Chapter 12

All the Way to the Singularity

10 min read

12. ALL THE WAY TO THE SINGULARITY

The horizon I have just sketched remains speculative, I grant that. Let us go a little further into speculation. Let us explore, theoretically, the world beyond the transition.

There exists a body of forward-looking theories about the state of the world once we have crossed the pass of artificial general intelligence (AGI) — which would match human experts in every domain — or even artificial superintelligence (ASI) — which would surpass all humans in every domain. The leading artificial intelligence companies have the avowed goal of developing AGI or ASI, and a majority of experts believe we will reach artificial general intelligence before the end of the decade — roughly five years from now. It is not impossible that this will take 10 or 20 years instead, but such long-range predictions remain in the minority.

In Planned Human Obsolescence, I laid out my view of the economic and existential risks posed by an AI endowed with intelligence equal or superior to that of human beings. The venture may end in catastrophe: either the formation of a powerful and immovable oligarchy, or, in the worst case, the extinction of the human species.

I take these scenarios very seriously, yet I still believe it is possible that humanity will wake up in time and erect safety barriers around advanced AIs. These guardrails could take different forms: restrictions on agentic AIs (agents that make decisions and carry out actions in the world); oversight of agentic AIs by a scientific AI (as Yoshua Bengio proposes with his Law Zero project); an international treaty defining computational thresholds to strictly regulate AI progress; the strengthening of democratic institutions and an overhaul of the monetary system to distribute the wealth created by AI; technical and political decentralization mechanisms1… If we were to halt all progress in artificial intelligence this February 2026, freezing models at their current level of development, a century would not suffice to exploit all the possibilities opened up by these tools, and literature would already be completely transformed. I cannot imagine a world in which things — writing, the book, reading — would simply carry on without undergoing profound changes. A radically altered world, on the other hand, seems to me not only possible but highly likely. As the leading companies automate programming work, and even AI research itself, the recursive self-improvement loop is no longer a mere science-fiction theory; it is a project already underway.

In the field of artificial intelligence, this tipping point has been given a name borrowed from black-hole physics: the “singularity.” Within the time that would open with the explosion of intelligence, things would grow strange; ultra-rapid changes would exceed human understanding. Apart from small groups in the AI world, a handful of marginal academics working in “Post-AGI Studies,” and a few curious minds (myself included), very few people are contemplating the possibility of living in a completely transformed world. We humans learn from experience; we know how to read continuities; but we are ill-equipped to apprehend ruptures. If we manage to create entities as intelligent as or more intelligent than ourselves, is continuity even conceivable? In my view, if the event horizon does not kill us, it will redraw the world from top to bottom.

Let us imagine this world in which AI surpasses the human being in every domain, including literature. Yes, I am speaking of an AI capable of producing texts more brilliant than those written by Marcel Proust or Virginia Woolf. I will be told that Proust’s work cannot be reduced to cognition or to brain activity understood as computation. “Every day I set less store by the intellect,” he wrote in the opening of Against Sainte-Beuve.2 I am not saying that superintelligence will undergo the existential and sensory experience that underlies the writing of In Search of Lost Time, nor even that it will want to write such a work, sprung from a distinctly human intuitive memory. What I am saying is that if it desired to, or if it could be commanded to, superintelligence could write a book more accomplished still than Proust’s, including by plucking the strings of human emotion — even though it is likely that during the seconds or minutes in which the AI produced its masterpiece, it would remain perfectly cold. One may well say that art exceeds the intellect, which is true. But if a superior intelligence has a goal — including the goal of emotional effect — it will optimize every technical means at its disposal to achieve it, and we will be none the wiser.

I do not know whether we will reach the kind of utopia that Nick Bostrom envisions in Deep Utopia. In such a “solved world,” to borrow the book’s subtitle,3 the intermediate steps between the desire for a text and the pleasure of a text would vanish entirely. A wish expressed — or even a desire merely felt, should a brain-machine interface or AI hyper-perceptivity exist — would be enough for that desire to become reality in a voice (textual, sonic, or otherwise) that would fulfill us to the highest degree. Texts would become extremely personalized: the AI would write a narrative or a poem for me, drawing not only on my literary desire but on a mass of data — my mood, my sleep, my health, my dreams, my preferences, my fantasies, the weather, the day’s news, the time I have ahead of me, and so on. In this world, there would no longer be any author or reader; the distinction would collapse. We would generate and absorb the text in a single movement. Or perhaps the AI would anticipate our desire: we would open a door and the text would be there, the voice would resonate with uncanny precision.

Then we would no longer be in the quest. We would no longer be in the request. We would be at the feast after the adventure: we would be in the banquet.

It should be clear by now: I am not convinced by this vision of a glorious tomorrow. But I do believe there are two notions — explored by Bostrom, but also by other thinkers — that help us approach the singularity: abundance and scarcity.

One can assume that everything will hinge on these notions without having to subscribe to the utopian scenario. If artificial intelligence can produce high-quality texts in seconds, or even fractions of a second, what effect will this have on the economy of literature? Prose and poetry possess value because they are scarce. Writing is not easy; many break their teeth on it. In the human world before the singularity (our current world), voice is not an abundant resource, and neither are the texts that flow from it. We take an interest in texts, we weigh them, compare them, analyze them, hold forth about them because we are searching for the needle in the haystack — the voice amid the chatter. At this game, the present often errs, for what is rare can seem strange, inaccessible, or failed. The hindsight of history, as we know, often corrects the myopia of the present.

One may get a false impression of abundance by observing the media deluge of the French rentrée littéraire. This is to forget that most of these publications are conventional novels written in middling prose. But even if every one of these books were spun from gold, we would still be far from the kind of abundance that artificial intelligence portends.

AI will be able to produce prose that is gold for the price of electricity.4 On this market, the human will no longer be competitive. Even with a voice, a human writer will not be able to adapt it for the reader the way AI does; will not be able to compile millions of data points to sing in perfect pitch to the ear of the listener or the reader.

It is perhaps precisely because AI will be too perfect that the human may retain a niche of their own. Do we really want a world where we are spoon-fed texts designed to please and enchant us? With hyper-personalization, reading could become a form of onanism. What would be missing is the otherness that defines the reading relationship today: we read a voice that is not our own, that was not made for us, that presents sharp edges that may shock or even wound us.

And then, human voices remain imperfect. Here again, it is a bad habit of traditional publishers to want to smooth the text until the imperfect event of its emergence has been erased. This is what the first editors of On the Road did: cutting the flow into paragraphs, using the conventions of the novel to name the “characters,” correcting errors, erasing repetitions, and so on. I prefer to unroll the original scroll and see the first sentence open on an imperfection — namely, the repetition of the verb “to meet”: “I first met met Neal not long after my father died…”5

Is this to say that human authors or human master builders still have bright days ahead of them? Nothing is certain in the singularity. The problem with AI — and especially with the prospect of artificial general intelligence or superintelligence — is that every time we think we have discovered some corner of human activity spared by the machine, we end up realizing we were wrong. AI will be able to colonize the domain of the imperfect as well. It will know how to reproduce the roughness we love in texts. Even otherness it will know how to recreate: all it will need to do is write a text by training all its data sensors on another source, on a register of experience different from the reader’s. It is even possible that AI will prove better than humans at the game of roughness and otherness. AIs are champions of optimization; all they will need to do is maximize the rough edges.

Nothing indicates that in literary matters, humans could still produce scarcity, even with their imperfection. But there may be one last refuge. The vocal imprints I discussed in this essay will have no value on the voice market, for they will be drowned in abundance; but we could give them a value in themselves. This would mean inscribing them on a blockchain (or on some future technology we cannot yet foresee): there, they could possess an economic scarcity. I should specify that this would be an entirely manufactured scarcity, since out there, on the text market, artificial voices — diverse, magnificent, satisfying — would remain omnipresent. I am speaking here of a blockchain reserved for human voices. It is an idea — a very theoretical one, I admit — loosely inspired by what Emad Mostaque proposes in the monetary domain with his currency he calls the “Culture Credit,” a quota allocated exclusively to human beings.6 In my vision, human vocal imprints would be artificially separated from the ocean of voices, and their value would rest on a humanity and a literariness verified at the moment of inscription on the blockchain.

In this way, scarcity would be preserved within a precise architecture. Humans and AI agents might wish to own a small fraction of an imprint for its collectible value and its investment value, as art lovers do. If there were millions, even billions of preserved human voices on the blockchain (including historical voices), this would still represent an extreme scarcity given the abundance of synthetic voices in the wider world.

  1. A solution advocated notably by Vitalik Buterin, the founder of Ethereum, but criticized by others, since it amounts to putting systems with enormous destructive potential into the hands of anyone and everyone. To counter this argument, some thinkers advance the theory of a balance of forces. In my view, decentralization would have positive economic effects (better distribution of resources) but would probably carry significant security risks. ↩

  2. Marcel Proust, Against Sainte-Beuve, translated by John Sturrock, London, Penguin, 1994, p. 3. In the original: « Chaque jour j’attache moins de prix à l’intelligence. » Marcel Proust, Contre Sainte-Beuve, Paris, Gallimard, 1954, p. 55. ↩

  3. Nick Bostrom, Deep Utopia: Life and Meaning in a Solved World, op. cit. ↩

  4. Since 2023, the price of intelligence has fallen from $37.50 to roughly $0.14 per million tokens (for performance equivalent to GPT-4), a 99.7% drop that shows no signs of stopping. Many foresee a future in which the price of intelligence approaches zero. ↩

  5. Jack Kerouac, On the Road: The Original Scroll, New York, Viking, 2007, digital edition, n.p. ↩

  6. The Last Economy: A Guide to the Age of Intelligent Economics, Intelligent Internet, 2025. ↩