Australian Researchers Train Lab-Grown Brain Cells to Play Doom, Demonstrating Biological Computing Potential
Key Takeaways
- ▸Lab-grown brain cells containing ~200,000 human neurons successfully learned to play Doom, adapting from novice to progressively better performance
- ▸The Cortical Labs CL1 chip converts digital game environments into electrical signal patterns, enabling bidirectional communication with biological neural networks
- ▸Biological computing demonstrates 20-watt energy efficiency compared to silicon AI, offering a sustainable alternative computing substrate with applications in drug screening, disease modeling, robotics, and personalized medicine
Summary
Australian biotech company Cortical Labs has successfully trained lab-grown brain cells to play the 1990s shooter game "Doom," marking a significant milestone in biological computing research. The "biological computer" uses approximately 200,000 living human brain cells—derived from stem cells harvested from blood donations—grown on a silicon chip called the CL1. After mastering the simpler game "Pong," the neurons adapted to the complex 3D environment of Doom, learning to navigate, target enemies, and complete goal-directed tasks in real time.
The breakthrough demonstrates that biological neural networks can learn and adapt to digital stimuli with remarkable efficiency. Researchers converted the game's digital environment into electrical signal patterns that the neurons could interpret, with different neuron activity patterns triggering specific actions like firing weapons or moving. While the neurons initially performed poorly—walking into walls and shooting randomly—they progressively improved their targeting accuracy, proving their capacity for real-time adaptation.
Cortical Labs emphasizes this is just the beginning of what biological computing can achieve. The CL1 chip can be reprogrammed for diverse applications beyond gaming, including drug screening, disease modeling, robotics, machine learning tasks, and personalized medicine. The human brain's remarkable energy efficiency—running on approximately 20 watts of power—far exceeds current silicon-based AI systems, positioning biological computing as a complementary technology rather than a replacement for traditional artificial intelligence.
- While neurons currently have a six-month lifespan and inconsistent programmability, researchers view this as proof-of-concept for entirely new classes of hybrid biological-digital systems
Editorial Opinion
This research represents a fascinating inflection point in computing technology—not a replacement for AI, but a fundamentally different paradigm for creating adaptive, learning systems. The ability to harness biological neural networks' remarkable efficiency and plasticity opens possibilities that pure silicon computing cannot match, particularly for personalized medicine and real-time adaptive tasks. However, the current limitations (six-month lifespan, consistency issues) highlight why this is early-stage research; the real value will emerge as researchers solve the engineering challenges of maintaining and programming stable biological systems at scale.
