There is an old physics textbook, still in print after more than half a century, that begins every idea the same way. Not with an equation. With something you can watch happen. A ball rolls down a slope, a shadow moves across a wall, a spinning stool tightens or loosens as you pull your arms in, and only after you have actually noticed the thing does the book offer you a name for it. Paul Hewitt built his career on a single stubborn premise: concepts before calculations. Understand it first. Compute it later, if you ever need to at all. Look first. Name later.
That’s the whole method, and almost nothing else built at scale is designed that way.
The termite is the first demonstration of it, and it doesn’t even require a mind. Take two structures that both had to solve the same problem, keeping something cool under a punishing sun, and watch how differently they arrive there. A termite mound in northern Australia grows as a flat wedge, its long axis running north to south, so precisely that researchers who rotated one experimentally into an east-west line watched its temperature swing from a stable 33-35°C plateau to spikes of 40-42°C within days. No termite measured the sun’s angle or calculated a thermal gradient. Selection did the measuring, over unthinkable stretches of time, and the mound is simply the readout.
A pyramid solves the same class of problem by the opposite route. Someone had to track circumpolar stars to true north. Someone had to flood shallow trenches and trust that water finds its own level whether or not anyone understands why. Someone had to discover, empirically, that wetting the sand in front of a sledge, a technique shown in a tomb painting, could dramatically reduce the force required to pull it, long before anyone had a word for the friction coefficient involved.
One structure encodes physics without a mind behind it. The other required a mind to go and get the data itself, standing under the actual sky, with actual water and actual rope. Different routes to the same respect for what can be observed.
That capacity, watching a thing closely enough to let it answer your question honestly, has no age limit and no particular gift required. What it has instead is an unusually fragile social life. A child asking why water always finds the same level is being tested by the water. The water doesn’t care what answer is hoped for. A child asking a question in most classrooms, most family dinners, most workplaces, is being tested by something else entirely: whether the answer matches what the person in authority already had in mind. Those are two different feedback loops wearing the same word, question, and only one of them actually trains observation. The other trains something adjacent and much less valuable, which is the ability to guess what’s wanted.
Most institutions that touch a growing mind end up, without quite meaning to, running the second loop. And the cost of that isn’t abstract. It shows up as a series of specific things, quietly sacrificed, one after another, the moment curiosity is allowed to actually lead.
Usefulness goes first, and not by accident. A curriculum built around genuine open questions can’t promise, on any fixed date, that a student will emerge able to compute a beam’s load or balance a reaction. Real understanding arrives when it arrives, sometimes in a flash, sometimes after weeks of confusion that looks, from outside, exactly like falling behind. A calculations-first curriculum guarantees something schedulable instead: this student, on this date, can produce this formula on demand. Institutions can budget for that. They cannot budget for the other kind, and Hewitt’s own book shows where that leaves things: what he himself calls the dominant liberal-arts physics text in the country, not the one engineering and pre-med students are actually assigned.
Time goes next, for a blunter reason: it’s the one resource nobody can manufacture more of. A class period is forty-five minutes whether or not real understanding arrives in that window. Institutions convert an un-schedulable variable into a scheduled one the only way they can, by fiat, deciding in advance how long a question is allowed to run before it has to produce a checkable answer. That decision has nothing to do with how long understanding actually takes.
Comparability follows for a related but distinct reason. The moment learners are allowed to follow real questions at their own pace, they stop being sortable against each other, and most gatekeeping mechanisms a society runs, admission, hiring, licensing, a single national exam standing between a student and a medical seat, need everyone tested on the same material on the same day so the results can be lined up and cut at a percentile. Idiosyncrasy, which is what real curiosity produces, is illegible to a ranking. So the ranking wins.
And last, the comfort of certainty, which is the most personal casualty and maybe the cruelest. Real inquiry means being visibly seen not knowing, in front of a teacher, a parent, a room of peers who may already have the answer. Watch how a room actually behaves: hands shoot up fastest for questions with a known answer, since arriving there first earns status, while a genuinely open question meets silence, since nobody wants to be caught guessing aloud. That isn’t a failure of courage in any one child. It’s what happens when not-knowing carries a social cost, and most rooms attach one without meaning to.
Notice what those four things actually are: deliverable output on schedule, a shared clock, a sortable population, participants who don’t need much tolerance for visible uncertainty. Which is to say, they are exactly the four things any institution needs in order to function at scale. Curiosity threatens all four at once. That isn’t a flaw in how any particular school or family or company is run. It’s what happens whenever real inquiry meets an operation that has to serve more than one person on the same timetable.
It has been solved before, at least partially, and by a method worth remembering precisely because it wasn’t a pedagogical trick. In the 1830s a Scottish missionary named William Adam surveyed indigenous education in Bengal and Bihar and estimated roughly a hundred thousand village schools, nearly one for every two villages, teaching in the vernacular under a single local teacher, paid directly by the families whose children showed up. That single arrangement rewired every casualty on the list without anyone theorizing about it. Time wasn’t rationed by a shared clock, because there was no cohort of thirty that had to move in sync. Comparability barely mattered, because nothing pulled a child into a distant ranking system, competence was judged locally and immediately: can this child now keep the family’s accounts, write the letter that needs writing. Usefulness wasn’t distorted into exam-certifiability, because the immediate customer was the family itself, and a teacher who stopped delivering something a family valued simply stopped being paid.
It didn’t solve everything. Adam’s own reports track how thin female education was, and how caste shaped who got taught at all. Decentralizing accountability to a village doesn’t erase a village’s hierarchies, it just changes which ones do the flattening. But within its limits, it was a working answer to a problem we still haven’t solved, and it is worth noticing how it declined. Adam was commissioned in January 1835, and within weeks Macaulay delivered his famous minute arguing for a centralized, English-medium, examined model of schooling, the start of a decades-long policy shift. The pathshalas were entering a system increasingly built around a different purpose: producing knowledge an administration could certify, standardise and compare. They weren’t out-taught. They were out-scaled.
Which is the warning worth carrying into whatever comes next. Something is now being built that could, for the first time, make patient, individualized judgment cheap rather than a luxury reserved for whoever can already afford a tutor or a small class. That is a genuinely new lever on an old problem.
But the same four pressures that hollowed out the village school are already built into the incentives of a tutoring product, just wearing different clothes. Usefulness gets redefined as whatever a parent dashboard can show progress on by Friday, not whatever the child was actually chasing that hour. Time gets rationed the way a class period rations it: a session has a natural length before a child’s attention, or a subscription’s unit economics, decides it’s over, so an open question faces the same clock pressure it faced in Hewitt’s forty-five minutes. Comparability creeps back in through streaks, levels, and grade-benchmarked scores, the same sorting instinct that made a national exam necessary, just personalized and gamified rather than administered once a year. And the comfort of certainty is the one a well-optimized product has the strongest reason to protect of all, since a tool that keeps a learner comfortable keeps a learner subscribed, endlessly patient at telling them what they already suspected they wanted to hear.
The honest, harder question is whether such a system could do the opposite: preserve the discomfort instead of dissolving it. A genuinely curiosity-preserving version would have to be willing to say, more often than any product manager would like, I don’t know, let’s find out, and mean it. It would ask a learner to observe something before it explained anything, and resist handing over the fact that observation would have produced, tolerating a wrong hypothesis long enough to let the learner test it rather than correcting it on sight. It would refuse the session boundary that cuts a question off before it’s answered, the leaderboard that turns understanding into a rank, and the dashboard metric that stands in for whether a child actually understands the thing. Sometimes it would need to send the child back into the physical world entirely, to actually watch the water find its level, rather than describing the result. And it would need to admit uncertainty in its own voice, not just prompt for it in the learner’s.
That is the same discipline the water example asked of the child at the very start of this essay, refuse the shortcut to the answer, trust what can be watched directly to do the teaching. It is also the harder, less profitable version to build, since a tool that keeps deferring the answer risks losing the user before the answer ever arrives. Nothing about the technology decides which version gets built. That is still a human choice, made under exactly the same pressure that has been deciding this question in every institution named so far. Hewitt’s method was look first, name later. The only honest question to ask of what’s being built now is whether it will help us look, or simply help us name faster.
And if the judgment behind it all sits with a handful of systems built by a handful of companies, the many small gates a healthier structure would need have not multiplied at all. They’ve just moved somewhere much less visible than a leaked exam paper that hundreds of thousands of people can see and march against.
So end where this began. Every one of those casualties, usefulness, time, comparability, the comfort of certainty, was really a description of what gets extracted from curiosity the instant someone insists it justify itself. Hewitt drew his own diagrams for forty years instead of letting a faster stock image ship in their place. A hundred thousand village teachers taught what families came to them to learn, none of it scaling, none of it certifiable at any distance greater than the next village over. None of them were running a cost-benefit calculation when they did it. The question, or the drawing, or the child in front of them, was worth attending to because it was true, not because the attention would pay off later.
It may not be the only human faculty built this way, but it is the clearest case of one: a thing that stops being itself the moment you ask what it’s for.