2. THE BAD BIRTH

In our field — literature — as in many other domains, artificial intelligence is generally perceived as a source of injustice. And rightly so: language models were trained on millions of texts accessible on the Internet, including our pirated books illegally uploaded to shadow libraries like Z-Library and Library Genesis, or to the Open Library project. The AI systems we interact with every day learned to speak thanks, among other things, to our literary texts. ChatGPT’s frequent use of the em dash — which many perceive as a telltale sign of AI-generated writing — suggests that literature had a real impact on training: it is in novels and essays — or more broadly, in the entire corpus of edited texts — and not on Reddit, that one tends to overuse this kind of typographical mark.¹ If LLMs had been trained only on public domain texts, we would have nothing to object to, but we know that is not the case. One need only query chatbots about contemporary, copyrighted authors to confirm that they ingested them during training.

Writers feel doubly cheated: first plundered, then imitated by an AI that becomes an unfair competitor within their own profession.

It should be noted that for now, the average continues to exert on artificial intelligences a pull comparable to gravity in the physical world, and they still struggle to tell a vivid image from a cliché. Moreover, they remain incapable of architecting an entire large-scale work, placing every piece in its rightful position, building the book as one builds a cathedral — to borrow an image from Marcel Proust.

But things are evolving rapidly. For now, I continue to write by hand (this essay, for instance), but I have been running tests on the machine regularly for over a year. The received idea that AI prose is bland is already far less true than it was in 2024. There is also a vast difference between free models and paid ones: Claude Opus 4.5/4.6 and Gemini Pro 3 — whose access is limited without a subscription — produce impressive results. Various techniques also exist for shaping AI-generated prose — I will return to these.

Piracy joins the long list of grievances held against artificial intelligence. Writing with AI, the argument goes, would mean endorsing the plunder and, in a sense, violating once more the rights of authors who have already been trampled — oneself included. All of this is true to a degree. The history of technological disruption is a violent history. It was not Silicon Valley that invented the modus operandi “Move fast, break things.” This kind of recklessness can be found throughout the history of text technologies. One of the earliest printers, Aldus Manutius, seized the works held in monastery libraries, which he claimed he wanted to “liberate” from the monks’ “harsh and gloomy prisons.” The printing press put the copyist monks out of work; the Benedictines petitioned the authorities against the printers, and in 1476, Parisian scribes physically attacked a press. The world of print — with which we associate ourselves so unselfconsciously today — also traces its origins to a “bad birth.”

One might counter that copyright did not yet exist at the time. True. And I do not deny the dispossession taking place today. I would even go so far as to say that AI likely marks the beginning of the end of an era: the copyright era inaugurated in 1710 with the Statute of Anne, then consolidated and mythologized in the nineteenth century by the French Romantics. The idea of a text originating in the subject, in the Author-Self, is being undermined by the new aesthetic regime that artificial intelligence is ushering in. What meaning does this Self-as-origin-of-the-text retain if writing no longer comes from within — interiority, mind — but from without, from the machine? An outside that has acquired the gift of language through an extraordinarily complex remixing, in which the fragments of text that once belonged to us have become drops of water in a churning ocean.

I am well aware that these claims about copyright will not be popular. I am not preaching the end of copyright in the manner of “free culture” advocates who, under the guise of universality, feed platforms that deprive creators of their livelihood. What I am saying is that copyright, in its current form, will not be enough to protect us — and that clinging to it like a life raft does not constitute a strategy. Let us be clear-eyed: the subjective regime of literature is already in crisis. The number of published books has exploded, their shelf life in bookstores has plummeted, the share going to each author has shrunk to almost nothing, and very few writers make a living from their pen. The book has persisted less as a livelihood than as a badge reflecting symbolic capital — status, recognition, prizes, grants, literary festivals, invitations, and so on. Copyright as a means of subsistence is already largely a fiction. It still works for books that find a wide audience, but writing that is “an assault on the frontiers,” as Kafka put it — writing that searches, that takes risks, that defies expectations — no longer has a viable economic model.

The copyright regime is already exhausted, and an entirely new paradigm is waiting to be born. We are in the in-between. How will practitioners of “literature in the making”² earn a living if copyright no longer holds? This is a complex question, and before addressing the economic problem, we must first pass through aesthetic territory by asking: what form can the art of writing take in the age of intelligence?

---

1. In French, prescribed usage actually calls for the en dash, not the em dash. Yet ChatGPT tends to import American typographical conventions even when writing in French — an imperfection that will no doubt be corrected sooner or later. ↩

2. To borrow a phrase François Bon used in the era of publie.net: la littérature en train de se faire. ↩