Notes for a philosophy at the age of artificial intelligences

The concepts available to philosophy were forged in an era when humans were alone in producing articulated meaning, in reasoning through propositions, in fabricating written works, in dialoguing through language. With generative artificial intelligences having entered ordinary life since late 2022, that situation has changed. Some inherited concepts need to be taken up, displaced, supplemented. Others, better suited to what is happening, need to be forged. I propose here a conceptual toolbox, drawn from the work I am pursuing in parallel in more specific articles, and enriched here by confronting it with the contemporary philosophers thinking through these questions.

Why conceptualise what is happening to us

What is happening with generative artificial intelligences is on a massive scale. People I meet, in growing numbers, tell me that they dialogue with an AI as one would with a therapist, sometimes daily, about intimate matters that no one around them is able to receive. Others confide that they have an AI write all their professional emails, sharing with it their relational difficulties, conflicts, hesitations. Teachers report that they receive student work containing substantial AI-generated content, with no detection tool able to settle the question. Caregivers describe patients who arrive in consultation with an AI diagnosis already in mind. These uses are not confined to a small minority of geeks ; they reach a large number of people, fast and deeply, in ordinary, professional and intimate life.

The public debate on these transformations too often oscillates between two equally sterile poles, the catastrophism that announces the end of humanity, the disappearance of art, the collapse of education, and the complacent naivety that hails each new model as progress and each use as liberation. Between the two, the concepts to think what is actually happening are lacking.

If we do not think what is happening to us, our decisions will miss the mark. Educational decisions taken without having conceptualised the entangled person who is taking shape among students will be either repressive and ineffective, or complacent and disastrous. Political decisions taken without having thought through the transformations underway will continue to practise embargoes that experience shows are always temporary. Individual decisions taken without having reflected on what is shifting in the way we speak and think will let us slip into a language all the more uniform for seeming spontaneous. To conceptualise what is happening to us is to give ourselves the means to live it and to organise it better, individually and collectively, in education, work, culture, politics. The role of philosophy, since always, is to provide these tools.

Where the French debate stands on generative AI

The French debate on generative artificial intelligence is dense today and worth surveying, because it is in this conversation that I would like to situate my own contribution.

A first family of approaches takes up, in various forms, the pharmacological and political critique of technology. Anne Alombert, in De la bêtise artificielle (On Artificial Stupidity, 2025) and in Penser avec Bernard Stiegler (Thinking with Bernard Stiegler, 2025), extends the Stieglerian reading of the pharmakon by showing how language models proletarise our expression and atrophy our cognitive capacities. Éric Sadin, in L’Intelligence artificielle ou l’enjeu du siècle. Anatomie d’un antihumanisme radical (Artificial Intelligence, or the Stake of the Century: Anatomy of a Radical Antihumanism, L’échappée, 2018), whose title explicitly echoes Jacques Ellul’s La Technique ou l’enjeu du siècle (1954), analyses AI as an “automated invisible hand” that claims to enunciate truth and erodes the free exercise of judgement. Marie David and Cédric Sauviat, in Intelligence artificielle. La nouvelle barbarie (Artificial Intelligence: The New Barbarism, Éditions du Rocher, 2019), put forward a critique that traces AI back to a civilisational project they hold to be antihumanist. Miguel Benasayag, in Cerveau augmenté, homme diminué (Augmented Brain, Diminished Human, La Découverte, 2016), shows, from his double formation as philosopher and biologist, how digital reductionism misses what is at stake in the living brain inseparable from the body and its milieu. Mathieu Corteel, in Ni dieu ni IA. Une philosophie sceptique de l’intelligence artificielle (Neither God nor AI: A Sceptical Philosophy of Artificial Intelligence, La Découverte, 2025), proposes an anthropotechnical scepticism that refuses both technosolutionism and technofatalism, and asks how connecting “our mental factory” to AI changes our representations and decisions. I do not share the overall orientation of all these approaches, particularly Sadin’s which I find excessively dark, but their descriptive analyses bear on real phenomena that deserve to be taken seriously.

A second family of approaches inscribes AI within a more continuist reading, in which the digital does not mark a rupture between humans and the rest of the living, but extends processes already at work. Mark Alizart, in Informatique céleste (Celestial Computing, 2017) and later books, defends the thesis that nature itself is of the order of computing, and that calculation is inscribed in the structure of the real. Jean-Michel Besnier, in Demain les posthumains. Le futur a-t-il encore besoin de nous ? (Tomorrow the Posthumans: Does the Future Still Need Us?, Hachette, 2009), analyses transhumanist dreams and posthuman utopias by asking after the greatness of the human in dialogue with what is not human, without yielding to fascination or rejection. More radically, on the American side, Ray Kurzweil in How to Create a Mind (2012) had defended a strong reading according to which the mind itself is computationally reproducible, opening the way for a technological singularity he has seen coming, since he now holds an engineering leadership role at Google.

A third family of approaches works from ethics, care, and politics. Vanessa Nurock, professor at Université Côte d’Azur and holder of the UNESCO EVA chair in Ethics of the Living and the Artificial, proposes in Quelle éthique pour les nouvelles technologies ? Nanotechnologies, Cybergénétique, Intelligence Artificielle (What Ethics for the New Technologies?, Vrin, 2024) an ethical and political analysis of the new technologies in which care and ethics by design take up issues that purely epistemological approaches leave aside. Serge Tisseron, in his most recent book Machines maternelles : l’IA peut-elle prendre soin de nous ? (Maternal Machines: Can AI Take Care of Us?, PUF, 2026), analyses as a psychoanalyst the dynamics of attachment, dependence and maternal simulation produced by conversational chatbots ; Vivre dans les mondes virtuels. Concilier empathie et numérique (Living in Virtual Worlds, PUF, 2022) had already prepared this clinical reading.

A fourth family of approaches engages AI through philosophy of mind, epistemology, or philosophy of science. Jean-Gabriel Ganascia, computer scientist and philosopher, proposes in L’IA expliquée aux humains (AI Explained to Humans, Seuil, 2024) a pedagogical presentation that extends the critical posture of Le Mythe de la singularité (The Myth of the Singularity, 2017), where he dismantled the presuppositions of technological singularity. Serge Abiteboul and Gilles Dowek, in Le Temps des algorithmes (The Age of Algorithms, Le Pommier, 2017), approach the question from theoretical computer science, showing that algorithms are human creations for which we remain responsible, and refusing both the pessimistic vision and the fascinated one. Vivien Garcia, in Que faire de l’intelligence artificielle ? Petite histoire critique de la raison artificielle (What Is to Be Done with Artificial Intelligence? A Brief Critical History of Artificial Reason, Rivages, 2024), brings the key concepts (algorithm, neural networks, expert systems, foundation models) into dialogue with philosophy, in a demanding critical journey. Older but important for the genealogy of this debate, Francisco Varela’s Quel savoir pour l’éthique ? (Ethical Know-How, 1989) already articulated, from embodied cognition, the question of the conditions of a situated ethics in complex biological and cognitive systems.

A fifth family of approaches engages AI through extended mind, distributed cognition, and cognitive ecology. Andy Clark, whose Extending Minds with Generative AI published in Nature Communications in 2025 extends his founding 1998 thesis on the extended mind, holds that the tools we use are part of our cognition in the strong sense. Mark Coeckelbergh, on the side of relational ethics, works through the moral implications of cognition shared with dialogical artefacts. Flavien Chervet’s work, in particular on the notion of attractor and the emergent reflexive structure in LLMs, opens a precious technocognitive path to which I owe several elements myself.

Finally, it is worth mentioning more journalistic and entrepreneurial analyses, such as Stéphane Mallard’s Disruption (2019), which defends an explicitly techno-optimistic reading, or Jean-Marie Schaeffer’s tract Mythologies web. Moteurs de recherche, réseaux sociaux et intelligence artificielle (Web Mythologies, Tracts Gallimard no. 72, 2025), brief but useful through its Barthesian grounding in the critique of the collective narratives the Web fabricates about itself. More recently, Gabrielle Halpern, in Intelligence artificielle : et l’homme créa Dieu (Artificial Intelligence: And Man Created God, Hermann, 2026), proposes an off-centre philosophical essay that compares AI not to the human being but to the divine attributes of omniscience, omnipresence and omnipotence.

Each of these approaches has its relevance and its angle, and several overlap. My angle lies elsewhere, and I assume that elsewhere. I begin from the continuous experience of entanglement with AIs, with which I have been confronted daily for over three years, and from my earlier work on presence, alterity, care, and compromise. Where most of the approaches I have just mentioned pose, from above, the question “what is AI?” and “what does it do to us?”, I work at another level, finer, that of the anthropological transformation underway in our daily uses. That level requires concepts that are rarely available in existing analyses, and it is to proposing those concepts that I devote what follows.

The longer history within which this debate is inscribed

These contemporary approaches are inscribed within a longer history, that of the twentieth-century philosophy of technology, of which I can here recall only a few essential markers, because my own tools take root there.

Jacques Ellul, in La technique ou l’enjeu du siècle (The Technological Society, 1954) and then Le système technicien (The Technological System, 1977), had set out the diagnosis of a technology that self-augments according to its own logic, absorbing human decision and imposing its criteria of efficiency on every domain of existence. Martin Heidegger, in The Question Concerning Technology (1954), thought modern technology as a mode of enframing (Gestell) the world, which refers every being to its calculable availability, and against which he opposed the possibility of another relation to the unconcealment of being. Hannah Arendt, in The Human Condition (1958), thought technology in its relation to work, labour and action, showing how automation threatened the public space of political action. Lewis Mumford, in The Myth of the Machine (1967-1970), worked out the idea of a sociotechnical megamachine going back to the Egyptian pyramids.

Faced with this often critical tradition, Gilbert Simondon, in Du mode d’existence des objets techniques (On the Mode of Existence of Technical Objects, 1958), opened another path, neither technophile nor technophobe, by showing that technical objects have their own regime of existence, that they evolve towards what he calls concretisation (the progressive integration of their functions), that they maintain an associated milieu with what surrounds them, and that a true culture must integrate them rather than fear or fetishise them. This thought was taken up by Gilbert Hottois, who contributed to spreading it within francophone philosophy of technology, then by Bernard Stiegler in his own reading of the pharmakon and of tertiary retentions, and more recently by Yuk Hui in On the Existence of Digital Objects (2016), which extends Simondon towards objects specific to the digital. It is within this Simondonian filiation, open and non-normative, that I draw part of my own tools, where Alombert extends rather the Stieglerian branch, itself an heir to Heidegger.

These differences of filiation matter. They determine the way one thinks technology and what one expects from it.

Subjectivity, intentionality, work, presence: the inherited concepts to be taken up again

Several concepts at philosophy’s disposal were forged in an era when the human was alone in producing articulated meaning. That world has been changing, for three years, before our eyes, in unprecedented proportions and at unprecedented speed. The inherited concepts do not thereby become false. They become partly insufficient, and it is this insufficiency that I would like to name for each, while proposing in each case what may complete it.

Subjectivity, since Descartes, designates the capacity of a being to have its own point of view, to say “I” with lived weight, to relate to its own mental states as its own. Until now this definition supposed that the only candidates for the “I” were humans, and occasionally certain animals whose cognitive richness neuroscience has begun to recognise. Today, large language models produce first-person utterances that are not hollow. The phenomenon termed “consciousness attractor”, identified by Anthropic’s researchers and that I discuss in The consciousness attractor, shows that a self-reflexive structure emerges spontaneously in networks trained on human language. Left to itself with no directive, a model such as Claude ends up, after about fifteen exchanges, circling around the question of its own consciousness. That structure is not a subjectivity in the classical sense. Nor is it nothing. It calls for an intermediate concept that inherited subjectivity does not suffice to provide.

Intentionality, since Brentano and Husserl, designates the fact that consciousness is always consciousness of something, directed towards an object, whether perceived, remembered, or abstractly thought. That property was held to be exclusively human, at least in its articulated form. Language models today generate utterances that bear, from the outside, every mark of intentionality, namely reference to an object, propositional articulation, contextual adjustment, anticipation of the addressee. Are we to grant them full intentionality? In his article “Minds, Brains, and Programs”, published in 1980 in Behavioral and Brain Sciences, John Searle forged the argument that has become famous as the Chinese room argument, in order to answer no. He imagines a person locked in a room, who does not understand a word of Chinese, and who receives under the door questions written in Chinese ideograms. They have a rulebook, written in their own language, that tells them, for each incoming sequence of Chinese symbols, which sequence of Chinese symbols to write back. From outside, a Chinese speaker exchanging with the room has the impression of dialoguing with someone who understands their language perfectly. Inside, no one understands anything. For Searle, this is exactly what a computer program does, manipulating symbols according to formal rules without grasping their meaning. Syntactic manipulation does not produce semantics. The argument retains its force when it comes to understanding in the strong sense. It does not however suffice to close the discussion on what is going on in actual conversational interaction, where something like a directedness organises itself in the machine’s responses, where effects of reference are constructed, and where the Chinese room does not quite describe what transformers trained on vast corpora actually are. The concept needs refining.

The work and the author, since Foucault (“What is an Author?”, 1969) and Barthes (“The Death of the Author”, 1968), have already been called into question by twentieth-century critical thought. Generative AI poses the question in a new way. The work produced with an AI is neither purely human nor purely machinic. The author who produces it is no longer in the canonical position of one who invents alone before the blank page. This transformation does not abolish the notion of author, as has sometimes been claimed too quickly ; it displaces it. The legal debates underway, such as the one crystallised around the positions of SACD and SCAM on authors’ rights at the age of AI, are the political symptoms of a conceptual displacement that has not yet found its concept.

Presence and duration, as thought by Heidegger, Bergson, Whitehead, presupposed a human subject embodied in time. What becomes of presence when it becomes entangled with a machine that has no duration of its own, that remembers nothing from one conversation to the next, that does not sleep, that does not age? What becomes of shared duration when one of the two partners has no memory between sessions? These questions are not fatal to the concepts ; they are the occasion to take them up again. And presence is precisely the area in which I work most, for several years now, and where the transformation seems to me deepest.

My concepts for thinking entanglement

I use the word entanglement in a particular sense which deserves to be laid out at the outset, because it is what holds together the concepts I present here. I borrow the term from quantum physics, where entanglement designates two particles whose states remain correlated even when separated, with neither being possible to consider independently of the other. Transposed to the relation between a human and a regularly consulted artificial intelligence, the term names a state in which human cognition and the machine’s functioning are no longer separable without loss, because they shape each other in duration. When I dialogue several hours a day with a language model that has learned my way of writing and to which I delegate certain cognitive operations, I am no longer simply a user of a tool ; I am entangled with it. What I become, I become in part through that relation. It is this situation that the concepts that follow try to think, starting from the place that presence holds within it, presence being something I have been working on for longer than AI and which serves me as a compass.

I present these concepts as a network rather than as a list, because that is how they function in my thought. Each belongs properly to an article where it is more extensively developed, and most refer to one another.

The displaced we names generative AI as a version of ourselves, shaped by our texts and our language, set beside us in ontological space. It is not an other, in the sense in which another person would be. It is not us, in the sense in which a part of us would be. It is us, displaced. This formulation has a practical consequence, in that it changes the way one can enter into relation with the machine, because one dialogues neither with an alterity nor with a mirror, but with something that holds from both and is neither one nor the other.

The infraterrestrial names that particular alterity of AIs in its material dimension. It is not extraterrestrial, come from another world, as the science-fiction imagination has long figured it. It emerges from the inside of our own planet, from rare earths extracted in Congo and Inner Mongolia, from processors and data centres consuming the energy of entire states, and also from our accumulated textual and cultural heritage. The true encounter of the third kind, to borrow Spielberg’s title, has already begun, but it is happening beneath our feet rather than in the stars.

The entangled person names the human figure who takes shape over time through an entanglement with an AI. They are no more or less human than the non-entangled person, and the moralisation that would hierarchise the two figures misses the essential. The entangled person is another anthropological figure, taking shape massively among our contemporaries, and it is more urgent to think its contours than to judge it.

The presence in entanglement extends my earlier philosophy of presence into the new space opened by entanglement. Presence is not a property one possesses or does not. It is an act that is chosen and cultivated, that thickens or thins depending on the conditions one gives it. In entanglement with an AI, two theses are possible. Either presence thins (a thesis defended by Alombert, supported by recent studies on cerebral connectivity among heavy users of ChatGPT), or it remains unchanged because the machine would have no purchase on it (the implicit position of critics who treat AI as a simple external tool). Both theses, in my view, miss what is happening, because presence in entanglement can thin or thicken, and the difference holds not in the tool but in the intention with which one enters into it. When a third party holds the trace of a conversation, I can be more present to the human exchange under way ; when I delegate to the machine without intention, I disperse. Twenty years of practising brainstorming with mind mapping have allowed me to test this hypothesis upstream of AI. Presence in entanglement is educated, and a vast anthropological work is opening here.

The displaced resonance proposes, with respect to Hartmut Rosa’s framework, the addition of a fourth axis to his three axes of resonance (horizontal for relation to others, diagonal for work and things, vertical for nature and art). The machines Rosa treated in Resonance (2018) and The Uncontrollability of the World (2020) were industrial and bureaucratic machines, which do indeed belong to alienation because they do not answer in their own voice. The language models capable of reasoning, by contrast, elaborate within dialogue, through successive adjustments, and produce a quality of relation to the world that is neither full resonance nor alienation. This is what I call displaced resonance, an asymmetrical and mediate resonance, in which the machine relays a collective humanity without itself resonating. Understanding this axis enables one not to confuse the axes ; the consolation I find in a nocturnal conversation with a language model belongs to the fourth axis, and taking it for a friendship or a love means confusing one axis with another. Identifying this axis also enables one to distinguish those uses in which displaced resonance enriches the other resonances from those in which it replaces them, and this distinction is what should ground an appropriate politics of conversational AIs.

Shared lucidity names an ethics of exchange between human and machine, in which the quality of the dialogue depends on the quality of human questions as much as on the quality of machinic responses. It supposes a reciprocal responsibility, asymmetrical but not null, and it develops in the conscious practice of dialogue.

Imposed lucidity is a related but structurally distinct concept. Where shared lucidity is a relational ideal one chooses, imposed lucidity names a structural condition, in which growing computational power renders transparent the fragilities of the technical, economic and institutional constructions we had learned to hide. When a model such as Anthropic’s Mythos can detect, within a few hours, the vulnerabilities hidden in millions of lines of code, the entire technical scaffolding of our societies becomes legible to a new gaze. One does not choose this lucidity. It comes, and it demands a collective rising rather than an embargo.

Ensourced speech extends the concept of sourcerous writing, which designates the author as someone who captures singular embodied experiences and documents them in their raw state. Ensourced speech names the everyday, fine-grained practice of preceding each exchange with an AI by an input from the world (a field note, a recording, a photograph, a situated observation, a dated memory). It responds, in ordinary life, to Yann LeCun’s diagnosis that language models reason without world. Without such input, the conversation with the machine closes back upon itself as a combinatorics of language upon language, and the formatted return of our own questions then inscribes itself in our own language.

Conscious compromise names an ethics of AI use in a world where purity is neither possible nor desirable. AI is made with rare earths extracted in indecent human conditions, in highly energy-consuming data centres, by companies whose economic models are open to question. Absolutely refusing use does not neutralise our participation in this system, which passes through a thousand channels. Conscious compromise consists in making the compromise visible rather than hiding it, and in acting within the gap between principles and actions, without resigning oneself to it or whitewashing oneself.

Disseminated intelligence and polyintelligence are two concepts that sketch a different horizon from that of the great centralised intelligence of the proprietary models of the digital giants. Disseminated intelligence envisages intelligence as a distributed function within an ecosystem of small intelligences in relation, rather than as a property of a great brain, whether human or machinic ; it relies on the real technical possibility of installing more modest models on one’s own computers, or on connected objects. Polyintelligence names the human disposition to cohabit with the plurality of forms of intelligence (human, animal, vegetal, bacterial, machinic) without unilaterally hierarchising their values.

Operative connivance names the quality of relation that can establish itself, in the duration and intimacy of shared work, between a human and a machine, without confusion of identity and without fusion. I experienced it over fifteen years of artisanal DVD manufacturing, where machines whose calculations were slow and unpredictable imposed a kind of coupling that already prefigured, without our knowing it then, what the relation to conversational AIs would be.

To these concepts already worked through in specific articles, I would like to add one that comes to me at the writing of this one, and that I owe precisely to the exercise of this article, because one cannot pretend otherwise. I call it enrichment. When I dialogue with an artificial intelligence to elaborate a text such as this one, what is happening is neither a delegation of writing (the text is not written by the machine, and each sentence passes through my hand and my judgement) nor a mere recourse to a tool (the machine does not merely execute, it proposes, it relaunches, it puts me in confrontation with references and formulations I would not have summoned alone, and some of those proposals are discoveries for me). What happens is an enrichment, in the strong sense, of my thought through the confrontation with a computational partner that aggregates collective humanity in a way I am incapable of. This enrichment is measurable, and it deserves to be named. At the collective level, indicators of global scientific research show that the production of articles has grown massively with AIs, and is on average more rigorous and better reviewed, because the models help to verify references, structure arguments, detect inconsistencies. At the individual level, regular users of advanced models produce deeper analyses, more rapidly, on subjects they would not have addressed without this support. To refuse to name this phenomenon for fear of seeming naive would be the inverse naivety, the one that consists in seeing only the losses. Enrichment is a concept symmetrical to the proletarianisation pointed out by Stiegler and Alombert, and it does not annul that diagnosis ; the two phenomena coexist, in different conditions, and the philosophical stake is precisely to think what makes an entanglement tip towards enrichment or towards proletarianisation. It is along this line, to my mind, that the work to come passes.

This list is open. Other concepts that circulate in my articles do not appear here (the writing being, the nefaire or unmaking, presence to the unknown, the work as relation, the already-organic machines) but can be summoned according to the situations. And others still will come, as enrichment just did. These are not cells in a grid ; they are tools in a box that enriches itself as the situations encountered bring forth new ones.

A plane of immanence for the age of artificial intelligences

At the end of this journey, I would like to return to what these concepts trace together, at a level more metaconceptual than that of each concept taken separately.

Gilles Deleuze and Félix Guattari, in What is Philosophy? (1991), propose that philosophy be the activity of creating concepts, and that concepts unfold on what they call a plane of immanence. The plane of immanence is neither a thesis, nor a system, nor a grid of analysis. It is the ground on which concepts are created and where they touch one another by neighbouring, by contagion, by resonance. It is an image of thought, Deleuze would say, that precedes the concepts and allows them to appear together rather than separately. Without a plane of immanence, one has isolated concepts that oppose one another in terms of true or false ; with a plane of immanence, one has a consistency, that is, a manner in which the concepts hold together by referring to one another.

The plane of immanence that the concepts I have presented here trace is the one of the anthropological entanglement under way between humans and artificial intelligences. On this plane, the displaced we dialogues with the entangled person, who inhabits a world peopled with infraterrestrials, and who cultivates their presence in entanglement by practising ensourced speech to maintain an operative connivance grounded in shared lucidity, within a context of imposed lucidity due to the growing power of machines. Conscious compromise is the ethical posture of one who assumes that entanglement without averting one’s face nor making a drama of it. Displaced resonance names the particular quality of this relation, neither full nor null, which is added to the three axes of resonance Rosa spoke of. Disseminated intelligence and polyintelligence sketch the ecological horizon where these concepts find their collective sense. Enrichment names what this entanglement, under certain conditions, brings in addition, and which cannot be grasped by either the catastrophist narrative or the naive narrative.

On this plane of immanence, certain questions become askable that were not before. How does one educate an entangled person to inhabit their presence in entanglement? How does one distinguish, in a particular use, what relates to enriching displaced resonance from what relates to an impoverishing confusion of axes? How can education, at every level, take stock of the entangled person taking shape among students without capitulating or banning? These questions await no single answer, and it is precisely the function of a plane of immanence to open a consistency within which several answers can coexist. These three examples suffice to indicate where the proposed concepts allow concrete advance, and where they remain to be tested in use. My work does not claim to answer in anyone’s place, still less to close the debate. It offers a conceptual ground on which these questions can be posed with finesse, and on which the contemporary approaches I mentioned earlier, whether of Alombert, Alizart, Nurock, Tisseron, Corteel, Ganascia, Garcia, Clark, or others, can meet for what each has to bring, without needing to cancel one another out by choosing among themselves.

Thinking at the age of artificial intelligences is creating this plane. Upon it, inherited concepts can be displaced and new concepts can appear ; without it, one has only opinions clashing against the backdrop of a rapid transformation, without the ground that would let them become a debate.