Human brain cells on a chip learn to play Doom
Mar 17th 2026
An Australian startup and an independent developer trained living human neurons on microelectrode chips to play Doom in about a week, showing faster learning than some silicon systems but still far below human skill and with many unknowns about how the neurons process the game.
- Cortical Labs grows clumps of human neurons on microelectrode arrays and previously trained them to play Pong in 2021.
- A new Python interface let independent developer Sean Cole teach the neuron chip to play Doom in about a week.
- The Doom system used roughly a quarter as many neurons as the earlier Pong demo and beat random play but performed far below skilled human players.
- Cortical Labs says the biological system learned faster than comparable silicon machine learning setups and could improve with newer algorithms.
- Researchers call the result a meaningful step toward biological or hybrid computers for real-time control tasks like robot arms.
- Experts caution we still do not understand how the neurons interpret visual input or the game objectives, so applications remain experimental.