Key takeaways
- Cortical Labs trained 200,000 neurons on CL1 to play Doom, advancing bioinformatics.
- Brett Kagan says 20 watts of brain efficiency could then meet AI’s power demands.
- CL1 cultures last approximately 6 months; FDA and NIH oversight could shape future uses.
In a Melbourne laboratory, a dish of 200,000 human neurons learned to strafe and shoot in Doom, thanks to a silicon interface. Cortical Labs’ CL1 chip translated the game world into electric models and read back the spikes in the form of movement and fire, pushing the cultivation dish from Pong reflexes to 3D navigation. The piece is still clunky, but it hints at energy-consuming biological computing alongside today’s power-hungry AI, a direction the team believes complements conventional models. Extend the lifespan by six months and tighten up the consistency, and the same wet software could run robots or filter drugs, not just hunt pixelated demons.
Human Neurons Confront Doom in Lab Breakthrough
Some experiments feel like a glimpse into the next chapter of computing. Researchers at Cortical Labs report training a group of 200,000 neurons to play Doom, the 1993 first-person shooter that helped define modern gaming. The neurons, grown from human stem cells and connected to a silicon interface, learned to navigate corridors and shoot enemies, hinting at a path for biocomputers that complement today’s AI systems.
How human neurons learn to play
The team started with Pong-level behavior, then moved to the 3D requirements of Doom. The neurons received structured electrical signals related to the game state and responded with patterns that the system translated into commands such as move, turn, and shoot. At its heart is the custom CL1 chip, which converts visual events into stimulation via electrodes, then reads cell activity to drive actions in real time.
Performance is far from esports ready. Cells often misfire or overcorrect, then improve over repeated sessions as training continues. The goal, the researchers say, is not a perfect goal but the demonstration of targeted learning within a living neural network, under conditions that a computer can orchestrate and measure.
The promise of biological effectiveness
Energy is making headlines. While today’s large AI models consume megawatts in cloud data centers, the human brain operates at around 20 watts. This efficiency inspires research into hybrid systems that can reduce energy requirements for learning, adaptation and control. Brett Kagan, chief scientific officer of Cortical Labs, sees this work as a partner to AI on silicon, not a replacement, especially for tasks that benefit from continuous learning with tight energy budgets.
For U.S. companies training base models on Nvidia GPUs and racing toward large-scale inference, even a partial move to biological coprocessors could matter. Think local learning loops for robotics or edge devices, while conventional chips handle precision math and large-scale retrieval. The short-term question is where the trade-offs lie in terms of latency, reliability and cost.
A future beyond gaming
The game is a practical test bed, but the broader target is science and industry. Biological computing could enable drug screening of patient-specific neural tissues, new disease models, and adaptive controls in robotics. The interfaces remain fragile, with a typical lifespan of around six months and outputs that are not yet fully standardized or programmable on a large scale.
https://www.youtube.com/watch?v=-CSEEXKTuY
Regulatory and ethical guardrails will need to keep pace, particularly in the United States, led by the FDA and NIH, if medical uses advance. However, the laboratory result is concrete: living neurons can be trained to act on complex digital tasks. From Doom to data centers, the journey began, silently and efficiently, inside a dish.
