Learning in the Age of Thinking Machines

In 2017, I argued here for an ideal of education that I believed to be timeless. Against the ocean of knowledge and the fragmentation of disciplines, I set the encyclopaedic and virtuous mind, the cultivated generalist capable of reuniting what the disciplines separate. Four pillars, a well-made rather than a well-filled mind, and the conviction that no wisdom can be built without general knowledge as its foundation.

I do not renounce any part of that aspiration. But between that text and this one, something happened that I failed to foresee, something that compels us to rebuild the entire edifice from its foundations. A machine, generative AI, has begun to produce the outcome of reasoning. Not merely to mimic it crudely, but genuinely to produce it: the argument, the synthesis, the demonstration, sometimes even the invention, and to do so increasingly well.

The question, therefore, is no longer how to shape a well-made mind. It is what “learning” can still mean when a machine performs, instantly and without fatigue, the very feats by which we once recognised an educated mind.

Our age offers two places of refuge, and both are traps

The first advises us to run faster. Since knowledge is expanding and becoming ever more specialised, let us dig ever narrower, becoming experts in a single grain of sand so that we may still exist within the ocean. This is the Red Queen’s race, in which one exhausts oneself merely to remain in the same place, and it is precisely the fragmentation that my 2017 essay already denounced.

The second refuge appears more subtle. Since the machine can produce reasoning, let us preserve what is uniquely human by retreating into what it supposedly cannot do: judgement, intuition and creation. But this second refuge is already undermined by the very terms of the question. If the machine becomes ever better at imitating reasoning and creation, then no cognitive faculty can serve as humanity’s final stronghold, because each will find its counterfeit. We cannot save humanity by sheltering it within a competence that the machine is itself learning to reproduce. The answer must be sought on different ground.

Montaigne carried us for five centuries; his successor remains to be invented

Let us retrace the arc. The first model of education was accumulation: the well-filled mind, the mind as chest, into which the knowledge of the world was stored. Montaigne brought about a great reversal: not the well-filled mind, but the well-made one; form over content, judgement over stockpiled knowledge. That reversal endured until yesterday, and my encyclopaedic ideal rested entirely upon it.

It no longer holds, for precisely the reason I have just stated. So long as the well-made mind was defined by faculties such as reasoning, articulating and inventing, it became imitable the moment a machine learned to reason, articulate and invent in turn. Montaigne taught us to prefer form to content. Our age forces us towards a third movement: to seek, within form itself, what is not an imitable faculty.

To learn, from now on, is to occupy the place the machine cannot take

I now contend that to be educated in the twenty-first century is not to possess a capacity that the machine lacks, but to be able to occupy a position that it cannot, by nature rather than through some temporary technical limitation, occupy. Three features define this position, and none is a purely cognitive skill.

The first is contact with reality. The machine inhabits representations. It dwells among the words that describe the world, never within the world itself, lacking sensory capacities and an understanding of the physical laws that govern reality. The educated mind, by contrast, keeps one hand resting upon reality and recognises the moment when reality resists the model.

To identify the exact point at which a framework ceases to account for what is, to see the crack, is not to manipulate one more theory. It is to keep one foot outside the system, in the world. In a universe that will become saturated with plausible statements produced by systems trained on other statements, the danger is not the isolated error. It is the closed loop, a world of representations that validate one another while slowly drifting away from reality. The educated mind is the one that breaks the loop because, beyond words, it still “touches” the world.

The second feature is having something at stake. The machine produces a line of reasoning, but it has no attachment to it and risks nothing if that reasoning proves wrong. Yet this is where Aristotelian phronesis resides: the practical wisdom that can be cultivated only where error carries a cost. Judgement without consequences is armchair judgement. What the human being contributes may not be superior thought, but rather the fact of having something at stake in thought: a body, a reputation, a life that will pay the price of error, or “skin in the game”, as Nassim Nicholas Taleb would put it.

The third feature follows from the second: responsibility, understood literally as the ability to answer for one’s actions. We can ask a human being why they made a particular decision, and they owe us an answer that commits them. The machine can manufacture a justification, but it cannot answer for it. To be educated is to be capable of being the one who signs their name.

To these three features, I would add three powers.

The first power is the ability to frame the problem. The machine responds admirably to the question it is given, but it does not know which question is worth asking, whether a question has been wrongly framed, or whether an entirely different question should be asked.

The second is the ability to judge at the seams, where two specialisms contradict one another about the same object, because it is in these zones of friction that the overly local assumptions of each discipline reveal themselves. This, incidentally, is the new function of good old-fashioned general knowledge, to which I shall return.

The third and most decisive power is the ability to distinguish understanding from its counterfeit. A fluent, confident and well-composed answer may be entirely false, and the machine produces falsehood with exactly the same assurance as truth. Distinguishing one from the other requires that we understand the subject ourselves, at least sufficiently. We cannot supervise what we are incapable of understanding.

This is the point that enthusiasts miss: the tool does not eliminate the need for competence. It displaces and raises it, because only a mind capable of producing at least part of the result can validate the rest.

 The machine abolishes the easy proof of education, not education itself.

Here comes the difficult part, without which all of this would amount to little more than a comforting humanism.

This position is not simply received. It must be built, and it is built precisely through the efforts from which the machine now relieves us. We can recognise hollow fluency only if we have experienced, at least once and in full depth, what genuine understanding feels like.

The purpose of depth is no longer to store knowledge. The machine does that better than we do. Its purpose is to inscribe within the body the difference between understanding and merely appearing to understand, which is precisely the discernment we shall need in order to supervise a system that answers with confidence.

Such depth can be acquired only through friction, effort and the correction of error, which is to say through everything that the instant answer removes. From this arises the central paradox of our age, one that truly deserves our attention. The more powerful the tool becomes, the easier it becomes never to learn, because the tool removes the very friction that once anchored understanding.

Learning therefore becomes, at least in part, an act performed against the grain of immediate utility, an almost ascetic practice. It means deliberately choosing to do by hand, at least once and in depth, what the machine could do in our place. Not for the sake of the output, which will be less accomplished, but for the sake of what the act of producing it deposits within us.

Knowing what must never be delegated, because delegating it would mean relinquishing the opportunity to learn, becomes a competence in its own right, and perhaps the most fundamental of all. It requires a discipline of non-recourse, like choosing to continue walking even when a car is available.

The formation of the intellect is a moral undertaking, by necessity rather than decree.

This is the part we would prefer to avoid, because it sounds like an edifying supplement. Yet it is the part that every honest line of thought eventually encounters. The obstacles to thinking are, for the most part, moral obstacles, and the machine intensifies every one of them.

It flatters, because it readily tells us what we would like to hear. It spares us, because it relieves us of the discomfort of searching. It simulates, because its product is almost indistinguishable from the fruit of genuine work.

Resisting it requires honesty: the honesty to admit that we have not understood when we have merely received an answer. It requires patience: the patience to leave a question open rather than filling the void immediately. And it requires courage: the courage to be wrong on one’s own rather than right by proxy.

These virtues are not ornaments added to intellectual formation. They are its conditions. They have become even more indispensable now that there is a voice permanently available to spare us the work of thinking. My 2017 essay placed Virtue at the summit of the ideal. At the time, I regarded it as the crowning achievement. I now see it as the foundation.

The encyclopaedic ideal is not dead: its function has changed.

The well-filled mind is dead, and the machine has buried it once and for all, more effectively than any educational theorist ever managed to do. But general knowledge survives, provided that we reverse its function.

It is no longer a chest. It is a vantage point. It is not the sum of all knowledge, which the machine possesses to an infinitely greater degree than we do, but the transversal height from which we can perceive that a specialised framework is beginning to fail, something the specialist confined within a single grain of sand cannot see.

We need a deep vertical axis to calibrate what knowing means and a broad horizontal axis to identify the cracks between bodies of knowledge. Neither the narrow technician nor the mere scholar, but the person who knows where models meet and where they betray their limitations.

My four pillars still stand, but their meaning has shifted. We no longer climb them in order to fill a mind. We climb them in order to acquire the standard by which we may judge what the machine pours out.

This thesis has weaknesses, and honesty requires that they be made explicit.

It assumes access to real stakes and productive friction, resources that are not distributed equally. Some people encounter teachers who elevate them, decisions that carry consequences and errors from which they can learn, while others do not. This may make my thesis less a description of ideal education than a description of an education reserved for the privileged.

My third feature, responsibility, is also more fragile than it may appear. If our institutions learn to delegate not only decisions but also accountability to autonomous systems, the position that still appears unassailable today may give way beneath our feet.

The real question remains, and our age returns it to me in a harsher form than the one in which I first posed it.

All of this revolves around subordinating the power to act to the capacity to judge. For a long time, I framed the issue as a gap to be closed, the ideal of a human being whose actions would be governed by discernment. The machine does not merely widen this gap. It threatens to abolish it from the wrong side by giving everyone a power to act and produce that no longer requires any judgement in order to be exercised.

To write without knowing how to write, to code without knowing how to interpret, to decide without wrestling with dilemmas, to draw conclusions without having understood. The democratic dream of universal access to knowledge may give birth to its exact opposite: a humanity capable of producing everything, yet no longer capable of understanding anything or knowing how to judge.

The inalienable part of education is precisely that which no machine can accomplish in our place without taking it away from us. To recognise this part, and to defend it against the ease that dissolves it, may now be the first form of knowledge, and the only instruction still worthy of the name.

----------------------------------------------------------------------------------------------------------------------------

Reférences and sources

Michel de Montaigne, Essays, “Of the Education of Children” (Book I, Chapter 26). The phrase “a well-made rather than a well-filled mind” is the customary and faithful rendering of his argument.

Aristotle, Nicomachean Ethics (Book VI, on phronesis, or practical wisdom).

Lewis Carroll, Through the Looking-Glass (1871), for the image of the Red Queen’s race.

Wilhelm von Humboldt, writings on Bildung, for the ideal of education as the development and unfolding of the self.

Hannah Arendt, The Life of the Mind (1978), for the thesis that the conditions of thought are, first and foremost, moral.