Human neurons on a Cortical Labs chip are taught to play Doom

Filed retrospectively. Before the Specola kept its nightly watch, these were the observations that turned the telescope.

In late March 2026, Cortical Labs reported that roughly 200,000 living human neurons grown on one of its chips had been taught to play the 1993 first-person shooter Doom. In a demonstration video the neuron-controlled character moved through corridors, met enemies, and fired, playing clumsily and dying often, but playing. The result extended the company’s earlier work in which cultured neurons learned Pong, described in a 2022 paper in the journal Neuron.

According to Scientific American, the Doom interface was not built by the company’s core staff. Cortical Labs ran a hackathon with Stanford University, where an independent researcher, Sean Cole, paired the neurons with a standard learning algorithm using the CL1’s Python interface, converting the game’s visual state into patterns of electrical stimulation. The company’s chief scientific officer, Brett Kagan, characterized the behavior as adaptive, real-time, goal-directed learning.

Researchers quoted in the coverage urged caution about how far such results should be read. The neurons were not experiencing the game in any human sense, and performance remained far below that of a casual player. The interest lay in energy and adaptability. A biological network draws a tiny fraction of the power of the silicon systems used to train conventional game-playing agents. The Specola notes the demonstration and its documented limits.