Melbourne-based startup Cortical Labs has successfully trained clusters of 200,000 living human neurons to play the 1993 video game Doom. The cells, cultured on a multi-electrode array, navigate and shoot via electrical stimulation and software controls, though they perform at a beginner level. This experiment adds to the eclectic list of platforms running Doom, which includes blockchain networks. Researchers used feedback rewards and artificial intelligence to train the neurons, which demonstrate adaptability despite not comprehending the game. Alon Loeffler, an application scientist, stated the project shows the inherent learning abilities of neurons outside the brain.
Cortical Labs connected living human neurons to software that converts Doom gameplay into electrical signals and neural activity into controls. The cells can move, react to enemies, and fire weapons within the classic shooter, as shown in a video posted on YouTube. This initiative continues the long-standing engineering benchmark of porting Doom to unconventional systems since its source code was publicly released. The game has previously appeared on unexpected platforms including blockchain networks, gut bacteria, pregnancy tests, PDFs, robot lawn mowers, and CAPTCHA challenges.
Alon Loeffler noted the public’s demand led to this project, saying, “the number one response we got is, can it run Doom?” The team developed a new platform for interacting with neurons via simple Python commands, which drastically accelerated development. Neurons learn through feedback signals, receiving small rewards for correct aiming and larger rewards for eliminating targets. Cortical Labs researchers used AI to refine how game information was encoded into the electrical signals sent to the cells.
Loeffler clarified the system’s limitations, stating, “The system doesn’t actually know it’s playing Doom. It’s getting electrical signals and then spitting out responses.” He emphasized that the neurons, while human-derived, lack structures for higher-order functionality and do not resemble human cognition. However, he observed evidence of neural adaptability outside the brain, indicating inherent learning capabilities. The gaming demonstration serves as a public-facing project while researchers explore practical applications for this biotechnology.
