Brain Cells in a Dish Learned to Play Pong. What That Says About Sensory Overload

Mon Oct 26 · Science

In 2022, a team in Melbourne grew about 800,000 brain cells on a chip, hooked them up to a game of Pong, and watched them get better at it within five minutes. [1]

No body. No eyes. No reward, at least not the kind you're thinking of. No treat for a good rally, no shock for a miss. Just a dish of cells, a grid of electrodes, and a paddle.

Sit with that for a second. A thing with no idea what a game is figured out how to play one.

How do you teach a dish?

The system was called DishBrain, built by the company Cortical Labs and published in the journal Neuron. [1] Electrodes on one patch of the culture told the cells where the ball was. Activity in another patch moved the paddle.

The clever part was the feedback. When the paddle hit the ball, the cells got a short, tidy, predictable pulse across every electrode at once. When the paddle missed, they got four seconds of random noise, a burst of electrical static with no pattern at all. [1]

That's it. The researchers weren't paying the cells in pleasure. They were giving them a choice between a world that makes sense and a world that doesn't. And the cells, apparently, preferred the one that makes sense. Rallies got longer than in the control conditions. [1]

The idea behind it: surprise is the enemy

The experiment was built to test a big, strange theory from neuroscientist Karl Friston called the free energy principle, often discussed as active inference. [1] The short version: anything that keeps itself together, from a cell to a person, is constantly guessing what's about to happen and trying to shrink the gap between the guess and what actually happens.

There are two ways to shrink that gap. You can change your guess, which is learning. Or you can change the world so it matches your guess, which is action. [1] Moving a paddle to hit a ball is the second kind. The noise stops when the world behaves.

If you've read our pieces on the free energy principle and Markov blankets, this is that idea leaving the whiteboard and showing up in a lab.

Now the cells play Doom

In early 2026, Cortical Labs showed its CL1 system, about 200,000 living human neurons on a chip, playing Doom. [2] The game's screen is translated into patterns of electrical stimulation, and the neurons' firing is read back out as moves: turn, walk, shoot. [2] The company's own description is refreshingly honest. The cells "play a lot like a beginner who's never seen a computer," they die a lot, and they're learning. [3]

So a dish of human cells is now wandering a 1993 shooter, badly. The universe is under no obligation to be less weird than that.

The fight over what it means

Not everyone was impressed by the original paper's language. The authors used the word "intelligence" in the paper and even "sentience" in the title. A group of researchers published a formal response in Neuron arguing that those terms weren't supported by the data and controls. [4]

That's worth holding onto. What's solid is the narrower finding: cultured neurons changed their activity in a way that kept the predictable feedback coming. What it says about minds is still a live argument.

So what does this have to do with loud rooms?

Here's where it gets personal for a lot of us.

The same "brain as prediction machine" idea has been used to think about sensory overload. In 2012, Elizabeth Pellicano and David Burr proposed that some autistic perception might come from leaning less on prior expectations, so the world arrives rawer, "too real." [5] In 2014, Sander Van de Cruys and colleagues proposed a related idea: that some brains treat prediction errors as important across the board, so they can't easily file small surprises under "ignore." [6]

These are hypotheses, not settled science. But look at the dish again. The cells learned to act on their world to make it more predictable, because unpredictable input is what they couldn't stand.

Now think about a restaurant at 7 p.m.: plates clattering, three conversations, a blender, music with lyrics. Every sound is a surprise you didn't ask for. If your brain doesn't turn the volume down on surprises, that room isn't just loud. It's a four-second burst of static that never ends.

And look at what we do about it. Headphones. Sitting with our back to the wall. Ordering the same thing every time. Leaving. From the outside, people call that rigid. Through this lens, it's the paddle. It's acting on the world so the noise stops.

That's not proof of anything. It's a better metaphor than most, and it came out of a dish.

The flat-circle part

There's something a little cosmic about it. The cells never asked to be in a game. They didn't know Pong existed. But given a world, they started shaping it toward sense, the same way every living thing has since the first cell put up a wall between itself and everything else.

Maybe that's what a mind is at bottom: a thing that would rather the world made sense, and does something about it when it doesn't.

Sources

1. Kagan, B. J. et al. (2022). "In vitro neurons learn and exhibit sentience when embodied in a simulated game-world." Neuron 110(23). https://pmc.ncbi.nlm.nih.gov/articles/PMC9747182/
2. The Verge (Feb 28, 2026). "Watch a computer powered by human brain cells play Doom." https://www.theverge.com/science/887047/human-brain-cells-computer-doom
3. Cortical Labs (2026). "Living Human Brain Cells Play DOOM on a CL1" (video). https://www.youtube.com/watch?v=yRV8fSw6HaE
4. Balci, F. et al. (2023). "A response to claims of emergent intelligence and sentience in a dish." Neuron. https://www.cell.com/neuron/fulltext/S0896-6273(23)00113-7
5. Pellicano, E. & Burr, D. (2012). "When the world becomes 'too real': a Bayesian explanation of autistic perception." Trends in Cognitive Sciences 16(10). https://doi.org/10.1016/j.tics.2012.08.009
6. Van de Cruys, S. et al. (2014). "Precise minds in uncertain worlds: Predictive coding in autism." Psychological Review 121(4). https://doi.org/10.1037/a0037665