Education After “Intelligence” Becomes a Commodity

As investors, builders, and parents, we spend a lot of time thinking about what makes software a good, enduring solution, while also pondering how to best educate our kids. Thinking through what makes an enduring technical solution naturally made us ponder what makes a resilient human in the age of AI.

Machines can now do a whole layer of what we used to call intelligence: recall facts, run the calculation, produce a fluent answer. That layer is getting cheaper every month. So the real question for education isn’t how to make humans better at it. It’s what humans should do once that layer is essentially free and universal.

What the Game of Go showed

When AlphaGo beat Lee Sedol in 2016 at the ancient Chinese game of Go, it looked like the old story: machine beats human, proof the machine is better. But what happened next was more interesting. Human play got better. A study of more than 5.8 million professional moves from 1950 to 2021 found that after superhuman AI arrived, human professional players made more original and more accurate decisions. They started trying sacrifices and shapes they’d once have dismissed, because the machine had explored ground they hadn’t. The emblem of that was what was called “Move 37″, a choice so alien that commentators assumed AlphaGo had made a mistake. It hadn’t. The move was brilliant, and it simply sat in a corner of the game that centuries of human players had never considered.

AlphaGo didn’t take Go somewhere it could never have gone. It compressed into a few years the exploration that might otherwise have taken humans generations. In the case of Go, maybe more centuries of human exploration.

In a sense, that’s what every useful technology has done through history. Writing didn’t invent memory, it extended it. The calculator didn’t invent arithmetic, it did it faster. AI is the same pattern, applied to answers. So the machine isn’t a teacher and isn’t an enemy. It’s an accelerant. It covers known ground faster than we can, which frees us for the part it can’t do.

The scoreboard

There’s a lot technology can’t do, starting with getting started: AlphaGo never invented Go. Humans built the board, wrote the rules, and decided what winning means. The machine only explored, brilliantly, inside a game we handed it. It optimized the frame; it never questioned the frame. Even AlphaGo Zero, the later version trained on zero human games, was still standing on the deepest human input of all, the definition of the game and the rules of engagement.

A scoreboard exists only once a human has decided what counts as a point. Inside it, the machine is untouchable, because it can play itself a billion times and read the score. But deciding what the score should be, or noticing that the scoreboard measures the wrong thing, has nothing to optimize against. A machine can select an arbitrary target or score itself against a random metric, but deciding what actually matters and carries value remains uniquely human work. Someone, whether a person, an institution, or a society, has to define the game before there is anything worth optimizing.

It’s the difference between playing in Newtonian universe perfectly and being the person who realized Newton’s universe was an incomplete frame. Feed a machine the rules of a pre-relativity world and it would model that world better than any physicist alive. What it would not do, on its own, is what Einstein did: decide the frame was incomplete and replace it, or augment it.

Blind spots make the same point from the other side. In 2023 a research team probed KataGo, one of the strongest Go engines ever built, by playing it over a million times until they found a strategy it badly misreads. A human amateur learned the trick and beat KataGo fourteen games out of fifteen, with no computer helping him play. The engine still beats those same humans easily in a normal game. But it had carried a gaping hole for years without noticing, because it only ever practiced against copies of itself that shared the blind spot.

Put the two together and you have the whole argument. Inside the game, the machine accelerates past us. The work that’s left, defining the game, noticing when it’s the wrong game, inventing the next one, remains out of the machine’s reach as there is no scoreboard that can be handed to the machine. Education’s job now is to teach kids to climb to that next level, and not to compete on the rung the machine already owns.

The floor above

Most of what makes a person valuable in this world sits one level above producing an answer. Roughly five skills live there, and they stack up in order.

Framing the problem. Real problems don’t arrive pre-packaged. “Cut food waste in half” has to be broken into parts before anyone can act on it: what to measure, what’s habit versus logistics, what to estimate versus check. Schools usually do this step for the student, handing over a clean question with the method just taught. AI can currently go thorugh the motions of creating sub-structurs for bigger problems, but it still strongly benefits from a human doing that for it, and that will probably remain the case for a long time to come.

Knowing when the normal answer stops working. Here is a scenario for anyone who has compared how it feels after an exam they were extremely well prepared for versus one they were less prepared for.”

After the exam they were extremely well prepared for, the person can replay their answers and identify exactly which questions they have answered incorrectly and why. After the exam they were less prepared for, they are much fuzzier about what might be right or wrong. They had made enough answers based on what seemed highly probable that, after a while, it becomes difficult to predict what their final score might be.

AI is like the moderately well prepared person, although not necessarily because it lacks information. It may know a great deal while still failing to recognise when a familiar rule is being applied to an unusual case. It will often produce the average or highest probability answer with reasonable confidence, even when that answer does not quite fit. Nothing about a wrong answer necessarily flags itself as wrong.

The future of education therefore needs to teach around the edges. Whenever possible, topics should be explored through three types of cases: where the rule works, where it partly works, and where it fails.

The goal is to develop students who notice when they have moved beyond the point at which the average or highest probability answer still applies.

Judging answers instead of just producing them. The machine gives you a fluent answer for free. The scarce skill is looking at several and knowing which is strongest, what’s wrong with the others, and what would raise your confidence. Defending your work out loud is the classroom version of this, and it catches the moments where a confident answer and a correct one have quietly split apart.

Knowing which fact matters. More information is rarely the edge by itself; machines can now process more of it than any person can. Knowing which single piece changes the answer is the edge. Give students more material than they need and let them find the part that actually decides it.

Owning the call. Someone still has to decide who gets the scarce resource, whose harm counts against whose benefit, and defend that when it goes wrong. A company can hand this to a review board. A person can’t outsource it, because in the end they are the one accountable. This belongs in every subject, not a separate ethics class, because it only counts when something is actually at stake.

The order isn’t decorative. You can’t judge answers to a problem you never framed, spot a risky case using judgment you never practiced, or own a decision you couldn’t tell was risky in the first place. Each skill is the ground the next one stands on. Skip a rung and everything above it wobbles.

But the foundation still matters

None of this means dropping arithmetic, grammar, history, or chemistry. It means changing who does the heavy lifting and who just has to understand it.

As the line often attributed to Picasso goes: “Learn the rules like a pro, so you can break them like an artist.”

Think of structural engineering. An engineer no longer has to calculate every load by hand or draft every drawing from scratch. Software can model forces, test designs, surface conflicts and revise plans faster than paper ever could. But the engineer still has to understand what the software is doing. They need to know whether the assumptions are right, whether the model reflects reality and where the structure might fail.

That is the role AI is starting to play in education. It can produce the answer, run the calculation and generate the first version faster. But students still need to understand the work underneath, because you can only direct a machine, challenge it or catch it when it is wrong if you understand the thing it is doing for you.

So the honest objection, “if machines will eventually build the higher floors too, why learn any of this,” has an answer. Understanding is what lets you climb and steer. The person who only presses the button is at the mercy of whoever, or whatever, understands the building. Understanding the foundation stays non-negotiable; being the one who lays it, brick & mortar, does not.

A classroom that spends all its time laying bricks produces students who can build one floor and stop. A classroom that covers the basics and also preserves time to ask “so what?” will give invaluable skills to post-AI students: debating whether a decision was right, telling a fight over facts from a fight over values, judging a choice from the view of someone it affects differently than it affects you. You reach these only by getting through the foundations first, not by skipping them.

This isn’t a leap into the unknown

In our humble opinion, the panic that AI forces us to reinvent school from scratch is overstated. Some of the systems we’ve studied and lived through have been building these upper floors for decades already, long before anyone worried about a chatbot taking people’s livelihoods.

France’s baccalauréat philosophy exam is rigorous in asking students to identify and formulate a problem, reason through it rigorously, examine a thesis, and justify a conclusion over four hours. Its Grand Oral asks students to defend a topic clearly and convincingly in front of examiners. The International Baccalaureate’s Theory of Knowledge asks how students know what they claim to know, while CAS ties learning to action, reflection, and projects beyond the classroom.

None of this solves the whole problem, but the blueprints already exist. The gap is access. France tracks students unevenly, and the IB is often expensive or selective.

Others have circled this too. Ethan Mollick argues students should become judges of AI output rather than producers of it, the same skill from the same observation. The OECD’s Learning Compass names creating new value, reconciling tensions, and taking responsibility as its transformative competencies, drawn up years ahead of the AI wave. What’s new here is the point that machines climbing past us is the reason to climb.

What you can do tomorrow

Schools move slowly. Curriculum takes years, retraining teachers even longer. Waiting on systems to catch up probably isn’t the most resilient plan for a kid growing up right now. The faster lever is the hours a child spends at home, which dwarf the school day and mostly go unused as learning time.

Here’s what learning at home could look like. A seven year old, fresh off watching Mary Poppins, ponders if jumping off the terrace roof holding an umbrella would let him fly (based on a true story!). Adults intercept the action on time, and instead of a lecture, the parents decide: let’s teach the kid the scientific way to ask a question like that.

That can became the whole exercise in one conversation. What’s a small first test that tells the kid something real without putting the kid or umbrella at risk? What would that test need to show before the next one makes sense? What’s the smallest version of the question still worth asking?

That’s framing the problem, spotting which detail actually changes the answer, and running a loop of small revisions instead of taking one answer on faith. No curriculum required. It just took seeing that a kid trying to fly off a with an umbrella is a learning opportunity. It will be engaging for the kid, and potentially entertaining for the adults involved.

This is the easiest thing in this whole piece to do, and the thing parents skip most, because the instinct is to use parental time to keep kids safe but not necessarily to engage intellectually when situations reveal themselves for further discussion. Engaging kids while they are in a moment of intense curiosity is when they’ll actually listen

Kids pick stranger questions than any curriculum would. The umbrella, the bug they won’t drop, the fight over whether a game’s rules are fair. Each one is a chance to run the whole thing in miniature: what’s really being asked, where the obvious answer breaks, what a better one looks like, what to check, what you learned, and whose problem it is if you’re wrong. The machines will keep taking the floor below. Our job, as schools and as parents, is to make sure kids are already standing on the one above it.