Human Brain Cells on a Chip Learn to Play Doom in One Week
Key Takeaways
- ▸Cortical Labs successfully trained human brain cells on a chip to play Doom in one week using a new Python programming interface
- ▸The biological computer learned faster than traditional silicon-based AI systems despite using only ~200,000 neurons
- ▸The breakthrough dramatically improves accessibility, allowing developers without biology expertise to program living neural systems
Summary
Australian biotech company Cortical Labs has demonstrated a significant advancement in biological computing by successfully training human brain cells on a chip to play the classic first-person shooter game Doom in approximately one week. The achievement builds on the company's 2021 breakthrough when it taught similar neuron-powered chips to play Pong, but represents a major leap forward in accessibility and complexity. Independent developer Sean Cole accomplished the feat using Cortical Labs' newly developed Python programming interface, which dramatically simplified the process of programming biological computers compared to previous methods that required years of painstaking scientific effort.
The neuronal computer chip used approximately 800,000 living brain cells grown on microelectrode arrays capable of sending and receiving electrical signals. While the chip's performance fell short of human players, it exceeded random chance and demonstrated faster learning than traditional silicon-based machine learning systems. According to Brett Kagan of Cortical Labs, the breakthrough lies not in matching human brain capability, but in utilizing biological material to process information in ways that cannot be replicated in silicon. The chip used roughly a quarter of the neurons employed in the Pong demonstration, yet tackled a vastly more complex game environment.
Experts in the field have hailed the development as a significant step toward practical applications of biological computing. Researchers at the University of Reading are exploring similar technology for controlling robotic arms, viewing the Doom demonstration as analogous to a simpler version of arm control. The ability of the biological system to handle complexity, uncertainty, and real-time decision-making represents capabilities much closer to what future biological or hybrid computers will need for real-world tasks. The new Python interface's accessibility means that developers without extensive biology expertise can now program these living systems, potentially accelerating innovation in the field.
- Experts view this as a significant step toward practical applications like controlling robotic arms and handling real-time decision-making tasks
- The achievement demonstrates biological systems can process information in ways silicon cannot replicate, opening new possibilities for hybrid computing
Editorial Opinion
This development represents a fascinating convergence of neuroscience and computer science that could redefine our approach to certain computing challenges. While the performance isn't yet competitive with traditional AI, the speed of learning and the accessibility breakthrough are noteworthy—transforming what took years of specialized research into a week-long project for a general developer. The real significance may not be in gaming performance, but in demonstrating that biological substrates can be practically programmed for complex tasks, potentially offering unique advantages in energy efficiency and adaptability that silicon-based systems struggle to match.
