Human Brain Cells Learn to Play Doom in Cortical Labs Experiment

Cortical Labs Connects Human Neurons to Classic Game Doom

Melbourne-based startup Cortical Labs has successfully trained clusters of living human brain cells to navigate the 1993 video game Doom. Researchers connected approximately 200,000 human neurons to software that translates gameplay into electrical signals and interprets neural activity into in-game controls. This allows the cells to move, react to enemies, and fire weapons, though their current performance is comparable to beginners.

The "Can it run Doom?" Challenge

The experiment builds on previous work, including running Pong on neurons in 2021. Doom has long served as an unofficial benchmark for testing new systems, with its source code release in 1997 leading to its porting on numerous unconventional platforms. These include gut bacteria, blockchain networks, PDFs, robot lawn mowers, and CAPTCHA challenges.

Cortical Labs utilized their CL1 device, which houses the cultured human neurons on a multi-electrode array. Initially, the team used improvised low-level computing code, but later developed a platform allowing interaction with neurons via high-level controls using Python commands. This accelerated development, reducing the coding time significantly.

Learning and Adaptation in Neurons

The neurons learn through feedback signals, receiving rewards for correct actions like aiming at enemies and larger rewards for successfully eliminating targets. Artificial intelligence was employed by Cortical Labs to refine how game information was encoded into electrical signals sent to the neurons, aiming to guide the cells' responses.

While the cells demonstrated improvement, researchers noted that the neurons react to inputs rather than possessing true understanding of the game. The system does not recognize it is playing Doom; it simply processes electrical signals and generates responses. This necessitates a different approach to programming compared to traditional computing systems.

Despite using human-derived neurons, the system does not replicate human cognition. The cells lack pain receptors and structures for higher-order functionality. However, researchers observed evidence of neural adaptability and learning in response to the environment, highlighting the inherent abilities of neurons to adapt.