Entry Point#4: expression, thought and singularity

Reaction to Gaspard Koenig’s interview with François Lenglet (podcast, May 2026), relayed by Benoît Raphaël on 7 October 2026

François Lenglet hosted Gaspard Koenig on a podcast (May 2026). I had not seen it, and I discovered this interview with Koenig through a post by Benoît Raphaël on 7 October 2026.

Asked about the use of AI, Koenig’s answer stunned me. I reproduce the transcript (from 26:18, my translation):

I think that when you do that, you gradually give up your capacity to decide, to choose, to think, and who do you give it up to? You give it up to the mass. Because AI only ever gives you the average argument, really, the one it finds in all the data it is busy eating.

So it is something that, by definition, takes away your singularity as a mind, a thinking mind, and I find it an affront to my dignity, that is to say, not even on a practical level or in terms of efficiency, but on a moral level. It is not me appropriating this thought, because it is not my mind that arrived at this reasoning, so it is worth nothing, I might as well not speak myself.

These remarks belong to a moment (May 2026) when Gaspard Koenig had already spoken out on AI. I had commented on that intervention in my text Dancing with the Machine, or Staying on the Drip. Although the passage above is short, I find in it some of the markers of a certain critical discourse on AI.

Faced with this new “critical” formulation, I think it deserves better than a technophile and/or outraged response. The difficulty I see in this kind of critique is that it actually rests on a series of hypotheses which, in themselves, and provided they are treated as such, are perfectly admissible. However, the author who uses them treats these hypotheses as one-dimensional and indisputable truths. The critique then becomes ideological and political.

That does not make it any less interesting, notably because it raises a philosophical and, above all, very topical question in the age of generative devices: what makes a thought truly ours? That question leads to another, more universal one: what constitutes the act of thinking? But I will not address it here.

I will therefore try to offer a different reading of the question, avoiding the pitfalls of the French philosopher who dispenses the (usual) clichés and some debatable presuppositions: equating the value of a proposition with its origin, or even its appropriation; confusing statistical generation with the production of a thought; and assuming, above all, that a singular thought should be built autonomously, or at least cut off from certain outside influences.

An idea and its expression

In the conversation, François Lenglet explains that he uses generative devices to express a thesis he has (pre)formed. Koenig’s answer reproduced above comes in response to this experience and in this context.

First, I think a confusion is made between the idea any individual may have and the shaping of that idea. It is in the expression of an idea, that is, in the form that allows it to be shared, that singularity can be distinguished. An idea is, by essence, immaterial, unexpressed. It is in our head as a thought, and others can only examine it once it has been given form on a medium or through language. This revelation is only possible through expression (unless we imagine mechanisms of brain extraction, but that strays from the subject). Only by examining this expressed thought can we observe its singularity and that of its author.

Next, and with that written, I must immediately add that an idea is always inspired and/or influenced in its foundation and in its contours. There is no idea or thought ex nihilo. For every idea there is a context that made it germinate in the mind. An encounter, a memory, a reading, and all the events of life contribute, to varying degrees, to drawing the contours of that thought.

When I ask a generative device to produce a text from an idea, I can give it a very broad frame (and therefore refrain from giving details). I will then receive something close to the average. Conversely, I can also define the frame more narrowly: specify the thesis I want to defend, the authors I want to bring in, the objections I refuse to dodge, the tone I want to adopt or the arguments to address. The more precise the frame, the further the result moves from the mass and the closer it comes to what I had in mind.

The average argument, the argument of the mass, exists, but it is the output of a generative device left to itself.

Singularity is not one-dimensional

Past these first pitfalls, we can turn to the notion of singularity itself.

It is worth noting that the word is also used by technophiles and AI enthusiasts, such as Ray Kurzweil, to describe a runaway technological growth that would bring unpredictable changes to human society. Kurzweil predicted human-level AI around 2029 and a singularity in 2045 (see Ray Kurzweil, The Singularity Is Near).

In Koenig’s conception, singularity means what is singular, what is unique or set apart, and he seems to place this notion in one place only: the individual.

As sketched above, singularity can also be found in the idea or in its expression by the individual. By narrowing the scope of singularity, we deny it existence on other planes and in other areas. This singularity is not univocal for anyone. Each of us can express our singularity in a thousand and one ways, and that too is what makes us singular.

Lenglet describes precisely a case where he has had several ideas and asks the machine to express them. He thus uses his own singularity upstream of the “computational” generation. One can imagine him sharing scattered notes, a list or an outline so that the generative device rephrases, completes, proposes and finally shapes a text.

Moreover, the user’s singularity can also be expressed downstream of the production. In the interaction with the device and in keeping (or not) the generated results, the user can rephrase, complete, modify or delete passages of the generated text. In this movement, singularity is expressed once more.

Writing and thinking

Where I can nonetheless agree with Gaspard Koenig is that the contours of an idea can take shape through writing. I think of this quote by Leslie Lamport:

If you’re thinking without writing, you only think you’re thinking.

The writing process creates a friction that allows us to think about our own thinking. For many, writing is the process through which we discover what we do not know about what we are writing. It is by writing that we realise our ideas and our reflection are incomplete. Writing opens a space for reasoning and reflection that is not always present or available when using a generative device.

That said, discrediting every other form of expressing ideas seems to me a step not to be taken, and it is an aspect addressed in Dancing with the Machine, or Staying on the Drip:

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.

Koenig’s position is therefore based on a single mode of use. It is because of this simplistic view of use that the conclusion he draws ends in a position that is dogmatic and reductive, but above all open to challenge.

The weight of words

In his post, Benoît Raphaël considered that one should not dwell too much on “the class contempt that oozes from Koenig’s remark”.

On the contrary, I think it is necessary to respond, because this vocabulary is jarring in its arrogance and disdain. Koenig speaks of his own dignity. But by using this morally charged term, does he not cast AI users as the “undignified”? If using the machine to formulate an idea undermines the dignity of the person who does so, then those who use it fall, by extension, into a less dignified category than those who abstain. Should we see in this a new form of purity policing?

As I wrote here:

What makes a text thought rather than simply produced?

This question has no technical answer. It cannot be resolved by external markers (use or non-use of AI), by moral criteria (effort deployed), or by declarations of intent (sincerity of the author). It can only be resolved by a judgment on the text itself. A thought text is recognized by its depth but also by its rough edges, by its assumed contradictions. A merely produced text is recognized by its polished vacuity, by its homogeneity.

I persist in believing that the quality of content generated with GenAI can only be guaranteed by human investment in direction, structuring, and the mobilization of concepts.

When this attention is lacking, the quality of the text diminishes. The text becomes generic, hollow, recognizable in its characteristic GenAI vacuity.

Koenig’s hypothesis, used as a one-dimensional truth, is not only restrictive but also puts bounds on reflection on the subject. This lack of nuance creates neither relevance nor accuracy. It is a reductionism instrumentalised to serve an ideological thesis through a social status, and it bothers me deeply when this instrumentalisation comes from a supposed paragon of critical and considered thought.

Koenig has every right to be severe with AI. We have, however, too little hindsight on these devices. They evolve very (too) fast and their effects remain poorly understood by and for many, owing for instance to Anders’s Promethean gap. We are all, today, in a phase of exploration and learning with these devices. The fact that technological progress in the sector is breathtaking does not mean its integration into the real world is moving at the same pace.

I always find it a little comical that those who defend critical thinking, singularity or nuance dispense with them as soon as these techniques are discussed. It therefore seems important to me to tell things apart: fantasies and other extrapolations on one side, dogmatic condemnations and disqualifications on the other.

For a pharmacological approach

What seems more or less certain, however, is that there will be no going back. In this context, a dogmatic and technophobic rejection, without nuance, seems to me a completely sterile approach.

To borrow Yann Ferguson’s words, what must be considered first and foremost is how this adoption can become collective, controlled and compatible with everyone’s own requirements.

And none of this is specific to artificial intelligence, because every technique is a pharmakon. The consequence is an ambivalence inherent in every technique. This ambivalence ultimately requires us to break free by seeking nuanced solutions, which are necessarily more complex than a disdainful rejection or a blind embrace.

In this context, I can only reproduce Christian Fauré’s explanation of pharmacology. According to Fauré (my translation):

Pharmacology refers to the Greek pharmakon, which means poison as much as remedy; “it depends”, as the saying goes. Now, it is this “it depends” that pharmacological approaches must make explicit each time, by agreeing to grant a particular status to this essential ambivalence of the pharmakon; a status that will lead us to think by compositions rather than by oppositions; a status that also allows us to dismiss both technophiles and technophobes, who share a dual and binary vision of things that forces them to think by opposition. (Réseaux sociaux : Culture politique et ingénierie des réseaux sociaux, p. 108)

The use of large language models is frequently considered from the angle of delegation: the user would keep control of their intellectual activity while outsourcing certain operations of research, formulation, production, or even, for some, creation and reflection to the machine. Koenig goes further when he considers that the act of thinking is delegated too.

I think this representation is inaccurate and insufficient. One of its weaknesses is that it assumes an already constituted subject, using a device external to it (exosomatic) and remaining fundamentally unchanged by its use. Yet the experience of generative devices shows that the interaction cannot be reduced to a “mere” instrumental relationship. The machine’s proposals affect the user’s reasoning, the user reacts to them, these reactions change the context of the exchange and in turn shape the subsequent generations.

This is where the critique becomes interesting. Technique has always taken part in the constitution of human knowledge and memory. It is very difficult to reject it or to do without it. It has always been there, and we can agree that going back seems impossible. With that observation made, we must then think.

This starting point is fruitful, because outsourcing certain intellectual operations is nothing new, but here these operations take a direct part in shaping our know-how of thinking. I have already written about the effects of this practice, which can lead to a dispossession.

Ultimately, what is at stake here is a change of practice. Choosing between thinking alone and surrendering one’s thought to a machine is a binary alternative that many propose and that will inevitably lead to a dead end. We must move beyond this way of doing things and allow ourselves to think by compositions rather than by oppositions.

These notes are part of Duality, my digital notebook. If you would like to receive them directly in your mailbox, you can subscribe to the newsletter.