Biological Computing Milestone: Lab-Grown Neurons Learn to Play Doom

Researchers have successfully trained a lab-grown cluster of human neurons to play the classic video game Doom. The achievement, facilitated by biological computer chips, suggests a promising future for organic computing systems in handling complex, real-time tasks.

From Pong to Doom: A Leap in Complexity

Australian biotech company Cortical Labs previously demonstrated that neuron-powered chips could play the simple arcade game Pong. While groundbreaking, the recent Doom demonstration represents a significant escalation in complexity. The system utilized approximately 200,000 living brain cells—roughly a quarter of the count used in the Pong experiments—cultured on microelectrode arrays.

A major factor in this rapid progress is the development of a new interface allowing the chips to be programmed via Python. Independent developer Sean Cole used this accessible toolset to teach the neurons to play Doom within a single week. Brett Kagan of Cortical Labs noted that this speed contrasts sharply with the years of research required for the initial Pong prototype, highlighting the flexibility of the new platform.

Performance and Potential

While the biological player performed better than a random input generator, it currently falls short of skilled human players. Despite this, Kagan pointed out that the neurons learned much faster than traditional silicon-based machine learning systems. The primary goal, according to researchers, is not to compete with digital AI in gaming, but to utilize biological matter for information processing in ways that silicon cannot replicate.

Experts suggest that navigating the 3D, uncertain environment of Doom proves these systems are ready for more serious applications. Yoshikatsu Hayashi of the University of Reading compared the game's demands to controlling a robotic arm, a tangible real-world use case currently under investigation. The ability of the cells to handle decision-making without eyes or a brain structure suggests a robust utility for future hybrid computers.