stately plump buck mulligan

He runs the science team at a Melbourne company called Cortical Labs, and the paper that broke the story was published in the journal Neuron in October 2022. The title sounds like a science fiction novel. In vitro neurons learn and exhibit sentience when embodied in a simulated game-world.

The setup was simple, and that is what made it so strange.

Kagan and his team took some brain cells from mouse embryos. They took some human brain cells grown from stem cells. They placed them on a chip covered in tiny electrodes, the size of a small coin, and they hooked the chip up to a computer running Pong.

The electrodes could do two things. They could read what the cells were doing. They could also send small bursts of electricity back into the cells.

The team used those two channels to talk to the dish.

When the ball was on the left, they fired the electrodes on the left side of the dish. When the ball was on the right, they fired the electrodes on the right. The closer the ball got to the paddle, the faster they fired. The cells could move the paddle by sending their own signals back.

That was the whole game.

Then the team added one more rule, and this is the part that changed everything.

When the cells missed the ball, they got a random, chaotic burst of electricity for four seconds. Noise. Static. Pure unpredictability. When the cells hit the ball, they got a clean, steady, predictable signal.

That was the only feedback the dish ever received.

Within five minutes, the cells started getting better at the game.

The rallies got longer. The hits got more frequent. The dish was not winning, but it was clearly playing, and it was improving, and nobody had told it the rules.

It had figured them out by itself.

The reason this worked is the part that should stop you for a second.

Brains hate surprise. That is the thing they are built to avoid. Karl Friston, who is one of the most cited neuroscientists alive and a co-author on the paper, has spent his whole career proving this. The brain is not really a thinking machine. It is a prediction machine. It runs on a single quiet rule. Make the world less surprising.

The cells in the dish were doing the same thing.

The chaotic stimulus felt like surprise. The clean stimulus felt like calm. The only way to get more calm and less chaos was to stop missing the ball. So the cells learned to stop missing the ball, not because anyone trained them, and not because they wanted a reward, but because the only way to quiet the noise was to play the game well.

They were not learning Pong. They were learning to make their own world more predictable, and Pong just happened to be the world they were stuck inside.

The same thing your brain is doing right now.

Every choice you make today, every word you reach for, every plan you build for tomorrow, is your brain trying to make the next moment less surprising than the last one. The feeling you call thinking is mostly your head doing the same thing those cells did. Trying to quiet the static.

The dish learned Pong faster than any AI had at the time, using around 800,000 cells and almost no power, while the AI systems running the same game needed thousands of times more energy and far longer training runs.

Kagan said it plainly in his interviews after the paper came out.

He said the cells were not trying to win. They were trying to feel less lost. And the moment he said that, half the room realized he was no longer just describing the dish.