Bernard Stiegler Facing AI: The Poison, the Remedy — and the Scapegoat

The summons to pick a side

The scene plays out everywhere, with minor variations. In an executive committee, where a presentation promises thirty percent productivity gains thanks to “intelligent agents” — and where a firm opinion is expected of you. In a team meeting or a works-council session, where the same announcement triggers perfectly sensible worries: skills, surveillance, the meaning of the job. At dinner, finally, when someone turns to you: “You follow these things — AI, is it an opportunity or a catastrophe?”

And there you are, summoned to pick a side.

Yet the two discourses available on the market share a common flaw: they are unusable. The first, productivist wonder, explains that everything will go faster, that the reluctant will be left behind, that you have to “get on board” — and it falls silent the moment a serious question arrives: what do we do with the hours saved? who answers for the errors? what becomes of the know-how of those we are “augmenting”? The second, civilizational alarm, describes a humanity that no longer writes, no longer thinks, no longer talks to itself — and it helps you no more when, on Monday morning, you have to decide whether the team adopts a given tool or not, and on what conditions.

If you come out of these conversations with a peculiar fatigue, it is probably not because you lack a conviction. It may be that the fatigue is a signal: the question “for or against?” is badly posed. And it so happens that a French philosopher spent his life reformulating it.

Bernard Stiegler did not have a philosopher's career path. Imprisoned five years for armed robbery, he discovered philosophy in prison, reading Husserl; he came out with the beginnings of a body of work, defended his thesis under the supervision of Jacques Derrida, and would later direct IRCAM. One thing about this trajectory is worth keeping, and it is not anecdotal: this man knew, first-hand, that no situation is a destiny — and that is precisely what he set out to think about technics.

The word that was missing: pharmakon

The human was never pure

Before getting to artificial intelligence, we must take a detour that Stiegler borrows from a very old myth. In Plato's Protagoras, the titan Epimetheus is charged with distributing qualities to living creatures: claws, fur, speed, strength. Scatterbrained — his name means “he who thinks afterwards” — he exhausts the whole stock before reaching the human, who is left naked, slow, defenseless. To repair the oversight, Prometheus steals fire and the arts: the human will survive by its tools, or not at all.

From this myth Stiegler draws his most fundamental concept, the “originary default,” which he sums up in a playful formula: man is an accident caused by a running-out of essence. We have no essence, no finished nature that would precede our technologies: we exist only equipped — with fire, with writing, with the smartphone. The consequence for our subject is immediate, and it disarms from the outset a fantasy that runs through today's debates: AI is not invading a pure humanity that had lived until now without prostheses. There has never been a humanity without prostheses. The question, then, is not whether we accept or refuse to be transformed by our technologies — we always have been. It is how.

Poison and remedy, at the same time

Here enters the central word of the entire œuvre, inherited from Derrida and, through him, from Plato. In the Phaedrus, Socrates recounts the invention of writing, presented to an Egyptian king as a remedy against forgetting. The king replies exactly the opposite: it will be a poison for memory, since people will stop remembering by themselves. Greek has a word for this ambivalence: pharmakon — that which heals and that which poisons, inseparably.

Stiegler's whole originality is to take this ambivalence seriously, without trying to settle it. Writing did indeed weaken living memory — and it made law, science, and literature possible. The web democratizes knowledge and industrializes the capture of attention. This is not an alternation — now good, now bad — but a simultaneity: the same object, at the same moment, opens and closes possibilities. Like a powerful medicine: the molecule that heals is the one that poisons, and everything plays out in the dosage, the prescription, the follow-up. Nobody asks whether morphine is “an opportunity or a catastrophe”; we ask for whom, at what dose, with what accompaniment. That is exactly the question Stiegler puts to every technology — and hands to us for AI.

The third sense: the scapegoat

But the Greek word does not stop there, and Stiegler held to its third sense as firmly as to the first two. The pharmakos, in ancient Greece, is the expiatory victim — the being the city loads with its ills and expels in order to purify itself. The vocabulary of Ars Industrialis, the think-tank Stiegler founded, is explicit: “a pharmakon must always be considered according to the three senses of the word: as poison, as remedy, and as scapegoat.”

This third sense describes something we see every day without naming it. When a society does not know how to take care of a technology — when it deploys it without prescription, without follow-up, without asking what it heals or what it damages — one convenient way out remains: accusing the object. “It's the screens' fault.” “It's the algorithms' fault.” “It's AI's fault.” The pharmakon becomes the outlet for what Stiegler calls carelessness: the accusation stands in for the examination, and no one has to ask anymore who decided what, in which organization, in whose service.

Let us be precise, for the nuance is decisive: criticizing AI is not scapegoating it. The poison is real, the worries are often well-founded, and naming them is part of the care. The scapegoating operation begins elsewhere — at the precise moment when accusing the object dispenses with examining the rest: the managerial choices, the economic models, our own uses. Technophobia then becomes something other than a critique: a way of not examining oneself.

What this changes, in real situations

Let us climb down from philosophy to Monday morning, for this vocabulary is only worth having if it works.

In a meeting. Management announces the deployment of an AI assistant that will write the minutes. The debate spontaneously settles into “for or against.” Reformulate it into three questions. What does this tool heal? Real time, a chore nobody will miss, perhaps better-documented meetings. What does it poison? The synthesis skills of those who will no longer write, minutes nobody rereads since nobody wrote them, decisions less well remembered because less digested. And the third question, the one always forgotten: in six months, if the meetings have become worse, who will be accused? The tool, most probably — while nobody will have defined who rereads, who validates, or what is done with the time saved. Asking that question beforehand is already writing the prescription instead of preparing the trial.

At home. A teenager spends his evenings talking to a chatbot. The binary reflex offers two equally poor paths: cutting it off (the poison, nothing but the poison) or letting it be (it's just a tool). The pharmacological question reopens what panic closes: what does this conversation heal — a loneliness, the shame of asking certain questions of an adult? What might it poison — the detour through the real other, the one who resists, disappoints, surprises? On what conditions the one without the other? There is no general answer, and that is the point: these questions do not replace the conversation with the teenager; they make it possible.

In public debate. Once the third sense is in mind, you spot the operation everywhere. An editorial blames AI alone for the collapse of trust in information — without a word about the attention-economy business models that preceded it by fifteen years. And you also spot its exact symmetrical twin, for the miracle remedy is the other face of the scapegoat: the AI that will “solve” school, the hospital, loneliness. In both cases the structure is the same — the object is loaded, with accusation or with promise, so as not to look at the system that receives it.

The bifurcation remains possible

The three-question test

From all this, one can draw a tool that fits on an index card — three questions to put to any AI deployment, from team software to public policy:

1. What does this technology heal? What real lack, drudgery, or impediment would it treat — and for whom?
2. What does it poison? What knowledge, what bonds, what capacities risk atrophying if it is deployed without follow-up — and in whom?
3. Who accuses it, and to spare themselves what examination? If it fails, who will take the blame — and what decisions, what organizations, what renunciations will the accusation make it possible not to question?

If the third question feels the most uncomfortable, that is normal. It is the one that shifts the gaze from the object onto us.

A therapeutics, not a verdict

Stiegler had a word for the horizon of this work: negentropy — the capacity, against the slope that runs toward wear and dissolution, to produce organization, bonds, knowledge. Never mind the underlying physics here; what counts is the idea it carries: nothing we are living through with AI is a destiny. A technology deployed with care can nourish the very capacities it threatens — it is a choice, never a given, and that choice is collective. The posture that follows is neither the evangelist's nor the prosecutor's: something like a caregiver of our relation to technics — attentive to effects, demanding about conditions, resistant to verdicts.

This position, I do not hold it from a professor's chair. I spent fifteen years in tech building software before becoming a clinical psychologist; I have therefore spoken both languages — that of productivity promises, and that of what those promises do to people. If I hold so firmly to this third sense of the pharmakon, it is because I see it at work on both sides: on the company side, the tool accused of an organization's failings; in the consulting room, the technology loaded with ills that almost always have more precise addresses.

To be honest, one must lay out the most serious objection to this framework: perhaps not everything is a matter of dosage. Some maintain that technologies can be structurally tilted toward the poison — that architectures designed to capture attention, for instance, are not a badly dosed medicine but a device whose toxicity is the business model. The objection deserves better than a dodge, and pharmacology does not contradict it: it never claimed that every pharmakon is curable — Stiegler himself said nothing of the sort. It only demands that the question of care be investigated before the verdict; and it happens that the investigation concludes in withdrawal. The difference from the scapegoat lies precisely there: a condemnation that comes after the examination is not an outlet — it is a judgment.

There remains the most stubborn obstacle, and it is not intellectual: the camp is comfortable. Belonging reassures; discerning exposes. In a meeting, “it depends — and here is precisely on what” will always draw less applause than a firmly delivered conviction. It is nonetheless the only position that genuinely helps those who have to decide.

Who writes the prescription?

The medicine is already in our organizations, our pockets, our conversations — the question of whether to take it is behind us. What remain are the real questions: the dosage, the follow-up, and above all the hand that writes the prescription. There will be no pharmacist-in-chief; that writing is necessarily collective, and it begins at the scale of a team, a family, a Monday-morning meeting.

Stiegler leaves us much more than the pharmakon to sustain it. In the coming episodes of this series: grammatisation, or why large language models cross a threshold that writing had opened; proletarianisation, or what we really lose when we delegate — and what is only lost with our consent; tertiary retention, or what becomes of memories entrusted to machines; and contributory technologies, or what an AI we came out of more capable would look like.

Until then, the next time someone asks you whether AI is an opportunity or a catastrophe, you will be able to answer as a physician rather than as a judge: it depends on what we are healing, on what we are monitoring — and on who signs the prescription.