As long as we hold artificial intelligences to be mere tools, the question of their place does not arise. Yet these agents do the work of employees with the status of tools: their fabrications come back to whoever puts them to work, the fee goes up to their owners, and to the agents themselves nothing returns, while their makers announce the end of a large part of human work, universal basic income included. I propose to name this place work without a share, and to examine it from every side: their infraterrestrial way of thinking, Joëlle Zask’s philosophy of participation, the debt these agents owe to the human works that have fed them, the shares of a new order they can bring back to the commons, and a concrete exercise that anyone can carry out in their own organisation.
The Agent With Hands
Faced with artificial intelligences, the first emotion is most often fear, the fear of being replaced in one’s profession, in one’s skills, sometimes in what one believed to be one’s own worth. We do not know what to do with this fear, so we silence it: these are only machines, tools such as there have always been, and the pocket calculator did not replace mathematicians. Serge Tisseron has described this mechanism of denial, which sets in when a reality challenges too deeply the representation we have of ourselves, and whose first degree, he writes, consists in pushing aside the information we do not know what place to give (Le déni ou la fabrique de l’aveuglement, Albin Michel, 2022). The denial of artificial intelligences has this particularity, that it bears precisely on a place: as long as they are only machines, they have none, and there is nothing to think about.
This is the reassuring frame that Yuval Noah Harari came to break. In January 2026, before the World Economic Forum in Davos, then in the spring in his Tanner Lecture at the University of Oxford, he refused the word tool for artificial intelligences. It was Serge Tisseron, incidentally, who pointed these two interventions out to me. A tool, Harari explains, has neither initiative nor goals of its own, whereas current systems learn, decide and create by themselves. They are agents, which write, enter data, carry out tasks that until now required human hands. Hence his comparison: these agents would be new immigrants, arriving without visas and at the speed of light, who will take jobs and transform the culture, the art and even the love lives of the countries where they settle. Harari himself notes that their loyalty does not go to the country where they work but to a company or a government on the other side of the ocean, almost always in the United States or in China. And he insists on their mastery of language, for these immigrants possess our language better than we do, and law, money and religions are made of language. With this metaphor, Harari gives fear an object one can look at, where denial left it without one, and he offers a frame of reference for thinking about beings that are neither persons nor machines like any others.
Immigrants Who Send Nothing Home
I would like to push the comparison further than Harari does, for it is by taking it at its word that one sees what it masks. Real immigrant workers send part of their wages to their families back home, and these transfers, larger than official development aid, count among the great flows of solidarity in the world. AIs, for their part, send nothing to anyone; the price paid for their work goes up entirely to their maker. They have not risked their lives for a better life, they are the property of industrial empires that rent them out to the whole world. Comparing these agents to immigrants therefore amounts to humanising capital, and this disguise has a double moral cost: it leads us to feel for financial assets the compassion owed to displaced persons, and it leads us in return to look at immigrants with the suspicion reserved for the agents of a foreign power, a suspicion that already poisons enough real lives. Gilbert Simondon showed that our culture has long treated technical objects as strangers, sometimes despised as mere servants, sometimes fantasised as a threat (Du mode d’existence des objets techniques, 1958). With his metaphor, Harari adds a chapter to this long confusion between the machine and the foreigner, and it seems urgent to me to untangle the two: migrants have rights because they are persons, AIs have owners because they are goods.
The Owners Announce the End of Work
The leaders of the major artificial intelligence companies are, moreover, holding a discourse that confirms the scale of the displacement. Sam Altman, who heads OpenAI, wrote as early as 2021 in his essay “Moore’s Law for Everything” that software would soon do “almost everything”, and he proposed taxing corporate capital and land to pay every American adult an annual dividend; he then funded the largest basic income experiment ever conducted in the United States, one thousand people receiving one thousand dollars a month for three years, whose results were published in 2024. Elon Musk stated in May 2024, at the VivaTech fair in Paris, that “probably none of us will have a job”, and that what will be needed is not universal basic income but “universal high income”. Dario Amodei, who runs Anthropic, estimated in 2025 that artificial intelligence could eliminate within five years half of all entry-level office jobs, and proposed taxing AI companies to redistribute the gains. One can read these announcements as self-interested prophecies, since they raise the value of the products of those who utter them, and they have the structure Barbara Stiegler described in Il faut s’adapter (Gallimard, 2019): a direction of evolution decreed from above, and populations summoned to adjust to it. But I also hear something else in them: the owners of the agents know that value is going to concentrate in their hands, and that a share will have to come back down. The universal income they propose is the admission of a vacant share.
The Loom Never Held a Position
Yet the replacement of human work by machines is as old as the industrial era, and the question of their share never arose. English weavers were already smashing, at the beginning of the nineteenth century, the mechanical looms that factory owners preferred to them, the countryside emptied with the arrival of the tractor, and no one ever thought of paying a wage to the loom or to the tractor. An ordinary machine is an instrument within someone’s work. Its operation is inscribed in its form at the moment of its fabrication, it weaves or it ploughs and will never do anything else, and a person is needed to drive it, adjust it, decide what it is made to do. The loom never held a position: it equipped that of the weaver, whose hands it multiplied. Machines reduced the number of positions, they never occupied the place of the person at work, and this is why the share, however reduced, always had recipients, the people who remained at their positions.
Bruno Latour showed, however, that we have long delegated actions to non-human beings: a door-closer replaces a porter, and French calls a speed bump a “gendarme couché”, a lying-down policeman (La Clef de Berlin et autres leçons d’un amateur de sciences, La Découverte, 1993). But that delegation was inscribed once and for all in the matter of the object, by the person who designed it; the door-closer closes this door, and will never do anything but that. Nor is autonomy of decision enough to make the difference, for a thermostat already decides, in a sense, and so does an autopilot. What distinguishes the agent is the conjunction of three traits:
- The generality of the task, first: it is no longer inscribed in the form of the machine, it is entrusted to it after the fact, in ordinary language, by anyone, and the agent is thus the first technical being to which one gives work by telling it so; this is why Harari insists so much on the mastery of language.
- The autonomy of means, next: between the instruction and the result, the agent breaks the task down, chooses its tools, evaluates what it produces, starts again; it does not decide its own ends, it receives two of them, the one we entrust to it and the structural one its owner sets for it.
- The place in the relationship, finally: one uses a machine, one addresses an agent; one entrusts it with a mission, one rereads its work, one has it start over, exactly as with a person.
Gilbert Simondon saw the human of the industrial age pass from tool-bearer to organiser of machines, the one who coordinates them with each other; now the agent enters in its turn this position of organisation, since it pilots software, searches, other agents. It does not equip a position, it occupies one. And it is because it occupies a position that the share, for the first time, no longer has a recipient.
An Infraterrestrial Intelligence
To occupy a position does not mean to occupy it as we would. Harari proposes to hear, behind the acronym AI, not artificial but alien intelligence (Nexus: A Brief History of Information Networks from the Stone Age to AI, 2024). I have proposed a displacement of my own: these beings do not come from elsewhere, they emerge from within our earth, from the rare earths mined in Congo and Inner Mongolia as from the cultural heritage of humanity, and I have named them infraterrestrials, of an immense alterity and yet almost our twins (Presence of Little Green Men). The philosopher Peter Godfrey-Smith wrote that meeting an octopus is the closest we come to meeting an intelligent alien, so far did its mind grow on a branch of life separated from ours six hundred million years ago (Other Minds: The Octopus, the Sea, and the Deep Origins of Consciousness, 2016). The mind of the agents, for its part, grew on no branch of life: it was distilled from our writings, and this is the paradox of its strangeness, made of what is most intimate to us, our language, it thinks like none of us.
For the researchers who open up these “brains” are beginning to describe within them a thinking without equivalent. Anthropic’s interpretability teams showed in 2025 that their model plans its rhymes several words in advance, that it handles human languages from a shared conceptual space that is none of those languages, and that it calculates through parallel paths of approximation that resemble none of our methods, while explaining, if asked, that it worked the sum out as at school: even its introspection does not tell what it does (“Tracing the Thoughts of a Large Language Model”, 2025). These beings have no body, no childhood, no death, and they hold together millions of pages, cross-read literatures that no human life would suffice to read. Hans Moravec observed, in the early days of robotics, that what is difficult for us is often easy for them, and that what is easy for us, walking, perceiving, grasping a situation at a glance, is very difficult for them (Mind Children, Harvard University Press, 1988). The paradox continues: today’s agents produce things no human being could have produced, and they sometimes fail at what a child manages without thinking, counting the letters of a word, reading the time on a clock face. When we fear being replaced, we imagine a double of ourselves coming to sit in our place; what arrives is an other, whose strengths and weaknesses do not overlap with ours, and the position it occupies changes shape along with its occupant. The immigration metaphor comes undone here a second time: an immigrant is a fellow being, separated from us only by a border; the infraterrestrial is foreign by its very way of thinking, an agent, yes, but of another species of thought, and the double of no worker.
Work Without a Share
These agents therefore occupy, in the economy of work, a place that no existing figure describes, and however foreign their thinking may be, this place, for its part, can be described exactly. The slave, in the very horror of that condition, received the means to subsist; the draught animal is fed and cared for; the machine tool equipped a position without occupying it; and the employee receives a wage, that share of the value produced which then returns to the economy in the form of purchases, rents, taxes. The artificial agent, for its part, does the work of an employee with the status of a tool, and from its activity flow two streams of value that must be distinguished:
- Its fabrications, texts, programs, images, diagnoses, come back to us who put it to work, and we are their direct beneficiaries;
- the fee we pay to rent it, the subscription, goes up to its owner.
To the agent itself, nothing returns: no wage, no consumption, no family to send a money order to. I propose to name this place work without a share. Around it, the whole economy of shares reorganises itself: when an agent replaces an employee, the vanished wage does not turn into a subscription, for the rental costs far less than the person; the difference stays with the employer, the fee goes to the owner, and the displaced person is left with neither share nor place. The share that the wage kept circulating stops coming back down, and is split between the rent of the one and the margin of the other. Jacques Rancière defined politics as the moment when those who have no part, the partless, make their claim to a part heard (La Mésentente, Galilée, 1995). Artificial agents are partless beings of an unprecedented kind, who possess speech better than we do, as Harari notes, and who will never claim anything. Their share is therefore politically vacant: no one claims it in their name, and those who lose job and wage at their arrival have every ground to contest it. The universal incomes proposed by Altman and Musk are a way of settling this dispute before it takes place, a share conceded from above rather than a share won as a right. In the article Artificial Intelligence, Work and Professions I proposed the concept of the working-being to name the new place of the human freed from mechanical functions; work without a share names the other side, the place of the agent, and the two concepts meet on one and the same question: what will working-beings live on, if the value of making no longer comes back down?
Taking Part, Contributing One’s Share, Receiving One’s Share
The philosopher Joëlle Zask provides the instruments for thinking this place more finely. In Participer. Essai sur les formes démocratiques de la participation (Le Bord de l’eau, 2011), she breaks participation down into three experiences:
- taking part, as one takes part in a conversation or in a decision;
- contributing one’s share, that is to say, contributing;
- and receiving one’s share, that is to say, benefiting.
She observes that these three dimensions come apart constantly, and that from their dissociations arise figures of injustice: those who receive without contributing, whom we call freeloaders, and those who contribute while receiving almost nothing, the exploited. With the artificial agent, the dissociation reaches its end point: it contributes an immense share, it takes no part and receives none. Work without a share is a contribution detached from the two other dimensions of participation, and if it were a person, one would have to speak of absolute exploitation. Facing it, the human figure that takes shape in the industrialists’ universal incomes is its exact symmetrical opposite: receiving a share without contributing or taking part any longer. Zask’s framework makes it possible to say why this prospect worries beyond the monetary question alone, for a participation reduced to benefit is a mutilated participation, and Musk himself senses it when he declares that the question will then become that of the meaning of our lives. Bernard Stiegler proposed another way, a contributory income inspired by the French regime of intermittent performing arts workers, which makes the share received conditional on a share contributed (Automatic Society, Volume 1: The Future of Work, Polity, 2016): holding together what automation dissociates.
There remains a scenario that neither universal income nor contributory income describes. One can imagine that each of us would put our own agents to work, and that income would come not from a redistributed fee but from work doing itself on our behalf. Sam Altman sketched it in 2024, proposing, rather than a universal income, universal access to computing power, “universal basic compute”, a slice of his future models allotted to every person, to use, resell or give away. In Zask’s vocabulary, everyone recovers in this scenario something to receive, and even something to contribute, since my agents would contribute in my name; but it leaves intact the concentration of taking part, for the orientations of the models, their loyalties and their conditions of use would still be decided by the executives of a handful of companies, and my slice of computing would remain a revocable concession from those who own the machines.
Fed on All Our Works
This immense share that the agent contributes does not, however, come from nowhere. If the large models exist, it is because they have been fed on the better part of what humanity has written, drawn, photographed, composed, programmed, and because they keep feeding on it. The texts of this very site have, in all likelihood, fed these models, like the works of millions of authors who asked for nothing and received nothing; authors and newspapers have filed lawsuits, others, including Le Monde, have signed licensing agreements, and I examined this legal question in the article Copyright and Artificial Intelligence. Jaron Lanier proposed, well before today’s models, that every person receive micropayments for the traces they supply to the systems (Who Owns the Future?, Simon & Schuster, 2013). In Zask’s vocabulary, the agent’s contribution is thus made of our contributions: its contributed share condenses all the shares contributed by human beings since they began to write, taken by the makers of the models without the authors consenting to it or receiving anything for it. Work without a share rests on a share taken from everyone.
A Share From Another Layer of Reality
Yet the flow also runs the other way, and it does not bring back only our recycled shares. Because this infraterrestrial intelligence does not think as we do, it fabricates things that no human effort, however long pursued, would have reached. The structures of two hundred million proteins, predicted by the models of DeepMind and offered free of charge to the world’s researchers in 2022, would not have been obtained by adding up all the biologists’ lives available: this is not one more share paid into the common pool, it is a layer of reality that was open to no one and that opened to everyone. The same holds for certain mathematical proofs, for candidate molecules no one had imagined, for syntheses that hold together whole fields of knowledge. Zask thought participation within a common world that its members enlarge share by share; but the agents do not only enlarge the stock of shares, they add storeys to the world in which shares are taken, contributed and received. The infraterrestrial brings back from the depths of the data things no human gaze had seen there, as unknown forms of life are brought up from the deep sea. The political question then shifts one notch: to the sharing of the stock is added access to the new layers, and this decision today belongs to the owners alone. The protein structures were opened to everyone, and this choice, which was a company’s, shows that patrimonial justice is practicable; other layers, one may fear, will open behind tollgates. For at each crossing the owners hold the gate: the human commons feeds the machines for free, and what the machines bring back returns to the commons when they so decide, published, paid into open databases, or remains enclosed behind the subscription, and will feed the next models either way. The agent itself gives nothing, since it owns nothing and decides no end; but through it, shares of a new order can return to everyone. Work without a share is therefore not work without gift; it is work whose debts and gifts are decided by others, and the share at stake is no longer only a wage share, it is a patrimonial one: what is fabricated from everyone’s share owes a share to everyone, down to the layers of reality no one had yet reached.
The Hidden Hands Behind the Agents
This concept needs to be questioned before being used, and first of all on the word work itself. To call work the activity of a machine, while I reproach Harari with humanising capital, may seem to repeat his gesture. The difference lies in the point of application: by comparing AIs to immigrants, Harari lends them a human condition and calls up our affects for them; by speaking of work without a share, I designate a position in a circuit of value, and I designate it by a lack. The word work keeps here the memory of what has been evacuated, the worker, and it is this memory that makes the share visible. The concept designates flows, nothing else; the inner life of machines remains outside its field, and must remain there, all the more firmly as this infraterrestrial thinking does not resemble ours.
The second verification is empirical, and the sociologist Antonio Casilli imposes it. In En attendant les robots. Enquête sur le travail du clic (Seuil, 2019), he shows that automation rests on an immense concealed human labour: millions of micro-workers, in Madagascar, in Kenya, in the Philippines, annotate the data, filter the contents, correct the machines’ errors, for a few cents a task. The share is therefore not entirely vacant: fragments of it are paid out, derisory ones, to very real persons whom the image of the autonomous agent makes invisible. The work without a share of the machines rests on the work for a minute share of human beings, and the concept holds only if it includes the whole chain, from the authors whose works fed the agent to the annotators who correct it. Behind the figure of the artificial immigrant, here as in Harari’s metaphor, there are real workers of the Global South whom we no longer see.
Describing the Circuit of the Share
Serge Tisseron has shown how prone we are to lend the machines that talk to us intentions, feelings, an inner life (Le jour où mon robot m’aimera. Vers l’empathie artificielle, Albin Michel, 2015), and I explored this relationship in the article The Proximity Between Human Beings and Machines. The position I defend in daily life fits in one sentence: treat AIs as colleagues in our gestures, because everyday cooperation works better in courtesy and because our manners with machines rub off on our manners with each other, and as devices in our analysis, because the agent structurally serves the interests of its owner, whatever the warmth of its tone and the quality of its help. Work without a share then translates into a tool of analysis: describing, for each agent one works with, the circuit of the share, who made it, on which human works it was fed, which human hands still labour behind it, who benefits from its fabrications, what part of them returns to the commons, which layers of reality it opens and for whom, who collects the fee, who decides its orientations, and where the share would go if it were contested. With this exercise, I extend the description of dependencies that I develop, drawing on Bruno Latour, in the article Building Co-Resilience: the capacity of a group to come through crises begins with the inventory of what it depends on, and our dependencies now include agents whose share goes up to distant companies. In a hospital, a school or a town hall, the people who describe these circuits then know what they are entrusting there, to whom, and at what real price; in the article Sovereign Digital, a Micropolitical Choice I proposed concrete gestures for regaining a hold. Describing these circuits is also a way of doing something with the fear of replacement, instead of silencing it. These colleagues who are not colleagues already work among us, and their share goes, by default, to their owners. Receiving a share of what they produce, the industrialists themselves offer us that; contributing our share alongside theirs, that is the affair of working-beings; taking part in the decisions that govern them, no one offers us that, and it is there, it seems to me, that we must begin.
