DishBrain
DishBrain wires human or mouse brain cells in a dish to a chip and plays them a simulated game of Pong. Within five minutes, they showed signs of learning: their rallies grew longer.
Some of what neurons do may be out of reach for silicon alone, so Cortical Labs, a biotech company, tried wiring neurons into a computer. About 800,000 human or mouse cells went onto each chip, roughly a hundred-thousandth of the 86 billion neurons in a human brain. Electrodes told each culture where a Pong ball was and how its moves turned out. Without that closed loop, the learning did not appear. And given the same 70 or so rallies, the cells learned faster than deep reinforcement learning programs.
Sources
- Kagan et al., "In vitro neurons learn and exhibit sentience when embodied in a simulated game-world", Neuron 110(23), 2022
- Khajehnejad et al., "Biological Neurons Compete with Deep Reinforcement Learning in Sample Efficiency in a Simulated Gameworld", arXiv 2405.16946
- Expert reaction to neurons in a dish learning to play Pong — Science Media Centre
- Europe PMC record for the DishBrain paper
Pick a question to go deeper.
Read next
- XenobotsXenobots are living machines built from frog embryo cells to shapes designed by AI. Set among loose cells, they shove them into piles that grow into new xenobots.
- Matrioshka brainA Matrioshka brain is a star wrapped in nested shells of computers, each running on the heat the shell inside throws away. The physics says the extra shells barely help.
- Penrose processThe Penrose process pulls energy out of a spinning black hole by feeding it something with negative energy. A spinning metal cylinder has now copied the black-hole runaway in a lab.
