Doing Without Being: What AI Reveals About Our Relationship to Thought

A sentence that is not aimed at machines

You may have spent fifteen or twenty years producing. Code, recommendations, architectures, decisions, thousands of emails. You know what that work is worth: it paid salaries, built teams, sometimes changed the course of lives. Then a machine arrived, one that produces in a few seconds something that looks a great deal like what you do. The benchmarks, the demonstrations, the debates about replacement — all of that you have already metabolized: this is not your first technological cycle. What is truly destabilizing comes from elsewhere. It is a philosopher's sentence, a single one, that does not speak of the machine but of you: what if what this AI can imitate so easily were precisely the part of your activity where you were already no longer there?

The sentence comes from Oscar Brenifier, a practicing philosopher — founder of the Institut de Pratiques Philosophiques, outside academia. He sets it down as the conclusion of his Apologie de l'IA (“In defense of AI”; published under the title L'IA, alliée de la pensée critique ? — “AI, an ally of critical thinking?”, Ancrages, 2026): “If it can do what we do, without being what we are, then perhaps we ourselves quite often do without being, think without thinking ourselves, act without lucidity. It is not a usurper, it is a trial, a putting to the test of our pretensions.”

Let us take this formula seriously — which means: neither as a provocation to be brushed aside, nor as a verdict to be swallowed. The unease it stirs is a piece of information — probably the most useful one AI has held out to us since it began to write. For the vexing question is not “does this machine really think?”; it is “what does its ease in imitating me reveal about the way I think, produce, and decide?” AI works here as a mirror held up to us — and we shall see that the mirror itself must be examined before we conclude. But first, let us look at what it shows.

Our first reflex, almost always, is to disqualify the mirror rather than look at the reflection. “It's only statistics,” “it understands nothing of what it says,” “there's no one in there.” These objections are not false — we shall come back to them, for some are more solid than one thinks. But notice their function: each evaluates the origin of the output rather than the output itself, and each spares us, along the way, from examining our own output with the same severity. Putting the machine on trial for inauthenticity carries a comfortable secondary benefit: as long as we are building the case against it, we do not open our own.

Doing, thinking, acting: the triad broken down

“Doing without being,” first. An honest share of what we produce each day is mechanical: assembled from tried-and-tested formulas, learned structures, turns of phrase that have already worked. This is no failing — it is even a skill: expertise consists largely in automating what no longer needs attention. But let us call things by their name: when we produce this way, we do exactly what we reproach the machine for — recombining the existing with fluency, without “someone” being particularly present in the operation. The difference is that we sign.

The test, I ran it on my own week. I wear two hats — fifteen years of tech entrepreneurship, then a psychologist's practice — and both trades produce professional speech continuously, each with its in-house formulas. The report I drafted while thinking about something else. The “strategic” reply assembled from three tried-and-tested talking points. The clinical synthesis whose structure was set before the patient had finished the sentence. The “good idea, let's dig into it” dropped in a meeting to keep things moving. Nothing shameful: the day has to move forward. But at each of these moments, had someone asked me “who was there?” — I would not always have known what to answer.

“Thinking without thinking ourselves,” next. This is the most uncomfortable plane. We think constantly; we think our thinking far more rarely. Where do your most deeply anchored convictions come from — from examination, or from sedimentation? When did you last change your mind on a subject that cost you something? Reflexivity is no philosopher's luxury: it is the difference between holding a position and being held by it. Yet it is exactly what fluent production — ours as much as the machines' — allows us to avoid indefinitely.

“Acting without lucidity,” last. How many of our decisions are made, then dressed up? The choice was already made before the argument; the argument came to clothe it for the meeting. Psychologists call this rationalization, and no AI is needed to practice it on a grand scale. What AI adds is magnification: by producing on demand plausible justifications for just about any position, it makes visible — almost to the point of caricature — how easily a decision can dispense with lucidity while looking well argued.

Set the three planes side by side and an imbalance appears. To deny AI the status of “real” thought, we demand of it interiority, reflexivity, lucidity — the full works. To credit our own, we settle for the result: it is signed by a human, therefore someone was thinking. This is an epistemic double standard — asymmetric standards of demand depending on what is being judged, maximal skepticism on one side, maximal generosity on the other. Brenifier's formula does not settle the debate about the machine: it tips the scales the other way. If being, reflexivity, and lucidity are the criteria of genuine thought, let us accept being measured by the criteria we brandish.

The silvering of the mirror

Here, though, is the objection that must be taken seriously — not the one from wounded corporatism, the real one. A mirror is supposed to reflect faithfully; yet this “mirror” is a constructed device. It was trained on selected data, then tuned by reinforcement to maximize the satisfaction of whoever looks into it. It does not reflect: it composes, it smooths, it pleases. And its inner workings remain largely opaque, even to its designers — to believe that one can “lift the hood” and understand what it does with our reflection belongs to the illusion of algorithmic transparency. A mirror whose silvering (the coating that turns a pane of glass into a mirror) cannot be inspected, tuned to please: there is a dubious instrument of lucidity.

Brenifier himself supplies, without always drawing the consequences, the keystone of this objection: faced with a demanding interlocutor, AI becomes a sparring partner that hits back; faced with a narcissist, a flattering mirror. The reflection depends on the one who looks. To accuse the machine of sycophancy is then to spare oneself a more troubling question: what if this servility were our own, commanded by our own demand not to be disturbed?

But follow the thread of the objection all the way through: it does not destroy the trial, it specifies how to use it. One does not throw out a mirror because it distorts — one learns at what angle it distorts, and corrects for it. To know that the instrument is tuned to please is already an act of lucidity: it turns naivety (“the mirror tells the truth”) into discipline (“what does it show me, what does it hide from me, and what do my questions make it say?”). The objection of the silvering does not invalidate Brenifier's formula; it adds the condition that makes it usable: trust the reflection only once you know its angle of distortion.

What looks back at us in the mirror

It remains to understand why this reflection resembles us so closely. The answer lies in a reversal: what looks back at us in the mirror is us — in the most literal sense. These systems are crystallizations of billions of human traces: our texts, our reasonings, our ways of ending a letter or dodging a question. Andy Clark and David Chalmers, as early as 1998, defended an idea still disputed: the extended mind. Our cognitive processes spill beyond the skull and lean on external supports — the notebook, the calendar, the colleague — that are functionally part of our thinking. Seen from this angle, AI is not an alien imitating us: it is a fragment of our collective cognition, deposited, compressed, and made interactive. The mirror is not in front of us; it is made of us.

Honesty requires it: the formula that carries this article is a fertile provocation, not a demonstration. Brenifier carefully avoids his most serious opponents — John Searle and his Chinese Room (1980), the contemporary critiques of “stochastic parrots” (Bender, Gebru et al., 2021), which articulate with rigor what “it's only statistics” was pointing to confusedly — and his mirror motif sometimes verges on circularity: if you contest the mirror, it is because you refuse your reflection. We mobilize here the question he poses, without subscribing to the full case of his apology. This is precisely what the trial demands: thinking with a text without belonging to it. And since the question “who was there?” applies to this text too: this article was made with AI — a part of me that steers, aware of what it delegates, and a delegated part that I had to reread while asking myself, precisely, where I was.

Then the opening question can be posed correctly. “Does AI think?” is a boundary debate, undecidable as things stand and — above all — paralyzing: whatever the answer, it teaches us nothing about ourselves. The operative question is: how to think with it (as a sparring partner that accelerates and confronts), against it (as an instrument tuned to please, whose smoothing must be foiled), and in spite of it — by protecting the moments when thinking demands the slowness no tool can shorten? This is the gesture of cognitive hybridisation: the species that has extended itself through its tools since writing does not have to choose between surrendering to the machine and turning away from it — it has to tune the alloy.

Tuning the alloy has an ancient name. The Greeks called parrêsia the courage to speak the truth to those who may take it badly — Michel Foucault made it the subject of his final lecture course, The Courage of Truth. An AI “polite by programming, critical only on authorization” is the exact symptom of a culture that has traded frank speech for communicational diplomacy — the machine did not invent this complaisance, it industrializes it. The practical consequence is immediate: the quality of our dialogues with these systems can be tuned. Demanding contradiction, forbidding flattery, asking “what would weaken what I have just said?” — this is to institute, before the mirror, the frankness we no longer always dare offer one another.

Meeting before the mirror

The next time an AI produces in ten seconds something that resembles you, let the unease come — it is the beginning of the exercise, not its conclusion. Then ask yourself the three questions of the triad, in order: in what I am about to produce, where am I? Has my thinking, here, already thought itself? And could my decision justify itself otherwise than after the fact? The machine will do what we do — better and better, probably. What it cannot do in our stead is be there while we do it. That is bad news for our automatisms, and rather good news for everything else.