200,000 Brain Cells Learn to Play Doom
SkimNews Take
The ability of biological neurons to learn complex, real-world tasks like navigating a 3D environment suggests that their inherent plasticity offers a unique advantage over fixed silicon architectures for certain types of AI.
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Cortical Labs trained roughly 200,000 living human brain cells — grown from stem cells harvested from blood donations — to play Doom on its CL1 silicon chip, a step up from the neurons' earlier mastery of Pong
- The neurons started at beginner level, walking into walls and shooting them, before learning to target in-game enemies; even now, a single demon can take several attempts and multiple fired shots to kill
- The CL1 chip converts Doom's 3D environment into electrical signals, with specific electrodes stimulating the neurons when enemies appear and researchers monitoring thousands of dots of activity to train responses
- Chief scientific and operations officer Brett Kagan said the team is "just scratching the surface," with the chip's potential applications spanning drug screening, disease modeling, robotics, and AI-like real-time learning
- Kagan frames the CL1 as a more sustainable form of intelligence, noting the human brain runs on an estimated 20 watts of power — an efficiency he says silicon computing and AI have not replicated
- Cells have a six-month lifespan and aren't yet capable of producing consistent, programmable results, but semiconductor analyst William Keating of Ingenuity called the work "real science" making "real progress"
Why it matters: The brain's estimated 20-watt efficiency is the headline pitch — if biological computing can deliver goal-directed learning at a fraction of silicon's power draw, the energy economics of AI could shift — but the cells' six-month lifespan and inconsistent, non-programmable output mean commercial deployment is years off, not months.
Ask SkimNews
