship to language — which is transmitted mainly through writing, thanks to Gutenberg — could, in the eyes of readers of the early 16th century, recall memories of an era not so distant when texts circulated in ways that were less trustworthy, less legal, less certified than since the appearance of printing, where each text was necessarily a manuscript copy, and thereby the regime of authorship was different.
Are the frozen words of Rabelais in the age of AI an image of what Claude or ChatGPT offer us?
Generative AI opens a new era in which language may fall from the sky without needing to be attributed to an author. It speaks, as the saying goes. On this point, a linguistic tic I noticed among students I’ve taught seems emblematic of this paradigm shift: they often say “in the text they say that…”. This curious plural seems to signal a new regime of authorship in the post-Gutenberg era, in a world where language production is no longer reserved for human beings. There can be language like there is rain or traffic: by a strange effect, language joins Freud’s das-the-Id (the ça).
Back to the question of naming things, which lies at the heart of your essay. In the Orélien affair, there is much talk of “plagiarism.” But whatever one thinks about the substance of the case, that word seems not entirely appropriate and, in a sense, unfair. Should we invent a term to designate the act of signing a text written by artificial intelligence and the person who does it?
Exactly. In an interesting way, this affair shows that we are lost because the tools we have for judging it, dating from the Gutenberg era, suddenly become obsolete. The notion of plagiarism is, indeed, a good example.
Of course, if suspicion is proven, it will likely cost the author a handsome literary prize. Yet one cannot precisely brandish plagiarism, which would be legally unfounded. What should we call what this is? Will we read, love, or reward works co-authored by AI? What would it mean to credit AI as a co-author, when we know the algorithm itself also relies on an immense aggregation of other texts, without citing them?
Does the death of the author in the age of artificial intelligence vindicate the structuralists? Are we witnessing the revanche of Barthes or Foucault?
In a sense, yes. When Roland Barthes dreamed of “The Death of the Author,” extending essentially the Proustian wish to disengage from biographical critique, but this time for good, he clearly did not foresee the actual disappearance of authors through the invention of robots able to produce language. Yet the structuralists must be confronted with their contradictions: if we dream of analyzing a text as a pure combination of words (textus), freed from any consideration of the intention behind its production, what could be more marvelous than prose written by a person?
Be that as it may, the language produced by ChatGPT raises serious structural questions, both in terms of textual analysis and poetics, as well as in psychoanalysis and reflection on the constitutive dimension of language in the construction of the subject.
Halfway between Barthes (the zero degree of writing) and Foucault (the author function) in the AI era, with this controversy engineers seem to discover the problems of literature, and the literary folk those of the digital…
I am convinced that engineers and AI commentators would benefit from structuralist concepts to think about what they are seeing. They were completely wrong to skip this dimension; they are perhaps only now realizing it, given what they are discovering.
The paradox of generative AI is this: it pulls the digital from the purely mathematical or algorithmic era into a structuralist and literary era, in every sense of the word.
How could we resist this new era of algorithmically accelerated suspicion that opens up to all written productions in the ChatGPT era?
There are so many threats brandished around AI that I would not dare to say how to respond to each one! Perhaps there is one that is more within my reach, and within the orbit of what I reflect on in this book. I would summarize it as follows: what to do about the language of AI, whether to recognize it as such (in a conversation with an LLM), whether to read it between the lines (in a text signed but clearly fed by AI), or whether to receive it without recognizing it (in reuse that would slip past our radar)? To disappoint you, my answer is as follows: more than ever, we must learn to read.
In the world to come, the ability to read, as I mentioned at the start of this interview, will become one of the most irreplaceable and most discriminating skills there are. Critical reading, reflection on how meaning is formed, vigilance toward words—things that many think will disappear under AI—will instead be the conditions of our freedom in the age of its reign.
Why now, Je est plein d’autres?
In the age of AI, it seems to me that many prophecies of modernity, these lines by Marx, Rimbaud, Nietzsche, Mallarmé or Freud that we often pull out, take on a new meaning. One of them is the enigmatic “Je est un autre” that Rimbaud wrote in his letter to Paul Demeny on May 15, 1871, when he was only sixteen and bored in Charleville while Paris was alive with the Commune’s fervor — which would soon end in blood.
It is a sentence that can be reread through the lens of psychoanalysis, Lacanian a fortiori, of course. There one hears the idea that what in me says “I” is an imaginary and symbolic construction in which the image of the other and their language enter. But if we shift this to the era of generative AI, we must go further. What happens if I adopt the AI’s language, whether to copy it or to take its discourse seriously? Which other enters my language? Who speaks to me then?
Because the AI language is composed of an immense aggregation of languages (human and now also non-human), one can say that je est plein d’autres, including others I do not know or recognize as subjects. Others speak to me (speak through me) without me (that is, who?) even recognizing them as others. It is dizzying.
What are the dangers for our identity and our unconscious when we let being spoken happen, that is, when the machine speaks for us?
When we talk about their geopolitical, geoeconomic and geostrategic dimension, the dangers of AI are often addressed from a cognitive angle. All these approaches are essential, and their cross-fertilization even more so.
But in my view there is another approach missing to the linguistic aspect of AI (that is to say, a portion of it, I admit), one that draws on the humanities, the science of language, and psychoanalysis. We have too quickly discarded analytic critique in favor of cognitive sciences — in this topic as in others — and in doing so, I think we have a lot to lose.
The fundamental problem of generative AI viewed from this perspective is the question of the subject and its freedom. If I let AI speak to me, what do I give up as a subject? What subjection do I condemn my subjectivity to? What part of what in me says “I” am I alienating, and for the benefit of whom? I believe analytic illumination (drawn from Jacques Lacan, though other structural approaches could lead there too) is a key to rethinking humanism in the AI era.
Far from being a way to remove the AI debate from geopolitics, it is a way to give it a deeply political sense, which questions us as subjects within democratic systems whose social contract rests on the exercise of our freedom.
I ask you about page 143: “Are we heading toward a blessed era of real democratization of access to language, or toward an invisible diversification of forms of illiteracy?” Without claiming to be the solution, how could generative AI possibly help reduce illiteracy?
Undoubtedly a bit of both, even if there is neither blessing nor curse in all this.
If we are not careful, the development of LLMs risks, in my view, exacerbating inequalities by giving the surface impression of smoothing them out. People who struggle to write will become more invisible, even more ashamed, which will amplify avoidance strategies that lead to the reproduction of inequalities. Moreover, generative AIs are tools whose output quality depends on the quality of the prompt, which itself is formulated in language. In other words, people with writing difficulties, whose prompts will be less precise, are more likely to receive AI-generated content of lower quality — less precise, more standardized, with a less distinctive style — while having fewer means to critically read these contents. It’s a cycle that increases the loss of freedom and power that illiteracy represents.
On the other hand, I also believe AI can be a formidable tool to help overcome illiteracy, provided it is used correctly — that is, schools should teach how to use it to help oneself. Used well, by focusing on precise work on short excerpts already written by students rather than replacing them with synthesis, AI could serve as a personal secretary for each learner. It can accomplish what a teacher cannot do for thirty students: individualized correction, taking the time to explain each error and offer several ways to correct it. That every person should have access to a private tutor, a language-master tutor, free of charge, I believe this would be progress.
The book’s epigraph, “Mal nommer un objet, c’est ajouter au malheur de ce monde,” evokes the famous passage in One Hundred Years of Solitude when, after the insomnia plague and the resulting memory loss, the Buendía family labels every object in their home to remember its name, and thus its use. Can Camus’s phrase be read on two levels? It is important to name the ills (such as illiteracy), and naming things well is precisely what is at stake here to avoid falling into illiteracy…
Indeed, there is a double meaning, and the image of the Buendía family sticking words to things after the immense insomnia is very beautiful. This image also captures the fragility of language and its transmission: one must name each object, not to remember its name — one could swap “fork” and “radiator,” and that would not be catastrophic — but to remember its function itself. What García Márquez points to here is that words also form our being-in-the-world, that it is through them that the real — this absurd mass we stumble upon — makes sense, becomes a livable world.
Let us return to the book’s epigraph: “Misnaming an object adds to the misfortune of this world,” by Camus. It is a line that has been reused so much that I almost hesitated to place it as an epigraph. But it seems so just that I could not do otherwise. It hits the mark precisely thanks to its double meaning, which is more like a squared double meaning. The lack of mastery of language is a defect of freedom, in the political sense. By leaving a large portion of our population with a certain degree of illiteracy — for there are degrees, it is not black or white — we allow immense violence to operate; we amplify virtually all inequalities; in short, we add to the misfortune of this world. Not naming this situation in turn is to square that violence.
Why is there talk of a “new betrayal of the clerics”?
Because the irresponsible ones, here, are us: the intellectual elites of liberal democracies. We would rather lament the idea that “the youth no longer reads,” that they “no longer know how to speak,” than seriously ask what kind of world we are building, what relationship to language we have, and what methods we can invent to rebuild a relationship to reading and critical sense today, in the world we know.
On one side — to the right — there is a declinist music linking a decline in language to a civilizational decline. In this gap, many racist and classist imaginaries easily slip in. On the other, out of fear of lending credence to this Zemourian reading, some close their eyes. Colleagues of mine, gathered in the collective Les Linguistes atterrées, published a widely read tract titled Le français va très bien, merci. While it rightly recalls certain historical elements about the evolution of languages, their argument omits a vast malaise currently cutting through much of our society — not just at school — which I associate with illiteracy.
The trap formed by these two positions (these two postures, really) is what constitutes the new betrayal of the clerics: two elites face off, each distorting the problem while it worsens, deeply harming the foundation of our democracies.
We often hear that the children of workers wrote better fifty years ago than the children of executives do today. You mention in the essay the evolution of the level of dictations (p. 27). Why might this decline in language even lead to a stage of “no speech”?
It’s funny: the lament over the “language level of the youth” is something quite common, and I often hear people on the left maintain a rather conservative stance on language. I don’t particularly want to get into this debate, for methodological reasons: I analyze what I understand about our relation to language today, as I teach, in the AI era, at a time when I observe that a large portion of the population (I’d say one in four) is in a state of great distress regarding written language. There is indeed a problem, and this problem is fundamentally political. In a democracy, the tool we have to share our ideas, to make our voice heard, to exist as free subjects, is our speech — both spoken and written. It is guaranteed by a core rule, the freedom of expression.
Today, arguments that claim freedom of expression is under threat come more from extreme libertarians who cleverly equate freedom of expression with the right (or rather the non-right) to say anything — particularly horrors. But in truth, I believe freedom of expression is threatened from another end, less often considered. For this fundamental right is not decreed once and for all. It is realized in the exercise that each person can undertake. And this exercise, even before the question of access to media (which social networks have largely democratized), passes through mastery of language itself. A weak command of language is an obstacle to freedom of expression — to freedom itself.
You write (p. 65): “I experience this daily: students are drawn to the finest literature precisely because they sense its beauty, provided we know how to show them how to think with it, how to approach it (and not how to bring it to them), and how to put something of their own intelligence into the dialogue with these forms of the past.” How to approach literature?
By a thousand paths — anyone who loves reading knows that the path that led them there is intimate and, to a large extent, impossible to replicate. But school serves this purpose too, I am convinced. It is not about imposing from above a “great literature” that we must persuade students to adore, as the schoolmaster in Jules Vallès’s L’Enfant, who cries: “Kneel! Kneel before the divine Racine!”
But it is not either, as I say in the excerpt you cite, about bringing literature to the students, if by that you mean, for example, starting from their daily lives, their musical tastes, or a TikTok video to show them that, in their time, the Enlightenment philosophers were also influencers and the poets of rappers. This way of diminishing the border between today’s students and the great texts of the past disgusts me a little, especially since it is usually unproductive and implies ridiculing the teacher who wants to show that he has the codes of the youth. On the contrary, I say that my task as a teacher is to lead students to the texts.
What does that mean?
It means building a method to enable them to enter into texts whose difficulty must not be underestimated. A text written in 1550 is very difficult to read today — for everyone! The same goes for a poem by Rimbaud or a page of Proust. We must stop believing that we make the texts understandable to students. I also don’t think we make them love anything by asking them to identify adverbs, personifications, or metaphors.
On the contrary, we should share with them that reading, real reading — and the pleasure that goes with it — begins when we understand that we do not understand everything. When we discover that the ant is not worth more than the cicada or when we wonder what Rabelais had in mind when he compared the crew’s ears to “beautiful oysters in escalle.” It is there that we can introduce vocabulary, stylistics, and poetics to support this astonishment, to help it structure itself and advance.
Learning to read is slow and long; one should never take that for granted — not with adults, nor with adolescents. Entering into a text is a demanding journey, and it is by sharing with students this sense of discovery that we can, I believe, awaken them to the pleasure of critical reading.
The book rightly focuses on teaching “French” and on the choice (sometimes by default, it is stated) of terminal-year students toward the so‑called “literary” options. Should we not also discuss the opposite choice, which would be to turn toward scientific disciplines “because one is bad at French”?
Alas, no. For two equally unfortunate reasons, the sorting occurs this way. The first is the very French myth — and deleterious for our society — of the “maths talent.” The idea is propagated that you are born good or bad at math, and that if you don’t manage it, it’s because you’re simply not cut out for it. Fortunately, some mathematicians resist this and propose new teaching methods — I think of Hugo Duminil-Copin and Emmanuel Sander in De l’autre côté du tableau noir (Seuil, 2026). This doxa also reinforces inequalities, notably of gender.
To this is added another bad reason: the belief that literary specialties — for the baccalaureate track known as HLP (Humanities, Literature, Philosophy) — are not subject to an entrance barrier.
Why?
Because we lower the requirements; I believe, for example, that the HLP specialization should mandatorily include the study of an ancient language, which today is an additional option and is disappearing. We tell ourselves that after all, if we can read, we can try to say something about a literary or philosophical text.
You see that this reasoning does not hold: if we treat literary specialization as a soft subject where everyone has the right to have an opinion, we could just as well treat mathematics as an open space where everyone has the right to have a brain. Yet it reveals a society that has somewhat forgotten what it is for to learn to read seriously — and it is not the fault of the young.
In the chapter “The Great Paradox,” how can we counter the argument that certain terms or expressions (such as “transport passes” instead of “tickets”) paradoxically aim to be more precise because they intend to cover several kinds of tickets? Isn’t that a return to Camus’s misfortune epigraph?
Indeed, one could reply when I rail against hypernyms — i.e., the umbrella terms (vehicle rather than car, fuel rather than gas, transport passes rather than ticket, etc.) — that I am caught in my own trap, because these terms help avoid misusing language: we say “transport ticket” because there are dematerialized tickets, subscriptions, discount cards, and so on. In my view, it is precisely these expressions that abuse language by drowning its spontaneity in a bland, disembodied magma.
The bureaucratic jargon, which claims to restore strict alignment between words and things through these inventions, is deadly for the language. I dare to offer an example that may sound a touch conservative, though nothing in what I defend is conservative, but so be it. When the Civil Code had to be invented in 1804, you can imagine that many concepts exceeded the definitions of language in its customary uses. Was there a need for neologisms and jargon? Not at all. True precision does not entail sacrificing clarity — quite the opposite: any loss of clarity, even for the sake of precision, is a form of impoverishment.
But there you have my point of view. I concede, as I’ve said, that people may think differently — thank heavens! I do not defend a conservative view of language. I defend instead a social function of writers, of intellectuals, of those who work with and on words: the duty to watch over the language, not in the sense of preserving it (which would be conservatism), but in the sense of watching over a sleeping child. We observe, we can form hypotheses. That is our role: to know how to grumble, not to restore a prior state of language, but to help others awaken to language, to beware of words, to notice certain linguistic tics. I believe this is fundamental.