Failure Is
the Tuition.

Stakes are the teacher

Strip any real decision to its core and it’s this: choosing under uncertainty, knowing you’ll absorb the outcome. Remove the absorbing and you’ve removed the lesson. A competition where the worst case is “didn’t win” teaches performance, not judgment — the kid learns to please judges, which is the opposite skill: reading the answer someone already holds. Real decisions have no judge holding an answer. That’s the entire difficulty of them. School gives zero reps in this — and so, it turns out, do most attempts to fix school.

The baby paradox

Here’s the saddest design flaw in how we raise people. Every human alive learned their first skills through pure, unpunished failure — a baby falls hundreds of times learning to walk, and no parent calls it failing; we cheer every fall. Then school starts, and somewhere around the first report card, failure flips from how learning works to something shameful. Parents’ tolerance for our mistakes shrinks year by year — exactly as the mistakes become more instructive.

And the economics run backwards: failing at eight costs a scraped knee; failing at thirty-eight costs a house. We block failure when it’s cheap and leave people to meet it for the first time when it’s ruinous.

Even the machine learns our way — we just won’t allow it for our kids

The irony of the AI era: every model is built by failure — and by nothing else. Training works like this: the model guesses, the guess is wrong, the error is measured, and the model adjusts — repeated billions of times. Early in training, the model is wrong almost every single time; that’s not a problem to fix, that is the method. The reinforcement systems behind the most capable AI go further: agents fail millions of times in simulation — crash, lose, collapse — and every failure becomes data. No engineer deletes a model for guessing wrong mid-training; wrongness is the raw material intelligence is refined from.

Now the part that should sting. Those billions of failures aren’t free — companies pay for them, gladly, in compute bills the size of national budgets, because everyone understands the failures are the tuition and the intelligence is the return. We accept this economics completely — for machines. The same parent who’d nod at a GPU bill for training AI won’t budget twenty dollars of real failure for their own child. The machine is given the exact thing our children are denied: consequences without humiliation, failure without shame — failure as information.

A 20-baht education

Let me tell you what the alternative looks like, from my own house. I started teaching my daughter about money at kindergarten age — her own money, hers to hold and hers to lose. One day she came to me: a friend had borrowed her money and wouldn’t pay it back.

I’ll be honest about my reaction: I was happy. Her first lesson in lending cost 20 Thai baht — the cheapest this lesson will ever be sold for the rest of her life. Compare the price to the adult version: the friend who vanishes with a real loan, the partner who never settles, the customer who ghosts an invoice. Same lesson, four or five zeros more expensive — plus interest paid in trust.

And she learned it properly, the way only a paid lesson teaches: don’t borrow — and if you lend, don’t expect everyone to pay back. No lecture from me could have installed that. The 20 baht did it permanently. That’s the whole argument of this article in one coin: the tuition was going to be paid someday. I just made sure she paid it at kindergarten prices.

Decision-making requires strength

And the part almost nobody says: judgment isn’t only mental. Deciding means weighing damage — and you can only weigh what you’ve felt. The body learns its limits by taking hits: falls, bruises, losses, recoveries. That’s how a person calibrates how much impact I can survive — and that calibration is what courage under uncertainty actually is.

Now look at what our kids live with: cushioned, scheduled, supervised, screen-fed — protected from every small hit that builds the tolerance for big ones. A person who has never absorbed damage can’t estimate damage. So they either freeze at every decision, or gamble blind. We’re raising both kinds. The decision is the rep — and reps require weight.

What to do instead

Small, real, survivable stakes — early and often. The kid’s own money on the line, not the parent’s entry fee. A real stall at a real pasar malam where unsold stock is his loss. Praise the recovered failure louder than the safe success. And employers, same rule one level up: juniors need survivable failures on real work — punish hiding mistakes, never making them.

Published by IMA AI — August 2026. Written by a father who was happy his daughter got cheated of 20 baht — and by a builder of machines that are allowed to fail their way to intelligence.