Dancing with the Machine, or Staying on the Drip
I read Gaspard Koenig’s column on Abel Quentin’s Sanctuaires, then the response Benoît Bergeret wrote to it. I would like to comment on that response, but before I do I need to write a few words about Koenig’s text itself. And to do so, I have to take a detour.
The Engine of Philosophy is the Dispute
In the March 2026 issue of Philosophie Magazine, Nicolas Tenaillon analyses three great philosophical quarrels to show how thinkers sharpen their arguments through confrontation. The detour through Philosophie Magazine is also a wink: Abel Quentin and Gaspard Koenig are interviewed in it. Should we read something into that coincidence? I don’t (yet) really know.
In any case, Tenaillon mentions in particular the quarrel that, at the end of the seventeenth century, set Charles Perrault against Nicolas Boileau: the quarrel of the Ancients and the Moderns. The object of the debate? Imitation. Should one continue to imitate the Ancients because they are unsurpassable models, or break free of them in order better to celebrate the sublime under the reign of Louis XIV? Such was the question.
If you have read Bergeret’s text, you may already see where I am going.
The starting point of Koenig’s column is that the writer is supposedly able to think alone, autonomously, without any influence. For my part, I am one of those who consider that we are all “dwarfs sitting on the shoulders of giants,” and that any form of reflection is influenced upstream. This very reflection is yet another illustration.
Bergeret puts it differently, drawing on biology: we are holobionts, organisms that only live with their symbionts (a lasting and intimate association between two different species, where the larger organism is called the host and the partner is the symbiont), and human cognition works in the same way. We never think alone. We think with the books we have read, the traditions we have inherited, the culture we live in, the tools we handle.
The Moderns Facing the Ancients
The episode Tenaillon recounts is interesting for another reason. Boileau attacks Perrault, writing that “a careless author always thinks he writes better than Virgil.” An escalation takes place between the two authors. And Tenaillon specifies: posterity sided with the Moderns. It is not the poets Perrault defended, nor their Christian epics, that literary history has retained. It is Boileau, La Fontaine, Racine, La Bruyère.
Tenaillon quotes, in this connection, a sentence by Racine that I find telling: those who cannot reach the Ancients must surely decry them.
That sentence has resonated with me ever since I read it. Because it echoes, in reverse, a movement I observe today in the debate around generative AI. According to Racine, it is the newcomers who decry the Ancients because they cannot match them: the disqualification comes from below, from those who do not yet have authority. Today, the movement is reversed. It is the established authors, those who have mastered the verb and won their place, who decry generative AI and the texts it allows to be produced. The procedure is no longer that of the aspirant who disparages the master, but that of the master who disparages the tool by which aspirants might now equal or compete with him. The rhetorical structure is flipped, the defensive mechanism remains the same: disqualify what one does not practise in order to preserve what one is.
Bergeret spots this too when he writes that Koenig is defending less the book than a certain economy of the book, a cognitive monopoly, that of the virtuosos of linear verb, whose rent is wavering. Philippe Silberzahn speaks, for his part, of misplaced purism. The formula recurs in many authors who see arriving on the market people now able, thanks to the machine, to produce texts that could compete with theirs. The Moderns, this time, are not those who claim to surpass the Ancients at their own game. They are those who change the game, who find in the tool an access to a kind of writing that was closed to them. This shift makes the disqualification comfortable: one does not say that these new entrants write badly, one says that they cannot write, because what they produce would not really be writing. The boundary is drawn not on the terrain of quality but on that of the legitimacy of the procedure.
There is also something to be said about the mechanics underlying this defence. The wink I noted at the outset, regarding the March 2026 issue of Philosophie Magazine which interviews Koenig and quotes Quentin, is not innocent. Koenig has published his works at Éditions de l’Observatoire and directs there the essay collection “De Facto.” Sanctuaires, Quentin’s novel which Koenig endorses in his column, was published by the same house. When Koenig writes to defend the book against AI, he also defends, whether he wants to or not, a circle of authors and a distribution channel of which he is part. This does not disqualify his argument, but it clarifies its position. The defence of the sanctuary looks like a defence of the inner circle, where people knight one another, where one quotes the house’s author in one’s column, where one builds together the legitimacy of a genre that the machine would come to destabilise. It is a known mechanism: faced with new competition, one closes ranks before reopening the debate. Note that Quentin’s book came out at the beginning of May 2026… a chronology also worth taking into account (and, in the interest of transparency, Quentin’s book is in my pile of books to read in the coming weeks).
That said, Racine’s sentence sheds light on Koenig but it could also shed light on Bergeret, and on me as I write these lines. Bergeret, by placing Koenig on the side of conservatism and himself on the side of becoming, may be doing what he reproaches Koenig for: drawing a boundary between those who would have understood and those who would not. There is always someone one cannot reach, whom one has an interest in decrying. The spatial thinker liberated by AI of whom Bergeret speaks also has a rent to defend, or at any rate a new status to win against the virtuosos of the verb. The Koenig/Bergeret dispute is not a dispute between someone defending his interests and someone defending the truth. It is a dispute between two positions, each of which has its interests and its way of mobilising philosophy to make them presentable.
This disqualifies neither of them. But it forces — and forces me, as I comment on them — not to settle too comfortably in one camp while believing I escape the very logic denounced in the other.
The Purism of the Non-User
This is where I get stuck. I had already had occasion to express surprise at certain remarks on generative AI written by people proudly claiming not to have taken the time to use it. One should not criticise a musical instrument without having played it for a few hours. One should not criticise a language without having read a few of its texts. But generative AI seems to enjoy a particular status: that of an object on which one is allowed to pronounce judgement without prior practice. The problem is that “unpractised” criticism is bound to miss its target. I think Koenig’s critique is not aimed at AI as such but at one oriented way of using it.
On this precise point, Abel Quentin’s analysis of dispute (in the aforementioned issue of Philosophie Magazine) is interesting. Quentin, who claims to be a disciple of Jacques Ellul, sees in generative AI a pure and simple elimination of otherness. And Quentin concludes: “Being permanently on the drip of a digital courtier really does not help us learn to argue with one another. Real conflicts then become all the more violent.”
The image is apt. But it describes one use, not the tool. Refusing that use is not enough to condemn the thing.
Dancing, Rather Than Letting Oneself Be Drip-Fed
The metaphor Bergeret borrows from Nietzsche interests me. In Thus Spoke Zarathustra, Nietzsche opposes dance to the spirit of gravity, that solemn and serious demon by which all things fall. Dance is not the avoidance of gravity, it is its outwitting.
Bergeret transposes: AI exerts a force that pulls toward easiness, that draws in the gesture of writing, that pushes toward instinctive delegation. If we let it, it smooths voices, invents facts, steers ideas. But to dance with it is to decide when to use it and when to do without it, to ask it to disturb an idea rather than produce one in our place, to keep one’s hand on what must remain one’s own.
This is exactly the opposite of Quentin’s digital courtier.
And it is this third way that Koenig does not envisage. Between the drip (which he rightly denounces) and the sanctuary (which he proposes), there is reflective practice: the kind that tames the tool, tests its limits, decides at each step what can be delegated and what cannot. This practice requires apprenticeship like any technique, in Ellul’s sense — that is, a method employed to reach a result.
That said, one must take the measure of a problem this reasoning sidesteps a little too easily. Is the boundary between dance and the drip as sharp as I have just suggested? I am not sure. When I write in dialogue with an AI and decide, at some point, to keep a formulation it offers me, I can tell myself that I am dancing: I sort, I choose, I keep the upper hand. But how do I know whether the formulation I have just adopted did not slip into me a few exchanges earlier, at a moment when I was not sorting, and whether it is not now resurfacing as my own intuition while in fact being its product? The idea I think I have disturbed, was it really disturbed by me, or have I simply rephrased an orientation it had already imprinted in the thread?
This is the whole problem of [[counterfactual reasoning]]. To know whether I am really dancing, I would need to be able to compare what I have written with what I would have written without it. But that alternative text does not exist. I cannot run two versions of myself in parallel, one assisted and the other not, to measure the gap. The thought that emerges from the exchange presents itself as a totality, and I have no access to the decomposition that would tell me what part comes from me, what part from it, what part was born of the friction between the two. The very gesture by which I persuade myself I have kept the upper hand is itself affected by the machine I believe I am mastering.
This means that dance, as Bergeret describes it and as I claim it, is perhaps less an objectifiable practice than a story I tell myself in order to distinguish myself from the courtier on the drip. The person being courted does not know himself as courted. He believes he thinks, he believes he decides, he believes he chooses. The most effective drip is precisely the one that does not feel like a drip. And if I cannot demonstrate the difference between my dance and another’s drip, perhaps I am already on the drip without knowing it, telling myself I dance to reassure myself.
Vertiginous? Perhaps, but it does not invalidate a certain practice. It does, in any case, oblige one to more humility than the Nietzschean metaphor suggests. To dance is not a posture one adopts once and for all: it is a continuous effort, never fully successful, of which one cannot know with certainty whether it succeeds. At best, one can take care of the acts that make dance more likely than the drip: take time without the machine, write things one does not show it, accept producing less polished texts in order to check one is still capable of it, read authors who think otherwise, keep a journal of one’s own formulations before the machine reformulates them. None of these gestures proves one is dancing. All of them, together, increase the chances that one is not merely being drip-fed.
A Relation of Individuation
What particularly interests me in Bergeret’s reading is that he describes the relation to AI in terms of duration and reciprocal transformation. The relation settles in over time, and each transforms the other, the human in their relationship to text and idea, the machine in what it learns from this particular human.
This description echoes what Gilbert Simondon called individuation: a process by which a being does not pre-exist its relation but constitutes itself through it. The individual is not given in advance, it comes about by relating to a milieu, to tools, to other beings. Applied to assisted writing, this means that the writer who practises AI over time is not the same before and after, and that what comes out of the exchange is neither the idea they had in mind to begin with, nor the one the machine would have produced alone, nor even the one it would have written for another user.
This is what the courtier metaphor cannot capture: the courtier flatters, he does not transform. The symbiotic relation, on the other hand, transforms. It is another form of thought.
For a Practised Critique
I do not know, at this stage, whether posterity will prove Koenig wrong as it proved Perrault wrong. Nor do I wish it: there is something to protect in what he is defending, and Bergeret recognises this too when he proposes his functional sanctuaries, those protected moments of judgement formation rather than places forbidden by decree.
What remains is the necessity of a critique that goes through practice. One cannot think seriously about generative AI without having inhabited it. One cannot reject it wholesale without rejecting along with it the possibility of learning to dance with it — that is, of keeping one’s hand on what must remain one’s own while accepting that what comes out of the exchange is neither entirely us nor entirely it. The dispute between Koenig and Bergeret is, in that respect, a fine dispute. It has the merit of sharpening arguments. Tenaillon was right at least on this point: the engine of philosophy remains the dispute.
I dance. Not because I am sure I am right against those who dig in (to borrow Bergeret’s expression) but because I am sure I do not want to dig in. It is less a certainty than a stance, which may have to be revised in due course. In the meantime, I prefer to learn these new dance steps.