The Defeat of the Sons

Mechanised Education and the Powerlessness of a Generation Before Its Own Simulacrum

By Alexandre Ferran

Preliminary Postulates

Postulate I. A school produces less a body of knowledge than a certain type of human being. What it transmits on the surface (content) matters less than what it forms in depth (posture, sovereignty, the relationship to truth). This thesis is as old as Plato’s Republic, which devotes three entire books to paideia because he understood that the city depends on the soul it manufactures. Any discussion of education that confines itself to curricula misses the essential.

Postulate II. Intelligence is not the same as cognitive performance. Aristotle, in the Nicomachean Ethics (Book VI), distinguishes five dianoetic virtues: epistēmē (demonstrative science), technē (art or craft), phronēsis (practical wisdom), nous (intuition of first principles), sophia (contemplative wisdom). A school that only cultivates measurable epistēmē and useful technē is a school that amputates four-fifths of the human being’s intelligence. And that is precisely what the machine, today, does better than the human being: measurable epistēmē and useful technē.

Postulate III. An education that allows itself to be standardised is an education that allows itself to be defeated. The simulacrum is faster than what it simulates, as soon as what it simulates has consented to be reduced to a format. The multiple-choice test, the competency grid, the assessment framework: these are already machines before any machine exists. They prepare the ground for their own supersession.

Postulate IV. A people’s capacity for self-governance depends on the robustness of the education their elites leave to their children. When that education ceases to be an instrument of elevation and becomes an instrument of sorting, democracy is no more than a word. This thesis is defended by Marc Fumaroli (L’État culturel, 1991), by Pierre Manent (La Loi naturelle et les droits de l’homme, 2018), by George Steiner (Real Presences, 1989), and earlier by Tocqueville in the second volume of Democracy in America (1840), chapters 15 to 17 of Book I.

Postulate V. What is advanced in this text is advanced as hypothesis, not as certainty. The empirical data mobilised (Stanford 2025, PISA 2022, Chinese and Iranian employment statistics 2022-2025, the Research Centre of the Iranian Parliament) support the thesis without proving it. A proof would require rigorous comparative international studies that have not yet been conducted with the necessary precision, and that would need in particular to disentangle the specific effect of AI deployment from competing macroeconomic effects (sanctions, demography, sectoral structure) in each country. That is precisely the research programme sketched in the conclusion.

Luc Ferry, in a recent video address, formulates a diagnosis that has circulated widely and which he develops further in his essay IA : grand remplacement ou complémentarité ? (2025). In his analysis, the artificial intelligence revolution is a tsunami unlike previous industrial revolutions, now threatening the vast majority of jobs. Youth, in this account, is in the front line. Why? Because established workers already hold their position in the market, they are not available for hiring, and it is on new entrants that the pressure of substitution first bears. The diagnosis is sociological, economic, vaguely demographic. It has the merit of sounding the alarm.

But the figures we have before us say something that this reading does not explain. The study conducted by Erik Brynjolfsson and his team at Stanford, published in August 2025 (Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence), establishes an asymmetry that cannot be reduced to a question of market position. Over the period from late 2022 to July 2025, American workers aged 22 to 25 in the occupations most exposed to AI saw their employment fall by 13%. Over the same period, in the same occupational categories, workers over 30 saw their employment grow by 6 to 9%. This divergence is massive. It is not explained by the mere contractual stability of the latter group: if that were all, one would observe at worst stagnation, never growth. Something, in the very nature of the work produced by each group, makes the older workers not merely protected but augmented by the machine, while the younger workers are displaced by it.

The Stanford researchers themselves supply the decisive element of interpretation. What AI replaces first, they write, is codified knowledge, book-learning: formalised, codified knowledge transmissible by procedure, acquired in modern training institutions and measured in competencies. Whereas what resists, and even grows stronger under the effect of the machine, is tacit knowledge, what Michael Polanyi called in 1958 “what we know without being able to say it.” Situated judgment. Taste. A sense of timing. Background understanding that cannot be formalised. These are the qualities the machine cannot grasp, and that its experienced users bring to bear in order to orient it, correct it, and turn it into a multiplying instrument of their own intelligence.

This observation points toward a hypothesis of far greater scope than Ferry’s. If youth is being crushed and maturity augmented, the primary reason is not generational position. It is that the two groups did not receive the same education. Workers who are 35, 50, or 60 today were educated, at least in part, in a school system that still transmitted some measure of what cannot be measured. The young entrants were educated in a system that, over three or four decades, has progressively shed that portion. They were trained to be what the machine does better than they do.

That is the hypothesis we wish to examine here. We unfold it in four stages. First, we will look at what the machine says, by its mere presence, about the school that preceded it (I). Then we will attempt a genealogy of the mechanisation of minds, drawing on Foucault, Stiegler, Bourdieu, and others (II and III). We will then compare what the dissertation and the study of Greek knew how to build in a soul with what the competency grid builds there today (IV). We will then set our Western reading against what the 2025 data say about China and Iran, two countries whose trajectories illuminate ours as in a mirror (V), before concluding on what AI, paradoxically, might allow us to repair (VI).

I. The Mirror: What the Machine Says About the School That Preceded It

A machine only beats a human being on the terrain where the human being has agreed to be measured against it. As long as the race pits a man against a horse over a marathon distance, the horse wins; but the race does not pit the same things against each other if one considers the value of the effort, the meaning of the course, the memory built over time. The horse beats the man on speed, not on civilisation.

The digital machine, and in particular the large language model that runs our era, only beats the human being on what can be coded, segmented, and evaluated in discrete output. It fills in boxes because it is, in its very structure, a box-filler. It summarises because it is, by construction, a statistical compressor. It produces plausible text because its training consists of maximising plausibility. On these three operations (ticking, summarising, producing the plausible), it surpasses us: faster, more regular, infinitely less costly. It necessarily does so, because that is exactly what it is.

Now let us look at what the average Western school asks of a pupil today. It asks them to tick the right box in a multiple-choice test. It asks them to produce a five-line summary of a given text. It asks them to complete a literary analysis sheet according to a protocol. It asks them, at university, to answer “examination questions” requiring a calibrated formulation. It asks them to validate “competencies” listed in a European framework. It asks them, in medicine, to tick the boxes of a digitised patient record. It asks them, in law, to apply a case-law grid. It asks them, in recruitment agencies and support functions, to follow a “context sheet” and produce the expected deliverable.

On each of these tasks, the machine surpasses them. Not because the machine has become intelligent, but because these tasks are already the “mechanisable” portion of the human being. They were designed, over four or five decades, to be evaluable at scale, standardisable, interoperable, reproducible. They were, word for word, algorithmised in the heads of pupils, even before any machine existed to execute them. The machine simply arrived on a terrain that had, already, been waiting for it.

It is in this precise sense that AI is not an invader of an intact human city. It is the revealer of a city that had, unknowingly, long since transformed itself into a machine. The defeat of the sons is not a defeat against an external adversary. It is the exposure of what they had become before the battle.

II. The Genealogy: How a School Learned to Manufacture What the Machine Reproduces

To understand how Western schooling arrived at this point, one must be willing to look back a little. Not to conduct the archaeology of a local malfunction, but to recognise a structural movement spanning at least two centuries, which has accelerated over the last three or four decades under the effect of several converging forces.

1. The Modern Disciplinary Apparatus

Michel Foucault, in Discipline and Punish (1975), showed that the modern school, like the barracks, the factory, the hospital, and the prison, was born at the turn of the eighteenth century as an apparatus in the technical sense: an ordered set of practices designed to produce a certain type of body, capable of submitting to discipline and producing regular work. Compulsory schooling, in its French Ferry-era form (1881-1882), Prussian form (Humboldt and his successors), or Anglo-Saxon form (the American common schools), responds to a shared need: to form the citizen-soldier, the civil servant, the skilled worker that industrial nation-states require. This school already produces, from its inception, a partially standardised human being. But it retains, at least in its higher ambitions, the transmission of a humanitas (formation through the humanities, rhetoric, the essay, ancient languages) aimed, beyond immediate utility, at forming a mind capable of self-governance.

The rupture is therefore not in the establishment of modern compulsory schooling. It lies in the progressive evacuation, from the mid-twentieth century onward, of the humanitas component in favour of the apparatus component. The school ceases to be both factory and academy; it becomes pure factory.

2. Grammatisation and the Proletarianisation of Knowledge

The philosopher Bernard Stiegler, in a series of works produced between 2003 and 2019 (notably Technics and Time, States of Shock, Pharmacologie du Front National, Dans la disruption), offers a valuable theoretical framework for thinking through what happened. He calls grammatisation the process by which a body of knowledge, previously embodied in a living practice, is codified, segmented, and externalised into a technical support (writing, print, the computer program, the algorithm). Grammatisation is not harmful in itself: without it, there is no transmission, no accumulation, no civilisation. But it becomes toxic when the grammatised knowledge is no longer reappropriated by individuals, and remains external to them as a procedure they execute without understanding.

Stiegler calls this moment proletarianisation, extending the term well beyond its initial Marxian sense. The proletarian, in Stiegler’s sense, is not only the worker dispossessed of the means of production. It is every human being dispossessed of savoir-faire, of savoir-vivre, and ultimately of savoir penser (practical know-how, the art of living, the capacity for thought), because these forms of knowledge have been inscribed in machines that execute them in their place. The doctor who does nothing but tick the boxes of hospital software is a proletarianised doctor. The lawyer who does nothing but apply an automated case-law grid is a proletarianised lawyer. The teacher who does nothing but roll out a competency framework validated from above is a proletarianised teacher. And the pupil who does nothing but tick boxes and fill in grids is, from school onward, a proletarian of the mind.

The arrival of generative AI does not invent this proletarianisation. It makes it visible by completing it. Where yesterday’s machine executed fixed procedures, today’s machine executes the flexible, contextual, linguistic procedure. It takes the last portion of knowledge that classical grammatisation still left to the human subject: the situated adjustment of procedure to case. And it takes it precisely from those trained only in that adjustment, namely the young graduates of the mechanised school.

3. Levelling Down to the Mean

A third force, more political in character, has accelerated the movement. It has a name in French sociology over the past fifty years: democratisation from below. The idea, seemingly generous, is that a school is democratic when it reduces the gap between the best and the least able. This idea is not absurd if its aim is to raise the bottom. It becomes pernicious when it targets the top instead. And that, structurally, is what has been done. The progressive elimination of experimental preparatory classes, the weakening of excellence competitions, the dilution of curricula in literature, mathematics, and philosophy, the gradual disappearance of Greek and Latin from the majority of schools (the 2015 French secondary school reform is an emblematic illustration, and only one episode in an older and broader movement): all of this has contributed to shaving the peaks in order to bring the average closer. The result is well documented: the 2022 PISA surveys place France at its lowest ever recorded level in mathematics, and confirm a decline in reading that began in 2012.

This levelling-down, which presents itself as justice, is in reality its opposite. It deprives children from working-class backgrounds of the only lever that would allow them to access robust knowledge: a demanding education organised to bring them to the highest level. It leaves to the heirs, who already possess at home the transmitted cultural capital (Bourdieu and Passeron, The Inheritors, 1964; Reproduction in Education, Society and Culture, 1970), the monopoly on genuine humanitas. The multiple-choice test is given to those with nothing; the dissertation remains, in private schools or a handful of residual institutions, the privilege of the notable’s son. The common school, by ceasing to be demanding, has not erased reproduction. It has locked it in.

4. A Plutocracy That Conceals Itself

This analysis, if accepted, leads to a still harder hypothesis. And we insist on its status as hypothesis: we do not claim to demonstrate what follows; we propose to hold it as a working axis.

The hypothesis is this. The progressive dismantling of demanding schooling in the West does not proceed only from pedagogical errors or ideological enthusiasm. It objectively serves the interests of an oligarchy constituted by the capture of cultural capital as much as of economic capital. A population trained in argumentation, structured doubt, close reading of ancient texts, and historical perspective is a population that is difficult to govern. A population trained in MCQs, grids, and procedures is a population that obeys. The latter is useful to a financial capitalism that needs predictable consumers, interchangeable operators, and citizens incapable of standing up to a complex argument. The former would be dangerous to that order. It is no accident that pathways of excellence have grown scarcer in the public sector and concentrated in the private. It is no accident that the grandes écoles in France, and more broadly the elite pathways, function at scale as devices of social entrisme for the children of the urban haute bourgeoisie, whose share in certain cohorts is documented and overwhelming (see the work of the Observatoire des inégalités, and the recurring analyses of sociologist Annabelle Allouch).

What liberal modernity calls meritocracy functions, in practice, as a methodical plutocracy. The plutocracy no longer merely transmits wealth; it now transmits the intellectual conditions for keeping it. And it abandons to the people an educational system reduced to producing docile employees, whose very docility is exactly what AI will, tomorrow, render superfluous.

This reading is not new. It refines, for our era, an intuition found in Étienne de La Boétie’s Discourse on Voluntary Servitude (1576), in Tocqueville’s second volume of Democracy in America (1840), in Hannah Arendt’s The Crisis of Culture (1961), in Christopher Lasch’s The Revolt of the Elites (1995), in Marc Fumaroli’s L’État culturel (1991). Generative AI, by completing cognitive proletarianisation, does not invent the plutocracy. It reveals its skeleton with a clarity that past prosperity had obscured.

III. What the Dissertation Knew, What the Box Ignores

Let us reformulate the question in more directly educational terms. What did the school of former times build in a soul that today’s school no longer builds? And why does that formation protect, while training in competencies surrenders?

1. The Dissertation as an Exercise in Sovereignty

The classical French dissertation, whose settled form can be traced to the nineteenth century, is a strange exercise. A pupil is given a subject (a question, a quotation, a paradox) and asked to construct, over three or four hours, an argued development passing through three movements: a thesis, its objection or transcendence, a synthesis or reorientation. The exercise appears academic; it is in reality a training in intellectual sovereignty. It compels the pupil to posit something as true, then to contest it with the greatest possible force, then to decide, bearing the weight of the choice. It is a miniature of political judgment in Aristotle’s sense (Nicomachean Ethics, Book VI): phronēsis, which yields to neither a deductive science nor an applicable technique, but requires a subject capable of evaluating a singular situation in light of internalised principles.

The multiple-choice test, and more broadly competency-based assessment, is the point-by-point negation of this exercise. It supplies the thesis, supplies the alternatives, and asks only that one choose. It does not ask one to contradict; it does not ask one to decide. It asks one to recognise. The machine is, by construction, infinitely faster at recognising. It almost never errs on forced choices. A pupil trained exclusively on MCQs does not know how to do what the machine cannot do: hold a position, experience its opposite, emerge transformed. They know how to do what the machine does better than they do.

2. Greek and Latin as Disinterested Exercises

The retreat of Greek and Latin from Western secondary education is one of the most telling indicators of the mechanisation of schooling. These two languages, in the European pedagogical tradition, were not taught so that pupils would use them (almost no one reads Sophocles or Livy in the original in daily life), but because their study formed a certain quality of attention. To learn Latin is to submit to the patience of a complex syntax. It is to accept that a sentence may mean something other than what it appears to say at first glance. It is to train in logical rigour without calculation. It is to enter a world where meaning is never obvious, where translation is always a compromise, where perfection is never attained. It is to experience a knowledge without immediate return.

This quality of attention, which ancient languages formed, is precisely what generative AI does not possess and cannot simulate for long. The machine is performant on the surface precisely because it short-circuits patience. It delivers immediately. It does not experience meaning in the process of searching for itself. The recent reasoning models are better than their predecessors for precisely this reason: they are constrained to this reflective dimension. A pupil who has spent six years measuring themselves against Latin future participles and Greek dual forms knows, in their body, that there exist problems that cannot be resolved by clicking. They know what it is to wait. They know that the first proposed solution is almost always wrong. They have, without even formulating it, a partial immunity to being dazzled by the rapid answer. A pupil who has known only the click, information extraction, and autocomplete does not have this immunity. They take the model’s output for thought, because they have never experienced what thought demands.

3. Phronēsis and Tacit Knowledge

The Stanford 2025 study names tacit knowledge what resists the machine. This notion, as noted above, was formulated by Michael Polanyi in 1958 in Personal Knowledge. Polanyi shows that all explicit knowledge rests on a foundation of tacit knowledge, “what we know without being able to say it,” which can only be transmitted through shared practice, apprenticeship, and prolonged presence in real situations. Aristotle, twenty-four centuries earlier, had drawn the same distinction in sharper terms. Technē, art or craft, is a know-how learned through rules; phronēsis, practical wisdom, is a virtue acquired through habituation to singular situations under the guidance of a master. Phronēsis cannot be put into algorithm, because its objects are themselves singular, irreducible to a general case. That is why AI cannot reproduce it: aside from in philosophical works, these notions are not set down on paper. They are silent, therefore invisible to AI corpora. And it is precisely phronēsis that today’s school no longer takes the time to form, for lack of reproducibility and efficiency.

Pierre Hadot, in his work on antiquity (Spiritual Exercises and Ancient Philosophy, 1981; What Is Ancient Philosophy?, 1995), showed that the Ancients conceived philosophy not as a body of doctrines but as a way of life cultivated through exercises: reading, meditation, self-examination, dialogue. These exercises formed what Hadot calls a style of soul. Modern grammatisation, by replacing the exercise with the procedure and the style of soul with the competency grid, has caused to disappear precisely what the machine, by construction, cannot occupy.

4. What Simone Weil and George Steiner Saw

Simone Weil, in The Need for Roots (written in 1943, published in 1949), writes that among the fundamental needs of the human soul are “initiative and responsibility,” “truth,” and “hierarchy” understood as recognition of an order of things that surpasses pure utility. The school she criticises, already in 1943, is one that reduces the pupil to the execution of tasks without making them a free subject. She prophesies, unknowingly, the condition of the modern pupil before AI: a being dispossessed of initiative, asked only to validate outputs produced by a procedure over which they have no mastery.

George Steiner, in Real Presences (1989), argues that genuine education is an encounter with a real presence (the author, the work, the great predecessor) that addresses the pupil in their interiority and obliges them to situate themselves. The grid-based pedagogy forecloses this encounter. It does not propose a presence; it proposes a framework. The pupil is no longer required to measure themselves against Sophocles, Pascal, Hegel, Bergson. They are required to tick items on a list of capacities. The presence has disappeared. AI, which is by essence an absence (a text without author, a speech without a present speaker), merely installs itself in the void this pedagogy has hollowed out.

IV. The Old, Multiplied by the Machine: Why the Well-Educated Are Not Afraid

If the preceding thesis holds, then a corollary follows directly. Generations formed in a still-demanding school, even partially, must show an asymmetry of behaviour toward AI. They must not only resist, but seize it as a tool that multiplies their capacity. And that is exactly what the figures show.

The Stanford researchers explain the 6 to 9% growth in employment among workers over 30 in the most AI-exposed occupations through two mechanisms. The first is defensive: established workers mobilise tacit knowledge the machine cannot grasp. The second is offensive: these same workers use AI to augment their productivity on tasks where they were already competent, rather than letting it substitute for them. AI becomes their instrument where it becomes a competitor for the others.

This asymmetry is not a matter of generational luck. It is explained by a simple structure: to use a powerful tool well, one must know what to ask of it, know how to evaluate its output, and know how to correct its errors. These three operations require an internalised competence that goes beyond using the tool. A lawyer trained in the rigour of reasoning, in close reading of a judgment, in adversarial argumentation, can use an LLM to produce a first draft ten times faster and then correct it with discernment. A junior trained only in applying grids can neither judge the first draft nor correct it. They can produce the version, but they are no longer needed to do so. That is precisely why the first documented waves of redundancies in major law firms and consultancies affect associates and junior consultants, not partners.

What the machine requires, in sum, is a human being who has received an education superior to what is needed to execute the task the machine is taking over. As long as the human being remains a subject, they can orient, control, and augment. Once the human being has been reduced, by their school, to the level at which the machine can itself produce (the level of the object), they become interchangeable with it, and therefore redundant. The difference between the older worker who is augmented and the young worker who is displaced rests neither on luck nor on contractual seniority. It rests on the quality of the education each received. This is a truth that neither the labour market nor Ferry’s sociological analysis names, because naming it would indict those who allowed the school to drift.

V. Converging Tests: China, Iran, and the Two-Variable Equation

A philosophical hypothesis cannot stand alone. It calls for empirical tests. The first reflex, when defending the above thesis, is to seek in the world a country where education has not followed the Western drift, and where youth is consequently not being crushed by AI as ours is. It was in this spirit that, in an earlier version of this essay, we approached China and Iran: as counter-tests that would either validate or invalidate the thesis. The 2025 data, examined closely, require a displacement. These countries do not function as counter-tests. They function as converging tests that complete the hypothesis by showing that youth, everywhere in the modern world, is today in difficulty, but for partially distinct reasons that illuminate our diagnosis in reverse.

China: Rigour Without Humanitas Saves Nothing

The Chinese system is an instructive case precisely because it represents the inverse of French laxity. The gaokao (the national university entrance examination) is one of the most brutal selection devices in the world, organised around mass rote preparation, fierce competition, and a quantitative demand that produces every year a volume of technical talent without equivalent in the West. Chinese students have dominated international mathematics and science competitions for fifteen years. The elite universities (Tsinghua, Beijing, Fudan, Westlake) conduct STEM selection that France and the United States can no longer match. If raw scholastic effort were sufficient to protect a generation from AI, China should be the country where young graduates fare best.

Yet the 2025 data say the opposite. Youth unemployment among Chinese aged 16 to 24 (excluding students) reached a record 18.9% in August 2025, the highest since the methodological revision of late 2023; it remained at 16.5% in December. The number of new higher-education graduates reached 12.22 million in 2025, up 430,000 from 2024, with a new record forecast of 12.7 million for 2026. Graduate job postings collapsed by 22% in the first half of 2025. More than 20% of delivery workers on major platforms now hold a higher-education degree, and at least 70,000 ride-hailing drivers hold a master’s degree. On 27 January 2026, the Chinese Ministry of Human Resources announced that it was preparing official documents to respond publicly to AI’s impact on employment. Beijing is beginning to acknowledge what it preferred to keep quiet about.

This data has a significance that simple comparison of averages cannot capture. It indicates that purely quantitative scholastic rigour, cut off from the dimension of humanitas (Latin, Greek, the dissertation, close reading of ancient texts, formation in phronēsis), does not protect. On the contrary: it produces, at very large scale, exactly what the machine does better than the human being. The gaokao is a monumentalised multiple-choice test. It rigorises formalism without reintroducing tacit knowledge, situated judgment, or the style of soul. The result is that a mass of young Chinese, extremely qualified on paper, find themselves in the same relationship to AI as less rigorously selected young Westerners: interchangeable with it, and therefore replaceable by it. China does not invalidate our thesis. It completes it by showing that the humanistic quality of formation, and not its intensity alone, is what constitutes a barrier.

And the quality at issue here exceeds, the more one examines it, the strictly scholastic frame. The humanitas that France has lost, and that China never reconstituted after the Cultural Revolution, was not merely a programme of close reading and phronēsis. It carried also, beneath the surface, a connection to what cannot be calculated: what the idealist tradition names the invisible, the Principle, the sense of the Centre. A society that has made materialism its official metaphysics deprives itself of precisely what, in formation, exceeds the machine. The question surfaces here, without our claiming to resolve it: might the most completely materialised societies be the ones most rapidly replaceable by their own machines?

Iran: Humanitas Preserved but Stifled, and What Its Tech Window Reveals

Iran is a still more complex case, and it requires methodological honesty. The Persian pedagogical tradition has retained, despite the ruptures of the revolution, a strong emphasis on mathematics, theoretical physics, and classical literature. Iranian students have dominated the International Mathematical and Computer Science Olympiads for twenty years. The great Persian poetic tradition (Hafez, Saadi, Rumi, Ferdowsi), taught widely from primary school onward, maintains a living relationship to language and meaning. And this tradition is not purely literary. Hafez, Saadi, Rumi, Ferdowsi are also, and perhaps above all, the voices of a persisting spiritual idealism (Sufism, Sohrawardi’s metaphysics of light, ishrāqī gnosis) that has never consented to the evacuation of the sacred that Western modernity, in its Weberian disenchantment, has installed as self-evident. Iranian humanitas has remained, despite the revolution and the theocracy, connected to the invisible. And yet young Iranians suffer. Why?

The 2025 figures draw a paradoxical picture. Youth unemployment in Iran (ages 15-24) fell from 22.7% in 2022 to 22.64% in 2023, then to 19.4% in Q3 2024 and 20.2% in Q4, representing an apparent slight improvement over the very period in which generative AI has massively deployed worldwide. At first glance, this would suggest that young Iranians work more than before AI, contrary to the Western trajectory. But read carefully, this improvement is misleading. First, the absolute number of unemployed aged 18 to 35 increased over the same period (1.643 million in summer 2022, 2.115 million in summer 2023), meaning the rate falls less because the situation improves than because part of the youth exits the statistics (discouraged, emigrated, absorbed into the informal economy). The labour market participation rate of the population aged 15 and over fell in 2024 to 40.24%, its historical floor since 1990. Second, unemployment among women aged 20 to 24 reaches 34.9% in 2025. Third, 110,000 Iranian students are currently abroad and 70% of them do not intend to return. Iranian humanitas protects, but it protects elsewhere: in Dubai, Toronto, Paris, Berlin. The “multiplied elders” of Iran exist; they are simply no longer in Iran.

As for the specific question of whether AI produces in Iran the same age bias as in the United States (the young displaced, established workers retained in exposed occupations), no study equivalent to the Stanford one has yet been conducted, and the Statistical Center of Iran does not publish age breakdowns crossed with AI exposure. The Research Centre of the Iranian Parliament estimates that 20% of Iranian jobs will be affected by AI in the coming years, but total public investment remains below $50 million in 2025, against billions in the UAE and Saudi Arabia. AI is deployed predominantly in the judicial and security sectors (tools of control, not productivity), and in urban platforms (Snapp, Tapsi, with a substantial share of their drivers and delivery workers being higher-education graduates without prospects). The country does not yet have the mass deployment (law firms, consulting, programming, customer support) that produces, in the United States, the Stanford asymmetry.

And yet a qualitative signal in the reverse direction deserves to be noted. Where AI is densely present in Iran, that is, in Tehran’s tech ecosystem, young people are not being displaced: they are driving it. The country currently produces between 287,000 and 335,000 STEM graduates per year, making it the second-largest AI developer pool in the Middle East. 3,728 startups were active in 2025, growing at an annual rate of 14.1%, with 83% based in Tehran, for a total of $676 million raised during the year. Parliament voted in May 2025 for a National Artificial Intelligence Plan (187 votes to 33), with an initial budget of $115 million, and has scheduled the creation of a Tehran AI Park for 2027. The leading universities (Sharif, Tehran, Amirkabir) have massively shifted their programmes toward AI. Young Iranians working in these ecosystems are not being replaced by the machine; they are building it. And when they are broken, it is upstream: by US sanctions reinstated in 2018, by the closure of the international market, by the state’s security capture, by the spring 2025 war with Israel. Not by any LLM that would have replicated their job description.

The Iranian case thus paradoxically provides one of the clearest validations of the hypothesis advanced here. Where humanitas has been preserved by the school, youth, when it effectively gains access to AI, does not allow itself to be displaced: it seizes it as a tool, exactly as Western workers over thirty do in the Stanford study. The young Iranians who suffer are not killed by the machine; they are killed by the absence of the market where they could exercise what has been transmitted to them. When that market exists (the narrow window of Tehran’s tech sector, or more broadly the diasporas of Dubai, Toronto, Berlin), they themselves become the “multiplied elders” to whom the machine gives wings, and as early as twenty-five. Their resistance to displacement by the machine does not rest only on transmitted content. It rests on what, in the transmission, refused the evacuation of the sacred. The lock is not cognitive. It is strictly political.

Iran thus completes China in inverse mirror. China trains rigorously but without humanitas, and loses. Iran preserves humanitas but stifles its best before they can use it, except in the narrow window of its tech sector, where the thesis is positively borne out. Neither case contradicts the Western thesis. Both refine it, by showing that two conditions must be met for a generation to be not merely educated, but protected and multiplied by AI.

The Two-Variable Equation

Let us reformulate what these converging data suggest. For an education to produce young people capable of standing before the machine, two conditions must be simultaneously met. The first is cognitive: education must transmit a non-codifiable humanitas (ancient languages, the dissertation, close reading, tacit knowledge, phronēsis). Without it, even the gaokao, which is pure demand, collapses. The second is political: society must offer an open market where free minds can exercise themselves, that is, to found, create, contest, and produce. Without it, even Tehran loses its best to Toronto, except in that narrow window, proven by the Iranian tech ecosystem, where the thesis is positively verified: where humanitas has been received and finds a market, however restricted, young people are not displaced by AI; they become its operators.

Measured against this double criterion, France is not a particular case of unfortunate Westernism. It is a case of double failure: we have destroyed the cognitive condition (Greek, Latin, the dissertation, the long oral examination have receded everywhere in public education); and we offer a political condition compromised by the plutocratic capture of cultural and economic capital. Our young people are crushed twice over: they did not receive what resists, and the little they did receive finds nowhere to exercise itself freely.

East Asia: The First Signals of a Different Balance

Japan, South Korea, Taiwan, and Singapore share with China and Iran the emphasis on educational rigour, but differ from both on one or both axes. Japan and South Korea have preserved a humanistic dimension in their schooling (classical literature, philosophy, calligraphy, rhetorical rigour) while also offering an open market integrated into the global economy. The first available data on the effect of generative AI on graduate employment in these countries show a less univocal effect than in North America. South Korea in particular sees its young engineers augmenting their productivity through AI use rather than being substituted by it. Several years will need to pass before systematic comparative studies are available, and hasty conclusions must be avoided. But the initial signal is consistent with the equation we have formulated: where both conditions are approximately met, AI augments rather than substitutes, even for the young.

What the West Can Learn, and What It Refuses to Learn

If the reading proposed here holds, then the defeat of Western youth before AI is neither a technological destiny nor an isolated case of regional misfortune. It is the most fully realised version of a failure that takes other forms elsewhere. Everywhere in the modern world, youth is in difficulty before the machine: in China because it has been over-trained in a formalism the machine produces better; in Iran because the system stifles it before it can exercise what has been transmitted; in the West because what earlier generations received has been dismantled over forty years. No modern country currently gives its young people what the French school of the 1950s-1970s still gave to a Luc Ferry, a Bernard Stiegler, a Marc Fumaroli: an interior apparatus capable of standing up to any machine, because it was formed where the machine cannot go.

The West, which still possesses an open market and a residual humanistic tradition, is paradoxically best placed to repair this failure, provided it recognises that the failure is educational before it is technical, and political before it is pedagogical. But it may also refuse to recognise this. The slope of refusal is the more probable, because recognising this failure would require naming the responsibility of the elites who organised it. It is more comfortable to invoke the fatality of the tsunami than to question the ministries, inspectorates, educational publishers, and assessment bodies that made the MCQ and the grid the common standard. But the comfort of invocation protects no one. Young people continue to be crushed, while the old, through their former education, are saved. And China, Iran, and all the others are paying their particular version of the same price.

VI. The Way Out: Restoring a School That Forms the Irreplaceable

A critique without a proposal remains sterile. If the hypothesis defended here is correct, then the way out is clear, even if demanding. It requires restoring to Western schooling what forty years of mechanisation have made disappear. It also requires using AI itself, paradoxically, as one of the levers of this restoration. Here are some orientations, offered as invitation rather than prescription.

1. Restore the Essay Exam, the Long Oral, Greek, and Latin

These exercises are not a nostalgic luxury. They form the only type of mind the machine cannot reproduce: a subject capable of positing a thesis, experiencing its opposite, and emerging transformed. They are, properly speaking, the opposite of the multiple-choice test. Their return to the public curriculum (not only in a handful of prestigious schools) is a measure of protection for working-class children, not a bourgeois privilege.

2. Reintroduce Sustained Attention and Effortful Study Without Immediate Return

The study of ancient languages, but also classical music, life drawing, endurance sports, and demanding manual crafts, share a common virtue: they form an attention that does not dissolve at the first available output. This attention is precisely what the machine, by construction, does not have. Cultivating it in our children is a political act as much as a pedagogical one.

3. Make AI a Demanding Tutor, Not an Easy Answer Machine

This is the direction we are exploring concretely at Eiffel AI with the Aristote project (see our essay Three Prophecies, One Blind Spot, published on 4 May 2026). A well-designed AI agent does not give the answer; it poses the next question. It obliges the pupil to formulate their thought, argue it, defend it, correct it. It does what an ancient preceptor did, at the individual level: it keeps the pupil in the posture of a subject who thinks, not a user who consumes. This path, which we call augmented Socratic pedagogy, is the pedagogical opposite of the current use of LLMs as lazy oracles. It requires an explicit pedagogical intention and a careful design of the agent. It is, in our view, one of the very rare opportunities AI offers to education, on condition that it not be confused with its default use.

4. Recognise the Political Nature of the Struggle

It must be said plainly: restoring a demanding school is a political struggle, because the mechanisation of schooling serves specific political interests. This struggle cannot be led by pedagogues alone. It requires a collective awareness of what has been lost, what has been taken away, and those to whose benefit it was taken. A democracy that lastingly reserves serious culture for the heirs alone is a democracy condemning itself to become a polite oligarchy. AI, by completing proletarianisation, places this truth before our eyes. It does what neither Bourdieu, nor Stiegler, nor Steiner had managed to impose on public debate: it makes the defeat visible, because it costs us.

Conclusion: The Mirror and the Chance

Luc Ferry saw a tsunami. He did not see that the coastline on which it crashes had been flattened, over forty years, so that no dune remained standing. Western youth is not struck because it is young; it is struck because we trained it to be what the machine does better than it does. Earlier generations are not spared because they are established; they are spared because they received, in a school that no longer exists, interior instruments the machine cannot reproduce.

This truth, if we accept it, changes everything. It shifts the debate on AI from the technological to the anthropological terrain. It obliges us to recognise that defence against the machine does not pass first through regulation, prohibition, taxation, or universal income. It passes through a collective decision about what we want our children to become. If we want them to become interchangeable operators, then let us continue the current school, and prepare ourselves to see their entire generation devoured by an intelligence they have been trained to imitate. If we want them to become subjects, then let us restore the exercises, the presences, the demands that form subjects. The machine cannot do this work for us. The machine can only make the result of our choice more visible.

There is, in the current situation, a chance that the grid of defeat conceals. The chance is that, for the first time in four decades, the evidence can no longer be evaded. The figures are there. Young graduates are looking for work in a market that no longer needs them in the functions for which they were trained. Juniors are disappearing from large firms. Tertiary support functions are melting away. No ostrich policy can long conceal this reality. The moment has come, if not to repair, at least to establish the diagnosis. AI, in this story, is not the enemy. It is the severe but necessary revealer of a failure that was there before it.

We do not know whether Western society will be able to seize this chance. The forces that profited from the dismantling of demanding schooling will not relinquish their position without resistance. The pedagogies that institutionalised the MCQ and the competency grid today have a considerable administrative apparatus to defend. Educational publishers, assessment bodies, ministries, and inspectorates all have an interest in maintaining the existing order. But at a certain point, the pressure of evidence becomes such that no apparatus can contain it. We believe that moment is approaching. We believe the decade ahead will be, in the history of Western education, either the decade of a recovery or that of a definitive consecration of cognitive plutocracy.

One final hypothesis, in closing, that exceeds the educational frame alone. If earlier generations are today multiplied by AI, it is because they received, without knowing and without willing it, a patrimony that now serves them. That patrimony is not only scholastic; it is cultural, familial, social, and sometimes religious. It is what Simone Weil called rootedness. A rootless youth, with no access to the depth of its language, the memory of its predecessors, sustained attention, or the patience of meaning, is a youth disarmed before any power that presents itself, and singularly before a power that speaks. The defeat of the sons is not only the defeat of schoolchildren; it is the defeat of a civilisation that has forgotten what it means to raise someone.

No one can make this decision for us. No one can restore, on our behalf, what has been dismantled. But upon those who have received, it now falls to transmit, against the spirit of the age if necessary, what they received. That is doubtless the only human way of responding to what the machine places before our eyes.

The mirror is cruel. It makes visible a defeat we had chosen not to see. But the mirror is not the enemy. The enemy is what it reveals. And what it reveals, contrary to what the prophets of the tsunami say, is not a destiny. It is a choice that remains ours to reclaim.

Sources and References

Empirical Data Mobilised

Brynjolfsson, E., Chandar, B., Chen, R. (Stanford Digital Economy Lab, 2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Study analysing ADP employment data by age and AI exposure, late 2022 to July 2025.
OECD (2023). PISA 2022 Results, Volumes I and II, Country Notes: France. Lowest levels ever recorded in mathematics; decline in reading beginning in 2012.
Council of Europe and OECD, successive reports 2015-2024 on developments in the baccalauréat and secondary curricula in France and Europe.
China Ministry of Education, annual statistics on higher education and the gaokao, 2020-2026. 2025 graduates: 12.22 million (+430,000 vs. 2024); 2026 forecast: 12.7 million.
National Bureau of Statistics of China, Youth Unemployment Survey (methodology revised December 2023). Rate for ages 16-24 excluding students: 18.9% in August 2025, 17.7% in September, 16.9% in November, 16.5% in December.
Asia Society Policy Institute (2025). The 19 Percent Revisited: How Youth Unemployment Has Changed Chinese Society.
The Wire China (January 2026). China’s Labor Market Braces for an AI Shock. Notes over 70,000 ride-hailing drivers with master’s degrees and more than 20% of platform delivery workers holding higher-education degrees.
Shen et al. (May 2025). Generative AI, Perceived Job Displacement and Political Preferences in China. SSRN. Chinese experimental survey showing that young graduates are the most anxious about AI substitution.
Foreign Policy (November 2025). China’s AI Planners Are Fearful of Job Losses. Official recognition by Beijing of the problem; ministerial announcement of 27 January 2026.
Research Centre of the Iranian Parliament (2025). Estimate: 20% of Iranian jobs to be affected by AI in the coming years.
World Bank, Iran Youth Unemployment 1991-2025. Trajectory: 22.7% in 2022, 22.64% in 2023, 19.4% in Q3 2024, 20.2% in Q4 2024, approximately 20.1% in 2025 (urban areas), 15.9% (rural areas). Unemployment among women aged 20-24: 34.9% in 2025.
WNCRI (May 2025). Employment Crisis: Jobless Rate for Young Women Nears 35%. Age and gender breakdown of the Iranian labour market.
Iran Focus (2025). Iran Unemployment Rate Among University Graduates. 10.7% unemployment among higher-education graduates in winter 2025; over 40% of Iranian unemployed hold higher-education degrees.
Filterwatch (September 2025). Control Over Innovation: Iran’s Paradoxical AI Development. Iranian public AI investment below $50M; judicial and security deployment prioritised; authoritarian capture of AI.
Ts2.tech (June 2025). Artificial Intelligence in Iran: Recent Developments and Outlook. National AI Plan approved May 2025 (187 votes to 33); $115M R&D budget; planned 50,000-training programme judged insufficient.
Nikou, I. (2025). Riders of the Storm: The Rise of Snapp! and Workers Struggle in Iran. SSRN. On the absorption of Iranian graduate youth by platforms (Snapp, Tapsi, Digikala).
Ts2.tech / Cambridge Iranian Studies (2024-2025). Artificial Intelligence in Iran: National Narratives and Material Realities. Iranian diaspora: 110,000 students abroad, 70% not planning to return.

Authors and Works Cited

Aristotle, Nicomachean Ethics, Book VI. Distinction of the five dianoetic virtues, in particular epistēmē, technē, phronēsis.
Plato, Republic, Books III and VII. Theory of paideia and the political function of education.
Étienne de La Boétie, Discourse on Voluntary Servitude, 1576.
Alexis de Tocqueville, Democracy in America, Volume II, 1840.
Simone Weil, The Need for Roots, written 1943, published 1949.
Hannah Arendt, The Crisis of Culture, 1961 (in particular the essay The Crisis in Education).
Pierre Bourdieu and Jean-Claude Passeron, The Inheritors (Les Héritiers), 1964; Reproduction in Education, Society and Culture (La Reproduction), 1970.
Michael Polanyi, Personal Knowledge, 1958. Notion of tacit knowledge.
Michel Foucault, Discipline and Punish (Surveiller et punir), 1975.
Pierre Hadot, Spiritual Exercises and Ancient Philosophy (Exercices spirituels et philosophie antique), 1981; What Is Ancient Philosophy? (Qu’est-ce que la philosophie antique ?), 1995.
George Steiner, Real Presences, 1989.
Marc Fumaroli, L’État culturel, 1991.
Christopher Lasch, The Revolt of the Elites, 1995.
Bernard Stiegler, Technics and Time (La Technique et le temps), 3 vols., 1994-2001; States of Shock (États de choc), 2012; Pharmacologie du Front National, 2013; Dans la disruption, 2016.
Ivan Illich, Deschooling Society (Une société sans école), 1971. Radical critique of schooling as the mechanisation of learning.
Luc Ferry, IA : grand remplacement ou complémentarité ?, 2025. Position discussed in this essay.

For Further Reading

Annabelle Allouch, La société du concours, 2017. On social reproduction through elite pathways.
Stefan Collini, What Are Universities For?, 2012.
Pierre Manent, La Loi naturelle et les droits de l’homme, 2018.
Observatoire des inégalités, annual reports on the social composition of the French grandes écoles.
Eiffel AI, Three Prophecies, One Blind Spot, essay published 4 May 2026 (eiffel-ai.io/fr/journal/trois-prophetes-un-miroir).

This article was written by Alexandre Ferran, founder of Galaad and co-founder of Eiffel AI. The theses advanced are those of the author alone and are offered in a spirit of open inquiry, not of certainty. All hypotheses formulated here, in particular those concerning the political responsibility for the dismantling of demanding schooling and the international counter-tests, call for contradictory discussion and empirical studies more refined than current public documentation allows.

Paris - Bagnères-de-Bigorre, 20 May 2026