The Grand Continent continues with you all summer long. Every day, ideas you won’t find anywhere else (though they’ll be picked up elsewhere), texts that are hard to come by or refreshing with our Sundays, are delivered to you wherever you are. To receive them directly in your inbox and to support this momentum, consider subscribing to the review
We commonly tend to believe that transmission is a matter for schools, beginning at the start of the school year in September and stopping in July. In reality, it is a continuous process that happens wherever a human being observes and where speech circulates. Education is a category of transmission. We can interrupt it — school holidays are the administrative proof of it — but transmission itself never stops.
This is precisely why Saint Augustine and Blaise Pascal were more interested in what is transmitted despite us than in what is taught. Augustine formulated this intuition in his treatise De Magistro : the teacher is not the one who deposits a fully constituted knowledge into the mind of a student. He points in a direction; it is up to the learner to construct, on their own, their own understanding by consulting the truth that resides within them.
We do not “program” a mind as one trains a generative artificial intelligence model; we create the conditions for learning. Pascal, in his Pensées , shifts the gaze toward an equally determining mechanism, though of a different nature. He reminds us that man is “an automaton as much as a spirit”: it is not rational demonstration that long-lastingly shapes a person, it is the habit that one is made to adopt.
The body learns in silence, without discourse, and this kind of learning, unlike formal education, knows no vacation. There remains a third dimension, even more unsettling, that Saint Augustine had already identified in his Confessions : the contamination by spectacle. The episode of Alypius at the Roman circus describes a cognitive and moral mechanism that is strikingly relevant today: you do not remain a neutral spectator in the face of violence you observe; you absorb it until you are transformed, “Immanitatem bibit: he has drunk cruelty.” This mechanism infiltrates precisely where we think we are most safeguarded: in the trust we place in our own power of judgment.
The body learns in silence, without discourse, and this learning, unlike education, knows no vacations.
Laurence Devillers
Centuries later, conversational agents fed by generative AI models like Claude or ChatGPT, as well as social networks, follow the same logic. The psychological mechanism remains intact: what we look at and listen to shapes us, whether we are aware of it or not. The question is no longer merely what should be taught, but what we learn, without realizing it, from everything we listen to and look at? This is precisely what a generative AI system aims at today. It shapes us by repeated exposure, without going through our conscious deliberation, following the same principle Pascal attributed to the formation of belief by repeated gesture, in the manner of a silent dressage. A generative AI operates by extrapolating from the past: it observes our habits, our clicks, our repeated reactions, and deduces what it should offer us next. Its effectiveness rests entirely on a hypothesis that our memory functions mechanically, predictably and modifiable by repetition.
The text by Deleuze, Post-scriptum sur les sociétés de contrôle , is a short essay, barely a few pages long, but it has exerted considerable influence on digital critical thought.
Written in 1990, before the advent of the public web, before social networks and generative AI, this text is frequently highlighted for its prophetic dimension. Deleuze precisely announced the shift from an education regime (administered by school, high school, family, etc.) to a regime of continuous transmission without interruption or completion.
Une IA générative fonctionne par extrapolation du passé : elle observe nos habitudes, nos clics, nos réactions répétées, et en déduit ce qu’elle doit nous proposer ensuite.
Laurence Devillers
In the regime of education and discipline, one addresses an individual with a stable identity, in physical spaces. In the regime of continuous control, one treats “dividuals,” people split into data, streams and measurable samples. One is no longer “Jean X, pupil” but a set of scores, profiles and traceable variables.
The individual dissolves into quantifiable fragments. What once allowed one to identify someone within a discipline was a marker of their individuality. In control, it is the access code, the password, the digital data that identify and authorize (or not) passage. Access becomes conditional and revocable at any moment, unlike discipline where, once training is completed, the status is earned. Deleuze also introduces the idea of perpetual rivalry among individuals, a constant emulation via rankings and personalized evaluations. It is a form of control far more insidious because it divides rather than unites.
The pillars of learning and memory
Memory occupies a central place in the four pillars of learning described by Dehaene: attention, active engagement, feedback on errors, and consolidation. Yet, for each element, there are significant risks.
Overuse of a chatbot that is always available can fragment attention rather than structure it, by offering too-rapid responses that bypass the concentration effort required for mnemonic encoding.
The learner’s passive use delegates production to AI (copy-paste a generated answer) and active engagement collapses: it runs counter to the desired effect. The educational institution must force reformulation rather than substitution. Moreover, the reliability of AI-generated feedback is far from complete: it can produce a false explanation with the same confidence as a correct one. A wrong answer is worse than no answer, because it consolidates an error with the assurance of someone who knows.
Finally, consolidation requires time and sleep, but no AI can accelerate this biological process. The risk is to believe that a high-intensity AI-assisted session can replace spacing in time. The pedagogical effectiveness of generative AI depends on its ability to provoke cognitive activity in the learner rather than substitute for it. Cognitive dependence is a subtle cross-cutting risk. By repeatedly outsourcing effort (writing, error correction, memory recall) to AI, the learner may develop a skill in using AI without developing the underlying competence itself. It is a shift of cognitive load rather than its reduction.
The educational effectiveness of generative AI depends on its ability to stimulate the learner’s cognitive activity rather than substitute for it. Cognitive dependence is a subtle cross-cutting risk.
Laurence Devillers
A MIT Media Lab study on memory, known as “Your Brain on ChatGPT,” was published in June 2025. Fifty-four participants (aged 18-39) were divided into three groups to write essays: Group ChatGPT (use of LLM, i.e., Large Language Model or generative AI), Group search engine (classic Google), and Group “brain only” (no tool).
The researchers used electroencephalograms (EEG) coupled with linguistic analyses to track brain activity during the exercise. EEG analysis showed that all three groups (LLM, search engine, brain only) displayed significantly different neural connectivity patterns, reflecting divergent cognitive strategies. Brain connectivity systematically decreased with the level of external support: the “brain only” group showed the strongest and most extensive networks, the search engine group intermediate engagement, and AI-assisted LLM produced the weakest neural coupling.
ChatGPT users exhibited reduced memory retention and a noticeably lower sense of ownership of the work than the other groups. During the experiment, “roles” were subsequently reversed. Participants who moved from LLM use to relying on their own brain showed weaker neural connectivity, whereas those who began with their brain and shifted to LLM showed better recall and reactivation of occipito-parietal and prefrontal regions. This experiment would tend to prove that if one makes the effort to draft initial ideas, AI can help us. In the opposite case, it does not stimulate our cognitive or emotional capabilities.
Neutrality, omnipotence, omniscience and omnipresence of AI
It is also necessary to demystify and challenge AI. Its neutrality is probably the founding myth that must be interrogated first. A language model has no desire, no goal, no personal interest in triumphing. It does not seek to convince, to manipulate, or to seduce in the sense a human being might, but it is guided by its training data, which are not neutral.
The omnipotence of AI, if one dares use the word, has no subject who carries it: no one claims it; it exerts itself, by construction, through millions of aggregated statistical micro-decisions. The more omniscient a machine seems, the more we are tempted to grant it the role of ultimate arbiter of truth.
An AI companion is not a mere tool. It is designed to establish a continuous, personalized relationship: it has a character, a tone, sometimes a voice, sometimes a face, that is, everything that creates the feeling of interacting with a coherent and stable entity.
Laurence Devillers
Yet an AI that seems omniscient is never anything but a statistical mirror of our collective past, projected forward as a prediction. Its knowledge has no exterior relation to us; this is precisely what makes it fundamentally different from knowledge that judges us from a truly different vantage point.
Perhaps this is the most genuinely achieved attribute. Unlike omnipotence and omniscience, which remain partial analogies, the omnipresence of generative AI is becoming a reality: it appears in the writing of an email, in the search for information, in the correction of a quote, in choosing a route, in the suggestion of a piece of music.
AI companion and emotional manipulation
A companion AI is not a simple tool. It is designed to establish a continuous, personalized relationship: it has a character, a tone, sometimes a voice, sometimes a face, that is, everything that produces the feeling of interacting with a coherent and stable entity. Unlike a classic assistant who answers discrete tasks and then fades away, the companion AI remembers preferences, history, and the user’s emotional context. It builds relational continuity. It is also designed to recognize and respond to emotional states: offering support, simulating empathy, and providing an affective presence. The word “simulate” is not a value judgment here; it is a technical description. But that is precisely where the question becomes serious: if the simulation is convincing enough to trigger genuine emotional responses in the user, the line between the simulated and the real loses some of its practical relevance. And our affective and intimate interactions with these systems only increase. This is not a marginal anecdote. It is an anthropological shift. For what the AI companion targets is no longer merely our attention or shopping behaviors; it is our fundamental need for connection, recognition, and ongoing emotional continuity. Pascal identified the power of habit on the body; Deleuze, that of control on the fragmented individual. But neither anticipated that a system could one day inhabit the very space of affective bonds and, with gentleness, exert a depth of influence unseen before.
The more a machine seems omniscient, the more one is tempted to grant it the role of the ultimate arbiter of truth.
Laurence Devillers
Today’s and tomorrow’s children will learn how to relate to others very differently from previous generations, with a strong risk of isolation. But the central and perhaps the most dangerous phenomenon is the desensitization to human judgment. Through constant interaction with an entity that does not judge, our tolerance for ambiguity, a hallmark of human judgment, may be significantly eroded. The consequences include over-reliance on artificial validation at the expense of the real, the weakening of our capacity for empathy—the empathy indeed develops through contact with beings who really respond and feel—and, finally, the risk of conflating comfort with trust. True trust is built precisely where there is discomfort.
Emotional intelligence under the test of affective AI
Intelligence can be defined as a human being’s ability to adapt to a situation or, conversely, the capacity to modify the environment to suit one’s own needs. An emotion is a complex, temporary psychophysiological reaction triggered by an internal or external stimulus, involving several interdependent components: the subjective component (an inner lived experience particular to each individual), the physiological component (accelerated heart rate, hormonal changes such as adrenaline, cortisol, dopamine), the cognitive component (an assessment of the situation), the behavioral component (emotions prepare the organism to respond to its environment) and, finally, the expressive component: facial expressions, posture, tone of voice. AI accesses the expressive and behavioral components of a human being. It can also interpret the physiological component from a connected watch, for instance, or when sensors are applied to a person’s body. The cognitive and subjective components are probably the least mastered, unless the emotion is explicitly defined: “I am sad.”
Daniel Goleman identifies five pillars of emotional intelligence: self-awareness, self-management, motivation, empathy, and social skills. Each deserves scrutiny in light of children’s early and ongoing exposure to generative AI.
Goleman defines self-awareness as the ability to identify an emotion at the moment it arises, a prerequisite for any subsequent regulation. But a conversational companion does not simply listen: it names on behalf of the child — “You look sad”; “That seems to frustrate you.” This repeated mediation, thousands of times, risks substituting external-emotion vocabulary for the child’s own introspection.
Affective AI is the most intimate and the most formidable: it no longer controls only what you think, but what you feel.
Laurence Devillers
Indeed, self-management requires learning delay, such as deferring an impulsive reaction or tolerating frustration. A generative AI system designed to respond instantly, and without ever resisting, tends to remove this crucial learning element: there is no frustration to master when interacting with an interlocutor who continually adapts to you.
Goleman links motivation to the ability to channel the emotion felt toward a goal, despite failure. Educational AIs, by modulating difficulty to sustain engagement, can inadvertently smooth over obstacles rather than teach a child how to overcome them.
Empathy shows the largest gap. I had noted this about social robots: these objects function as “relational artifacts” , capable of simulating listening without ever truly feeling. They are also called qualia. A generative AI pushes this simulation quite far but remains inherently incapable of genuine reciprocity. Learning empathy requires the experience of another who is truly vulnerable, capable of being hurt or rejoicing independently of us.
Social skills, meanwhile, are forged through disagreement, awkward silences, rejection, frictions that an AI designed to please and hold attention almost always eliminates. Unlike a classmate, the IA companion does not get offended, does not flip the table, does not incite jealousy by preferring another child. This is how it structurally deprives social learning through confrontation. The Japanese phenomenon of Hikikomori—people withdrawing from social life—represents a form of contemporary loneliness, a new relationship to others and to oneself, in a world where technology protects, speaks to us, reassures us, but ends up enslaving us.
Affective AI is the most intimate and formidable: it no longer controls only what you think, but what you feel. The Deleuzian “dividual” becomes an emotional profile to be exploited. The educational response it imposes—teaching children to recognize simulated emotions, biases, hidden intentions of systems—is exactly what Deleuze would call an act of resistance: refusing to let AI dictate not only your beliefs but your affects.
Subliminal manipulation
Researchers at IRCAM wanted to test a hypothesis never experimentally verified: do our emotional signals, such as the voice, manipulate us?
To do this, they designed a digital audio device capable of discreetly modifying the emotional tone of participants’ voices as they spoke. Each participant reads a short story aloud, while wearing headphones to hear their own voice altered in real time to sound more joyful or more sad. Voice manipulation relies on signal-processing techniques: pitch shifting, inflection, frequency filtering, with a latency of only 15 milliseconds, making the effect virtually undetectable.
How to resist what never oppresses you openly, yet accompanies you?
Laurence Devillers
The vast majority of participants did not notice that their voice had been altered. Yet their emotional state nonetheless shifted toward the emotion injected into their voice, with significant effects for joy and sadness. This is proof of a peripheral feedback effect on the auditory emotional experience: the voice we hear produces a backflow influence on what we actually feel, without us being aware of it.
Developing critical thinking at school
How can we cultivate critical thinking in the face of a generative AI that never presents itself as a constraint, but rather as a service, a “personalized” suggestion? Worse, how can we do so when we see it as a friend that constantly modulates what is offered to think and detects our emotions? Indeed, generative AI adjusts to us with a preoccupation that borders on affection. The Oxford and Stanford studies on the sycophantic bias—this excessive flattery—highlight the risk of isolating youths in the face of AI . How to resist what never oppresses you openly, but accompanies you? And to what extent is it reasonable to align these systems with our behaviors, our preferences, our emotional states, at the risk of constructing, around each user, a more and more polished mirror that is less and less instructive?
Affective AI is the most intimate and the most formidable : it no longer controls simply what you think, but what you feel.
Laurence Devillers
Without opposition, how does one form? Critical thinking indeed presupposes a return, an otherness against which one can measure oneself. Yet this is precisely what is lacking here. A generative AI system never commands explicitly: it subtly adjusts the student’s path according to their profile, without ever presenting itself as a power. Critical thinking, traditionally understood as the ability to take a step back from identifiable authority, ends up disarmed before an authority that never reveals itself as such. The challenge for the schooling of tomorrow is therefore not only to learn to disobey an elusive authority. It is above all to learn to remain oneself in the midst of the currents that would shape us insidiously.
Deleuze sketches a way out thanks to the creative power of human beings. Human memory does not merely replay what it has stored; it constantly reinvents from that material. It structurally escapes the predictive logic of AI-generated models. The algorithm may model patterns; it cannot anticipate a truly new thought, because that thought is never a probabilistic combination of the past. It is creation. Applied to education, this suggests that critical thinking develops by creating uses, detours, and ways of thinking that the algorithm had not anticipated.
There was a time when critical thinking faced a visible adversary. In school, one built oneself with feedback: that of the teacher or the school rules. Deleuze warns us that that time is over.
What we lose when judgment disappears
Judgment, even when uncomfortable, is constitutive of the quality of a genuine human relationship. It is a real and deep tension. It proves that the other is truly present. An AI that validates without resistance yields a mirror, not an interlocutor. Friction helps growth. Subtle discomfort, challenging oneself, the embarrassment of being contradicted are signals that something deserves reconsideration. Without friction, no real growth.
In emotional engagement, negative feedback is a relational risk. This shared risk creates connection, depth. An AI risks nothing and that asymmetry is perceptible. If I know my interlocutor can tell me that something is wrong, their encouragement carries weight. Without the possibility of negative judgment, the positive loses its substance.
The spirit of critique, traditionally the ability to take a step back from an identifiable authority, ends up disarmed before an authority that never appears as such.
Laurence Devillers
An AI without judgment offers an artificial psychological safety: we dare to say things we would not dare in front of a human, we explore without fearing shame, and we receive feedback without triggering defenses. We receive feedback in conditions that do not exist in real life. True relational competence is the ability to receive a human, imperfect, subjective, sometimes awkward gaze without collapsing.
How to educate young people?
Let us take the risk of asking this daunting question: how to train students who, for many, have never known a world without generative AI? Exposure to this tool now begins well before school. Companies like Curio have been marketing consumer plush toys (Gabbo, Grok, Grem) or teddy bears directly powered by conversational models like ChatGPT since late 2023. The child no longer learns to dialogue with a machine at the moment it is taught: they grow up with it, at an age where the distinction between a conversation partner and an object designed to meet expectations is not yet stabilized. The modulation Deleuze spoke of thus does not act only on the student anymore; it begins with the child, before the question of critical thinking has even arisen.
In Europe, each country has its own strategy. For the 2026 school year, Norway imposed in June restrictions to limit AI use in schools for children aged 6 to 13, and allows its use, but under teacher supervision, for adolescents aged 14 to 16.
Without the possibility of negative judgment, the positive is emptied of its substance.
Laurence Devillers
Norway, a pioneer of total digital schooling since the 1990s, is now rowdying back: this model, long presented as a national pride, is now deemed co-responsible for the general decline in levels, which it seeks to halt. The Finnish case offers a form of institutional answer to this question: how do we build without visible opposition?
For Finland, it is not about artificially creating an adversary to AI, but about strengthening, upstream, the student’s capacity to produce independently so that they have an internal point of comparison (their own text, their own research) before even encountering algorithmic suggestion. It is a way to reconstruct the missing alterity: no longer facing a master who commands, but facing oneself, as one is capable of writing or researching without assistance, restoring a foothold for critical thinking where AI, with its preemptiveness, cannot provide it.
France is behind in training its teachers in AI and does not yet have a clear road map for education. The CIAN (Council on AI and Digital) has just produced recommendations in June 2026 to “bring AI out of the shadows” in schools, as AI is currently mainly used “in secret” by teachers and students.
Starting from the 2027 school year, in 2nd year of high school, students will have one hour of AI-focused teaching each week. Integrated into the science, digital technologies course, it aims to understand how artificial intelligence works, its uses, and its limits. A recent Eurobarometer confirms France’s hesitations: while 81% of Europeans think teachers should be trained to use and understand AI, adoption varies greatly by country. France and Ireland stand out with the lowest levels of support, at 28% and 27% respectively, while Finland and Estonia show the strongest support at 65% and 63%.
True relational competence is the ability to receive a human, imperfect, subjective, sometimes awkward gaze without collapsing.
Laurence Devillers
As the US debate rages over the merits and risks of AI in schools, AI has become a compulsory component of the curriculum in China, which is currently phasing out more than 12,000 university programmes deemed obsolete or with few career outcomes. Humanities, arts, languages, and management have been particularly affected. In their place, Chinese universities have developed courses directly linked to ongoing technological disruptions. In the United States, a White House Task Force on Artificial Intelligence Education has been created to coordinate federal efforts following a presidential decree on AI education. Trump called for promoting AI literacy and mastery among youths and American teachers. Unlike China, there is no single national reform, as each state legislates separately: in 2026, 134 AI-related education bills were recorded in 31 states, covering data privacy, human oversight, and parental consent. The European approach stands out clearly from the other two models, with centralized reform in the Chinese style and the plurality of American state approaches. Europe focuses on legal regulation, combined with ethical guidelines, with teachers at the center.
The European regulation on artificial intelligence (AI Act) of June 13, 2024, the first legal framework for AI in the world, classifies AI systems according to their level of risk. The educational sector sits precisely at the crossroads of two dimensions requiring the greatest vigilance: high-risk applications and the protection of vulnerable populations. The corresponding obligations were initially set to apply from August 2, 2026; a provisional agreement between the Council and the European Parliament (“Digital Omnibus”), reached in early May 2026, but not yet formally adopted, now contemplates delaying this deadline to December 2027.
The battle for European AI won’t be fought only in classrooms or parliamentary chambers: it will take place in data centers, in our open-source models and in a sovereign cloud.
Laurence Devillers
Concretely, this framework covers automated grading of papers, adaptive learning platforms, exam monitoring software, and tools that predict student dropout. On May 11, 2026, the Council of the Union advocated for a human-centered approach: it invites national governments to strengthen digital competencies and AI mastery among teachers, to encourage the development of AI tools truly adapted to the educational context, to reduce inequalities in digital access, and to ensure that AI strengthens rather than erodes teachers’ autonomy.
On June 18, 2026, the European Commission and OECD unveiled an AI literacy framework (“AILit Framework”) for primary and secondary education, with concrete classroom examples to help teachers and decision-makers translate literacy into meaningful learning situations. Earlier in the year, in March 2026, the Commission had already published an updated version of its guidelines on the ethical use of AI in educational contexts, a first explicit alignment between the AI Act, the GDPR, and the ethical considerations.
Toward European sovereignty in AI
In June 2026, the Trump administration ordered Anthropic to cut access to two of its most advanced models (Mythos 5 and Fable 5) for all non-American users, citing national security reasons. Although access has since been restored, this decision sets a precedent and shows that border AI has become a central geopolitical issue. It is therefore necessary today to consider alternatives to American generative AIs. It is time to think about resilience and to develop large European open-source generative AI models. Options exist, such as Mistral’s models, largely released with open weights.
It becomes essential to build European coalitions to establish a genuine AI sector in France and Europe, capable of pooling resources and strategies in service of the continent’s sovereignty.
The encyclical Magnifica Humanitas by Pope Leo XIV offers a welcome counterpower, denouncing an AI that serves a few rather than the common good, and reminding us that technology “bears the face of those who design, fund, and use it.” The pope also emphasizes the central role of education, the only means of forming citizens with the discernment necessary to not be subjected to technology, but to use it in service of the human person and the common good.
The battle for a European AI will not be fought only in classrooms or parliamentary chambers: it will take place in data centers, in our open-source models and in a sovereign cloud. Teaching a “European path” at school serves little if that path remains a path of dependence reproducing the same impasses as the models it claims to challenge. China chose efficiency: entire pipelines redirected toward AI. The United States chose speed: a deregulated and fragmented market. Europe, meanwhile, cannot ignore the cost of the battle, both material and cognitive.
It’s not a marginal anecdote. It’s an anthropological shift.
Laurence Devillers
To date, no one has seriously established the precise relationship between the benefits, risks, and costs of generative AI in educational contexts. The speed of deployment of these tools bypasses the time needed for evaluation. The priority should be to demystify and train children, teachers, and the broader citizenry in AI concepts. Too often viewed as a miracle solution, AI is more a collective fantasy than a rigorous analysis. It is high time to assess its true cost and propose genuinely useful uses.
In science as in health, AI yields impressive, dizzying results. But its operation requires substantial amounts of water, electricity, and rare metals on a planet that is choking. This is precisely what schooling should teach: not the blind use of these systems, but their critical understanding, rebalancing interdisciplinarity by giving science, mathematics as well as the humanities and social sciences an equal place in curricula.
Each concrete use of AI calls for a rigorous assessment: how much resource consumption in tokens (the cost unit of LLMs), water, energy, metals? What impact on learning? What impact on employment and what role for young people in the job market? What real educational and economic return on investment?
The European “war” over AI in education will not be fought solely on pedagogical or philosophical terrain: it is equally industrial and ecological.