A Journalist Notices Something
Good journalism often performs an unusual service. It notices historical change before anyone has found the concepts to describe it. This is rarely because journalists are better theorists than philosophers or historians. Quite the opposite. They arrive earlier because they are not searching for theories at all. Instead, they pay attention to ordinary conversations, recurring frustrations and subtle changes in people’s habits. They notice that familiar activities begin to feel different long before anyone can explain why, and only later do those observations acquire broader historical significance.
That was my reaction when I read Arnaud Leparmentier’s recent reporting from Silicon Valley for Le Monde. On the surface, his article concerns familiar themes: artificial intelligence, the changing labour market, the difficulties facing young graduates and the uncertainty experienced by software engineers wondering how long their expertise will retain its value. Companies reorganising around increasingly capable AI systems have become almost routine subjects of discussion. Yet one detail in the article stayed with me long after I had finished reading.
Several engineers described subtle changes in the way they approached difficult problems. They found themselves becoming stuck less often than they once had. When they reached an impasse, they were less likely to spend hours wrestling with it alone or to interrupt a colleague for another perspective. Instead, they reformulated a prompt, consulted another model or waited for the next iteration of the software. None of them presented this as a loss. On the contrary, they regarded it as an entirely rational adaptation to more capable tools. If a machine could help solve in minutes what previously demanded an afternoon of work, why insist on preserving the longer route?
Nothing about those remarks appears especially profound. Had I encountered them in isolation, I might have passed over them without a second thought. Taken together, however, they suggested that something more interesting was taking place. The engineers were not merely describing a faster way of working. They were describing a different relationship with intellectual difficulty itself.
My first instinct was to interpret this simply as another story about productivity. Every major technology reduces the effort required to perform certain tasks. Calculators transformed arithmetic, search engines changed access to information and GPS altered navigation. Artificial intelligence seemed merely to continue a familiar historical pattern in which the distance between a question and an acceptable answer steadily contracts.
The more I reflected on Leparmentier’s reporting, however, the less convincing that explanation became. Productivity measures how quickly we arrive at an answer, but it tells us remarkably little about the experience of getting there. What struck me was not that the engineers were finding solutions more quickly, but that they were spending less time inhabiting uncertainty itself. They encountered fewer moments in which their own understanding had genuinely reached its limits, and they relied less often on another person’s perspective when it did. Gradually, I realised that what interested me was no longer the answer at the end of the process but the process that seemed quietly to be disappearing.
Technology has always changed the way we solve problems. What artificial intelligence may be changing, however, is the way we experience having one. At first, that distinction seemed almost semantic, yet the more examples I considered, the harder it became to dismiss. Throughout most of human history, intellectual work possessed a certain stubbornness. A scientist whose experiment failed could not immediately ask reality for an alternative explanation. A lawyer confronting a novel case had relatively few ways of escaping uncertainty. An architect facing an unexpected structural problem could not instantly generate dozens of plausible alternatives. Progress depended upon remaining with a difficulty long enough for something to change—sometimes the evidence, sometimes the question itself, but often the person trying to understand it.
No educational institution invented that rhythm, and no profession consciously designed it. It simply reflected the way the world answered our questions: slowly, incompletely and often only after our first explanations had proved inadequate. As I continued thinking about Leparmentier’s article, I found myself wondering whether the most significant consequence of artificial intelligence might not be that it changes our answers, but that it changes our encounters with reality itself.
I had not intended to think about education. Indeed, I was not thinking about education at all. Not yet.
The Lessons Reality Used to Give Us
The more I reflected on this, the more I realised that what fascinated me was not artificial intelligence itself but the kind of world within which previous generations had learned to think. Reality has always resisted us in ways that education never had to invent. A student struggling to understand a mathematical proof, a physician confronted with contradictory symptoms or an engineer unable to explain an unexpected result were not participating in carefully designed learning experiences. They were simply encountering a world that refused to yield its answers on demand.
That resistance shaped the rhythm of intellectual life. Before arriving at a solution, we often had to exhaust our own understanding, pursue explanations that ultimately proved inadequate and reformulate the question because the original one no longer made sense. Eventually, we turned to another person, not because collaboration was inherently virtuous, but because we had reached the limits of what we could achieve alone. Looking back, it is tempting to describe these episodes as educational, yet they rarely felt that way at the time. They felt frustrating, inefficient and occasionally humiliating. Their educational value became visible only in retrospect, when we recognised that what had mattered was not the failure itself but the sequence of experiences it had made unavoidable.
Those experiences shared a common characteristic. They forced us to discover that our first explanation was incomplete, to distinguish confidence from understanding and to recognise when another person’s perspective contained something we could not yet see ourselves. None of these lessons had been deliberately taught. They emerged because reality imposed conditions that no teacher had designed and no curriculum had specified.
It was at this point that I realised I had been asking the wrong question. Until then, I had been wondering what artificial intelligence might change about learning. Gradually, however, another question displaced it. What if many of the experiences through which human beings have learned to think were never educational inventions in the first place?
The possibility was unsettling because it inverted one of our most familiar assumptions. We tend to think of education as the institution responsible for intellectual formation. Schools design curricula, universities organise courses and teachers create learning activities. Formation therefore appears to flow from intentional educational design. Yet the engineers in Leparmentier’s article suggested a different possibility: perhaps educational institutions have always depended upon something they did not themselves create.
Consider a student who finally walks into a professor’s office after struggling with an idea for several days. The educational significance of that conversation does not arise because asking for help was written into the syllabus. It arises because the student first encountered a difficulty that could not easily be bypassed. The conversation becomes transformative precisely because reality had already done part of the educational work before the teacher entered the picture.
The same pattern appears well beyond classrooms. Scientists often remember the experiment that refused to behave as theory predicted, lawyers the case that forced them to abandon an elegant argument, doctors the patient who did not fit any familiar diagnosis and architects the project that exposed a flaw in assumptions they had scarcely realised they were making. Years later, these moments frequently stand out as turning points in professional formation, even though none of them had been designed with education in mind. Their formative power derived from the fact that reality insisted on resisting human expectations.
This, I began to suspect, is what artificial intelligence may be changing. The technology does not abolish uncertainty; anyone who works seriously with today’s systems knows that they generate new forms of ambiguity of their own. Models hallucinate, disagree with one another and require careful interpretation. Yet they also alter something more fundamental. They reduce the frequency with which reality itself obliges us to remain inside uncertainty before offering a possible way forward. There is almost always another prompt to try, another model to consult or another explanation to consider before we have fully exhausted our own understanding.
The point is not that these answers are necessarily correct. It is that the experience of moving from not knowing to knowing begins to change. For centuries that interval was largely dictated by the world itself. Increasingly, it is mediated by technologies designed to shorten it.
When I reached this point in my own thinking, I realised that Leparmentier’s article was no longer simply about software engineers. Without intending to do so, it had led me towards a much older question: whether education had ever fully understood where some of its deepest formative experiences actually came from.
When Reality Changes
If many of the experiences through which judgement developed emerged because reality itself imposed them, then artificial intelligence raises a question that extends well beyond education. The issue is no longer simply whether students will rely too heavily on AI or whether professionals will delegate too much cognitive work to increasingly capable systems. The deeper question is what happens when the environment itself ceases to generate, with the same frequency, the experiences through which human beings have traditionally learned to think.
This is not an argument against technological progress. Every significant technological advance has removed forms of difficulty that previous generations simply accepted as part of ordinary life. We no longer calculate logarithms by hand, spend days searching library catalogues or navigate unfamiliar cities with paper maps, and few would wish to recover those conditions merely because they demanded greater effort. The point is not that difficulty is inherently valuable, but that some forms of difficulty proved unintentionally formative. Until recently, we rarely needed to distinguish between obstacles that merely slowed us down and obstacles that quietly shaped our judgement because both disappeared together. Artificial intelligence forces us to recognise that they are not necessarily the same thing.
The distinction becomes clearer if we consider the difference between obtaining an answer and arriving at an understanding. An intelligent system may provide an elegant explanation, propose alternative interpretations and even anticipate objections that its user has not yet considered. In many circumstances this is an extraordinary intellectual resource. Yet it also changes the sequence through which understanding develops. Instead of discovering the limits of our own reasoning before encountering a better one, we increasingly encounter sophisticated reasoning before we have fully explored where our own thinking might have failed.
Education has always depended upon that sequence. Intellectual maturity is rarely produced by exposure to correct answers alone. It develops through the gradual recognition that apparently convincing explanations conceal unnoticed assumptions, that problems first thought to be simple demand different questions, and that genuine understanding often begins with the recognition that one’s own interpretation is inadequate. Such experiences are remarkably difficult to design artificially because they depend upon authentic encounters with resistance rather than the mere transmission of information.
Seen from this perspective, the challenge posed by artificial intelligence is not primarily one of knowledge but of intellectual formation. Knowledge can often be transferred with remarkable efficiency; formation cannot. It develops through repeated encounters with uncertainty, revision and responsibility, experiences that educational institutions undoubtedly cultivate but that they have never supplied on their own. For centuries, schools and universities operated within a world that continuously generated these formative encounters for them. The classroom did not exist in isolation from reality; it depended upon the epistemic conditions of the wider world.
This is why so many contemporary debates about AI in education feel oddly incomplete. We devote considerable attention to assessment, academic integrity, personalised tutoring and future skills, all of which are important questions. Yet these discussions often assume that the educational environment itself remains fundamentally unchanged and that the challenge is simply to decide where artificial intelligence should or should not be introduced. If the argument developed here is correct, that assumption can no longer be taken for granted. The environment within which intellectual formation occurs is itself changing, not because schools have been transformed, but because reality is increasingly mediated by systems designed to reduce the very forms of resistance through which judgement has so often developed.
None of this allows us to predict the future with confidence. Human beings have always learned under changing technological conditions, and every major transformation has produced intellectual virtues that earlier generations could scarcely have imagined. Artificial intelligence may ultimately foster forms of judgement that are different rather than diminished. The mistake would be to assume, however, that because answers remain available, the experiences that once led us towards understanding remain essentially unchanged. What is changing is not merely the destination of thought but the terrain across which thought travels.
By this stage I realised that I had almost stopped thinking about artificial intelligence itself. Leparmentier’s engineers had not offered a theory, nor were they trying to explain a civilisational transformation. They had simply described small changes in the texture of their working lives. Yet those observations had gradually led me towards a question that reached far beyond software engineering: what happens when reality itself no longer educates us in quite the same way as before?
The Educational Question
None of this suggests that education should attempt to recreate every intellectual hardship that previous generations endured. There is little virtue in preserving inefficiency simply because it is familiar, and few would argue that scientists should work more slowly, engineers solve fewer problems or physicians delay diagnoses in the name of intellectual formation. If artificial intelligence enables people to discover treatments more rapidly, design safer infrastructure or devote less time to routine tasks, those are achievements to be welcomed rather than regretted.
The challenge, however, is subtler than deciding which tasks should remain human and which can safely be delegated to machines. It lies in recognising that many of the experiences through which judgement developed were never deliberately protected because nobody imagined they could disappear. They seemed inseparable from the very act of understanding the world. Reality itself ensured that difficult questions demanded patience, that uncertainty could not always be escaped and that our first explanations often proved inadequate before better ones became possible.
For centuries, education quietly relied upon that world. Schools and universities undoubtedly shaped knowledge, values and habits of mind, but they did so within an environment that continuously supplied its own forms of intellectual resistance. Teachers did not have to create every formative experience because many of them arrived uninvited, embedded in the ordinary act of trying to understand something that refused to yield. Education certainly cultivated judgement, but it did so in partnership with a reality that was itself profoundly educational.
Artificial intelligence may be changing that partnership more deeply than we yet appreciate. Not because it eliminates uncertainty altogether, but because it increasingly mediates our encounters with it. The world continues to resist us, yet its resistance is now filtered through systems designed precisely to make that resistance more manageable. In many circumstances this is unquestionably beneficial. The question is whether, in making reality easier to think with, we may also be altering the conditions under which people learn to think in the first place.
If that is so, then the central educational question of the coming decades is not simply how artificial intelligence should be incorporated into classrooms. It is whether we can recognise which formative experiences belonged to a world that no longer exists in quite the same form, and whether new experiences capable of cultivating judgement will emerge within the world that is replacing it. History offers no reason to believe that human beings cease developing wisdom whenever their tools improve. Every technological transformation has altered the conditions under which people think, and each has produced intellectual virtues that earlier generations could scarcely have anticipated. There is every reason to expect that artificial intelligence will do the same.
If this account is broadly correct, the distinctive responsibility of educational institutions may not be to preserve access to knowledge, which intelligent systems will make increasingly abundant, but to preserve encounters with realities that remain genuinely independent of the learner. Difficult texts that resist immediate interpretation, scientific problems whose answers cannot simply be prompted into existence, historical events that refuse present-day assumptions, and other people whose perspectives cannot be reduced to our own all confront us with a world that does not reorganise itself around our immediate purposes. Such encounters are often inconvenient. They are also among the most powerful conditions under which judgement has traditionally developed.
What seems less certain is whether the experiences through which judgement develops will remain unchanged simply because intelligence itself has become more accessible. Answers may arrive more quickly than ever before, while the journey that once led us towards them quietly contracts. If part of our intellectual formation depended upon that journey, then preserving the destination may not be enough.
I have often found myself returning to Leparmentier’s engineers. They were neither celebrating artificial intelligence nor lamenting its arrival. They were simply describing small, almost unremarkable changes in the texture of their professional lives: they became stuck less often, asked colleagues for help less frequently and spent less time inhabiting uncertainty before moving on. Those observations stayed with me because they seemed to capture something larger than the transformation of software engineering. They hinted that a change in our tools might also represent a change in the kind of reality within which human judgement has long been formed.
Whether that intuition ultimately proves correct remains to be seen. Yet if good journalism sometimes notices historical change before we possess the concepts to describe it, perhaps Leparmentier’s engineers were describing more than a new way of working. Perhaps they were offering an early glimpse of a world in which reality no longer teaches us in quite the same way as before.