How old is Egypt?
My daughter came home from school with a fact and a question. She'd learned that China is 4,000 years old, and that Iran and Egypt are the oldest countries in the world — but she wasn't sure how old they were.
What followed turned into an unplanned tour of geography and economics. The interesting part, looking back, is the shape of it: who spoke when, and where the AI was allowed in.
The transcript, annotated
"China is 4,000 years old — and Egypt and Iran are even older. But how old are they?"
the spark: a school fact with a missing number
"Let's take turns guessing."
commit to a guess before looking anything up
"Seven million!"
a first anchor — wildly high is fine; it's hers
"Maybe 7,000 is a closer number. And if Egypt is older than China, would its number be bigger or smaller than 4,000?"
calibrate by ordering, not by telling
She works it through: older means bigger than 4,000 — and lands on 7,000 as her guesstimate for Egypt and Iran.
she intuited her way there — the number is hers now
Only now does the AI join. Together we ask it: how do some countries last? Why is Russia so big, and why has it fought so often? What happened to Mongolia's huge empire?
AI enters only after the commitment — as the kid alongside her
Each answer gets tested against her basic model of humans: people need water, oxygen and food to survive.
ground every new fact in invariants she already trusts
The thread ends in landlocked middle Africa: paying more to obtain food and water, with less access to educational resources to sell — and therefore net poorer than the countries around them.
the thread's payoff — geography becomes economics
The moves
- Guess before you look. A committed guess turns the answer into a payoff instead of a fact to absorb.
- Calibrate by ordering, not telling. "Bigger or smaller than 4,000?" is a question she can win with reasoning she already has.
- Ground in invariants. Every new fact got tested against what she already trusts: people need water, oxygen and food.
- Pull the thread. One answered question buys the next: Russia → Mongolia → landlocked Africa.
Where the AI was allowed in — and where it wasn't
The AI never answered first. The guessing game was ours: commitment, calibration, her landing on 7,000 — no lookup. Only after she owned a number did we ask the AI together, and its job then was to be the kid alongside her — following her curiosity, answering at her level, staying child-led. Mine stayed steering: choosing the next question, planting a concept where it would take, and knowing when to step in.
An answer before the guess would have killed the game. That's the boundary.
No lesson plan would have routed from "how old is Egypt?" to trade access in landlocked countries. The question-asking was the curriculum.