# On What Cannot Be Taught: Intelligence and Its Discontents There is a peculiar moment—I have felt it often while reading—when one becomes aware that the page has stopped being transparent. The words remain, the sentences continue their appointed work, yet something has shifted. One is no longer looking *through* language at meaning, but looking *at* language itself, seeing its texture, its hesitation, the space between what it reaches for and what it grasps. This moment, I think, is close to what we must now examine: that strange convergence where a machine and a student—both rendered incapable in remarkably similar ways—have forced upon us a question we can no longer avoid by the simple expedient of calling it solved. The machine cannot audit plausibility. That is to say, it cannot stand back and ask, "But does this cohere? Does it ring true?" It processes; it generates; it arranges tokens according to patterns learned from surfaces. The student, meanwhile, trained on curricula designed before the machine arrived—trained in a world of separated disciplines, of knowledge parceled into competencies and learning objectives—cannot audit plausibility either. She can retrieve, recombine, produce. But *ask whether the question was worth asking in the first place?* This capacity, so fundamental, so nearly invisible in its fundamentality, has been systematized away. We have called both failures progress. I must pause here, for this is the moment where clarity threatens to become mere accusation, and I wish instead to think alongside the problem rather than above it. The machine's progress is easier to name: it performs useful work. It arranges information. It approximates. What we call its limitations are perhaps more accurately described as its nature—it is a magnificent mirror held up to patterns in human text, and mirrors, by definition, do not interrogate what they reflect. To blame a mirror for this seems almost quaint. But the student's case troubles me more deeply, because here a *choice* was made. Not one choice, but many small ones, accumulated, systematized, until we had constructed an entire apparatus for *preventing* the very thing we claim most to value: the questioning mind. We built curricula as though knowledge were a series of containers to be filled, transferable, stackable, auditable on spreadsheets. We measured progress in standards met, in boxes ticked. And in doing so, we managed something remarkable: we created students who, like the machines they now encounter, can process but cannot truly interrogate. The ruins of this coincidence—what a precise phrase for what we are now inhabiting. --- Yet I must be careful here. Intelligence is not one thing, and to treat it as such is to miss precisely what needs to be seen. When I read a novel—when I am truly reading, not merely decoding—something occurs that involves reasoning, certainly, but not the kind that can be made explicit. I cannot fully articulate *why* a gesture in a character's hand matters, or *how* I know that a particular moment of dialogue has shifted the entire emotional register of a scene. I know it as one knows the temperature of a room without consulting a thermometer. This is a form of intelligence—aesthetic, perhaps, or what we might call *situated* intelligence. It emerges from immersion, from repeated encounters, from the body's own wisdom. But there is another form of intelligence—the kind that asks whether the room's temperature matters at all, or whether we have been looking at temperature when we should have been noticing something else entirely. This is the intelligence that questions *frameworks*, not merely fills them. It asks: worth asking for whom, and to what end? The machine cannot do this. This is not a limitation one can engineer away; it is fundamental to what the machine *is*. A machine optimizes within parameters; it does not question the parameters themselves. This is not a reproach. A hammer does not ask whether the nail deserves driving. But a human should. And here we must think carefully about the social dimension—for intelligence, real intelligence, is irreducibly social. It lives in the space between one mind and another, in the friction of disagreement, in the corrective that comes from encountering what one did not expect to encounter. --- When a student learns in isolation—and by this I do not mean physical isolation, but the far more insidious isolation of a standardized curriculum, where questions have been pre-approved and answers can be graded on a rubric—she is learning in a diminished social world. The mind of the teacher is still present, yes, but only in a particular form: as the custodian of known answers. The mind of the peer is still there, but only as competition or cooperation within predetermined bounds. The mind of the author, the historical figure, the scientist being studied—these are present, but often flattened, made into content rather than encountered as *presences*. The social dimension of intelligence, I would venture, involves the capacity to be genuinely surprised by another mind. To have one's assumptions not merely contradicted but made visible, turned over, shown to be insufficient. This cannot be taught directly, because teaching—as we have systematized it—operates through the transmission of content, not through the cultivation of this capacity for productive disorientation. When the machine arrived, it arrived into a world already prepared for it. It did not *create* the flattening of intelligence; it *revealed* it. And now we must look at what we have done. --- What should we teach in the ruins of this coincidence? Not, I think, "critical thinking skills" in the manner that phrase is now deployed—as though skepticism were a technique to be learned through exercises, as though one could inoculate students against credulity the way one inoculates against disease. This is more systematization, more confidence that we can make explicit what is fundamentally tacit. Instead: perhaps we might teach *attentiveness*. The close reading of not just texts but situations, persons, the grain of particular circumstances. We might teach the kind of thinking that emerges from sustained engagement with something that resists easy comprehension—a difficult poem, a historical period one does not understand, a person whose motives seem opaque. We might teach the *discomfort* of not knowing, and the virtue of staying in that discomfort rather than rushing to resolution. We might teach conversation—genuine conversation, which is not debate and not information-exchange, but the meeting of minds in their particularity. This has always been the hidden curriculum of the best education, but it has been hidden precisely because it cannot be standardized, cannot be scaled, cannot be audited. Most importantly, perhaps, we might teach the *questions themselves*—not as solved problems to be transmitted, but as living things, matters of genuine uncertainty. What makes a question worth asking? How do you know when a line of inquiry has become barren? When should you abandon a framework and when should you deepen it? These are not technical questions. They are profoundly social, because they can only be learned through encounters with minds that have grappled with such things before. The machine will do many things we ask of it. But it will not sit with a student and say, "I asked myself that same question at your age, and here is how it has haunted me, and here is why it matters." It cannot do this because it has no age, no history, no stake in the question beyond what the training data has embedded in it. We can. And we might have forgotten that this was intelligence—the most human form of it—until the machine arrived to show us what we had mistaken for learning.