Chapter 6

Improbable Writings

9 min read

6. IMPROBABLE WRITINGS

Improbability may well become the new name of literariness.

We touch here on a vast and complex historical question: by what criteria do we distinguish non-artistic texts from texts that belong to the art of writing? Jacques Rancière has offered very compelling answers to this question, exploring in particular the transition — heralded by Madame de Staël’s On Literature in year zero of the nineteenth century (1800) — from the representative regime to the aesthetic regime of the art of writing. To summarize: under the old regime, texts were deemed artistic when they operated within mimesis, according to a framework of representation of the real, whereas under the regime inaugurated in the Romantic era, they are deemed artistic when they gather the signs of the real in a way that is sensible — felt, perceived.

Is it possible that at the dawn of the age of intelligence, we are beginning a shift from the notion of sensibility to that of improbability as the criterion separating artistic from non-artistic writing? We would thus be moving — I am riffing on the concepts here — from the aesthetic regime into the algorithmic regime of the art of writing. It would be mathematics that determines what is literature and what is not, and the most reliable criterion of distinction would rest on the notion of statistical probability. If a piece of writing falls within a region of high statistical probability, it might be an email, a report, a briefing, a legal opinion, and so on. If, on the other hand, a text falls within a region of very low statistical probability, it could be considered literary.

This distinction raises an obvious problem: literature likes to take up high-probability linguistic forms. In her Pillow Book, Sei Shōnagon accumulates inventories, lists, and factual enumerations. In Walden, Henry David Thoreau includes lists of materials and account statements. In The War, Marguerite Duras presents notations that resemble administrative prose. Even so, I do not believe these examples invalidate the thesis of statistical improbability. Everything is a matter of angle and context: the literary text will adopt a certain distance from the high-probability writing model. This distance can operate through enunciation, narration, the transition from the factual to the fictional, from the administrative to the poetic, or even through format: a list that runs to 200 pages, for instance, would be contextually improbable. There remains a knot here that is not easy to untangle and that will no doubt always be a problem and an object of debate. How many critics have faulted autofiction writers for their lack of literariness? “It’s nothing but testimony, diary writing — it has no literary value,” and so on. An entire text can be written in high-probability prose and yet, beneath the surface, make itself so flat, so extremely probable that it becomes, paradoxically, improbable.

Not only can the notion of probability become a new, complex — non-binary — criterion for distinguishing literary from non-literary texts, it can also separate the wheat from the chaff within the field itself. We all know instinctively that a large portion of texts published today relies on reproducible language and reproducible narrative models. I am thinking here of genre fiction (romance, crime, science fiction, etc.) — which I do not deny can be inventive within its conventions, but whose conventions themselves belong to high probability — and also of the vast majority of novels published today. While in the nineteenth century the novel was the “genre without genre,” to borrow Rancière’s phrase — a space of sensibility to the world undetermined by the categories of mimesis — it has become, over time, a genre of high normativity. One must understand that in the nineteenth century, a reader who opened Flaubert’s Madame Bovary or Zola’s L’Assommoir saw not pure fiction but a laboratory of contemporary society — hence the scandalous nature of these novels. The subtitle of Madame Bovary — “Provincial Manners” — offered a clue to the societal study within. Today’s novel no longer carries that sense of “experiment” (in the quasi-scientific sense) that felt so vivid and so real at the time. Telling a well-crafted story with themes drawn from contemporary society is not at all the same thing as using language and fiction as a genuine laboratory of the social. Accustomed to the codes of the novels that shaped their culture, most contemporary novelists implicitly accept a certain number of conventions. We find it natural to open a novel whose author and narrator bear different names, like a sign of facticity hung at the entrance to the book. Proust at least maintained the ambiguity and tended the illusion by suggesting his narrator was called “Marcel.” Alternation of dialogue and narration, division into scenes and episodes, elimination of repetitions, completion (even when the ending is narratively open): there is a set of specifications that most novelists follow.

This is why I suspect the novel will be the first casualty of artificial intelligence. In the territory of pure fiction, AI will excel: it will recreate all the artifices of the novel or the short story — that will be its path of least resistance. Its stumbling block, on the other hand, will lie in the territory of language rubbed against the real: memoirs, notebooks, journals, and the like. I am not saying these forms will be entirely spared by AI, but the machine will not be able to write such texts autonomously without breaking the nonfiction pact.

There are, however, ways to use intelligence for these friction-writings. In the case of a travel notebook, for instance, the human will need to supply the AI with the raw material (notes, fragments, dictated indications, etc.) and keep close watch while it braids the sentences. Unless, in a perhaps not-so-distant future, one could carry a camera along the road and obtain prose or poetry from it — a visual-to-text process whose conceptual ancestor is Stendhal’s mirror: “a novel is a mirror carried along a high road.”1 Beyond nonfiction, it also seems to me that AI will stumble against forms that reinvent fiction out of its own necessity — that is, once again, out of a relationship with the real. Texts whose fiction is not factitious but rather deployed from a diction that falters will be difficult to automate. When saying the real is not enough and fiction is born of that insufficiency, we plunge into a grey zone that AI will struggle to penetrate, because the conflict with the real can only be undergone; it cannot be guessed at or imagined as in conventional fiction.

In the highest strata of the novel, there is sometimes stylistic inventiveness, but it almost always remains contained within dialogue or narration. When language becomes improbable one level up — in the enunciation, that is, when the voice itself is called into question — one begins to slide out of the novel and into the récit. The récit is not a genre without genre; it is not a genre at all. It is the sheer genericity of telling that crushes the notion of genre. Dominique Rabaté, in Vers une littérature de l’épuisement and then in Poétiques de la voix,2 showed convincingly how the question of voice lies at the heart of the art of the récit. To recite is to displace narration toward the theatre of speech. It is this displacement — in which the situation of enunciation is made visible — that we observe in Thomas Bernhard (the narrator settled in his wing chair in Woodcutters), in Pierre Michon (“Since you insist, Monsieur, I am willing to linger a moment longer with you in the grand staircase”)3 and in several other authors who have, over the past fifty years, drawn the narrative art toward recitation.

The notion of improbability can help us distinguish novels that reproduce already consolidated sentence-level and narrative norms (high probability) from prose that ventures into unlikely syntaxes and arrangements.

Some will say this is merely a derivative of the notion of deviation, which has already been the subject of several theories of literariness. I believe the concept of deviation did indeed pave the way. According to this theory, prose or poetry is literary when it finds its region of deviation in the domain of language. In Kerouac’s On the Road, the deviation is highly visible: the sentence defies logic, punctuation, uniformity of register, and so on. But it is not always so visible. Deviation can also nestle in the middle, as it does in Duras. In the sparse, in the blank, an interstice opens where language finds its literariness. This is why, in my style tests, I included minimalist prose: contrary to what one might think, it is not a style of high statistical probability. There is a rarity to blank writing just as there is a rarity to spontaneous prose.

With the shift to artificial intelligence, deviation transits from the aesthetic to the statistical. The distribution of the sensible rested on aesthetic criteria; the distribution of the probable rests on algorithmic criteria. An improbable prose or poetry is not merely writing that deviates from norms; it is a text that had little chance of coming into being in probabilistic terms.

“The opposite of what is purely spoken, the opposite of the poem, is not prose. Pure prose is never ‘prosaic.’ It is as poetic and hence as rare as poetry,” wrote Martin Heidegger in On the Way to Language.4 In other words: poetry and prose are not common; they lie outside the linguistic phenomena that ordinarily present themselves. The notion of rarity, which we will encounter again at the end of this essay, goes to the heart of a probabilistic, quasi-Bayesian vision of literature. Taken to its limit, in this new regime, one could dispense with the notion of deviation altogether, which only makes sense in relation to that of “norm” — something not easy to define. We would no longer concern ourselves with determining whether a piece of writing departs from the norm. We would simply say: this writing is improbable. And the more improbable it is, the rarer it is — that is mathematical. Strictly speaking, today’s improbable could become tomorrow’s norm, since a rare form is liable to be duplicated and to pass in a flash into the domain of the probable. Literature as improbability would fit squarely within the hyper-fast temporality of the age of intelligence. In place of norms fixed across time, we would see the emergence of a shifting regime, constantly redefined at the mercy of algorithms.

  1. The Red and the Black (1830). In the original: « Un roman est un miroir qui se promène sur une grande route. » (Le Rouge et le Noir, Paris, Gallimard, coll. « Folio Classique », [1830] 1972, p. 557.) ↩

  2. Vers une littérature de l’épuisement, Paris, José Corti, 1991; Poétiques de la voix, Paris, José Corti, 1999. ↩

  3. Pierre Michon, The Eleven (Les Onze), trans. Jody Gladding and Elizabeth Deshays, New York, Archipelago Books, 2013. In the original: « Puisque vous m’en priez, Monsieur, je veux bien que nous restions un instant encore dans le grand escalier. » (Les Onze, Lagrasse, Verdier, 2011, p. 18.) ↩

  4. Martin Heidegger, On the Way to Language, trans. Peter D. Hertz, New York, Harper & Row, 1971. The passage also appears in Poetry, Language, Thought, trans. Albert Hofstadter, New York, Harper & Row, 1971. ↩