Every time before, we outsourced part of our minds and the brain noticed and rebuilt.
This is the first time we’re outsourcing it to something that has a mind of its own.
And this time, we get to watch.
https://t.co/HyK6iJtF6D
Your brain is smaller than your ancestors’ from 3,000 years ago. About a tennis ball’s worth of tissue, gone.
Jeremy DeSilva at Dartmouth’s theory: we got good at storing knowledge outside our heads. The brain stopped paying for it. So it pruned.
Andy Clark’s argued for 30 years that your mind doesn’t end at your skull. Your phone is your notebook. Your AI is your notebook.
Lose it and you don’t lose a device. You lose your map of the world.
London cab drivers grow bigger brains.
To pass an exam called The Knowledge, they spend 3 years memorizing every street in central London. Eleanor Maguire scanned 43 of them before and after. In the ones who passed, the hippocampus physically grew.
Now flip it.
A 2025 MIT study found heavy AI users showed weaker brain activity than people doing the same task without it.
The brain grows around what it does. It shrinks away from what it doesn’t. We are about to find out what it does with tools that think for us.
The skulls have a record of what civilization did to us, written in bone.
It's hard to think about that without wondering what it's going to notice next.
https://t.co/oKnYjbUtAk
Cro-Magnon had bigger brains than we do. By about the volume of a tennis ball.
Documented across continents since 1934. Human brains today are ~8.5% smaller than they were 10,000 years ago.
Same window we invented agriculture, writing, cities, money, religion, and law.
Jeremy DeSilva at Dartmouth calls it cooperative intelligence. We pushed the thinking out of our heads and into the group. The colony does the remembering now.
We didn't get dumber. We split atoms and send machines to other planets.
The brain didn't just shrink. It reorganized. We traded some capacity for the social and symbolic processing that lets us live in groups of millions.
Researchers at Cambridge and MIT looked at 30 AI agents being deployed by major labs in 2026. They asked one question of each: from the public docs, can you tell what it’s supposed to do, what it isn’t, and what happens if it breaks.
Four had answers. Four out of thirty.
There’s a working alternative. Waymo’s robotaxis stop and call a human when they hit something they don’t understand. The Pentagon endorsed this approach in 2012. 1,500 Amazon engineers petitioned for it this March. It’s being ignored.