At Cortical Labs in Australia, a cluster of 200,000 living human brain cells is learning to navigate the chaotic corridors of "Doom," the iconic nineties shooter game—and it's doing so far better than anyone expected. The researchers who orchestrated this feat of biological computing aren't simply playing a prank on lab-grown neurons; they're demonstrating that brain tissue integrated onto a silicon chip can adapt to complex challenges in real time, opening doors to applications that range from drug screening to sustainable artificial intelligence.
The biological computer at the heart of this research is called the CL1, a silicon chip seeded with human neurons harvested from blood donations and coaxed into life through stem cell technology. At first, as Alon Loeffler, Cortical Labs' senior application scientist, recalls, the neurons performed like a complete novice at video games. "They were walking into walls a lot, shooting the walls, turning around, doing funny things like that," he told AFP. But with consistent feedback—delivered through precise electrical stimulation from electrodes embedded in the chip—the cells began to learn. Enemies that once survived intact eventually fell to sustained fire. The neurons were gaming.
The feat required converting "Doom's" digital world into electrical signals the biological tissue could parse. When an enemy appears on screen, specific electrodes stimulate the neurons, triggering coordinated activity across thousands of cells. Different patterns of neural firing produce different actions: move left, move right, or fire the weapon. Researchers watch this activity unfold as thousands of tiny dots dancing across a computer monitor, adjusting their inputs to shape the neurons' learning trajectory. It's a closed loop of stimulus and response that mirrors, in miniature, how any brain learns.
What makes this work significant isn't just the novelty of neurons playing games. The human brain accomplishes its feats of reasoning and adaptation while consuming only about twenty watts of power—a level of energy efficiency that silicon computing and artificial intelligence have failed to match. Brett Kagan, Cortical Labs' chief scientific and operations officer, describes the CL1 as "a more sustainable and more powerful form of intelligence," one that might help solve the escalating energy demands of modern computing. The cells themselves have a six-month lifespan and can't yet deliver the consistent, programmable results that silicon systems guarantee. But industry observers take the work seriously. "This isn't wacky science or some bunch of scammers," said William Keating, CEO of semiconductor research firm Ingenuity. "This is real science and it's making real progress."
Kagan emphasizes that biological computing isn't intended to replace traditional AI—rather, it aims to give us abilities we've never possessed. The CL1 chip can be reconfigured for robotics, real-time learning tasks, and across healthcare, medicine, drug screening, and personalized treatments. The neurons have already graduated from "Pong," proving they can master more complex challenges. As the team continues to refine its approach, the question is no longer whether brain cells on a chip can learn, but how far that learning might take us.