They were walking into walls, shooting the walls, doing funny things
In a laboratory in Australia, two hundred thousand human neurons grown from donated blood cells have learned to navigate the digital corridors of Doom, adapting their behavior in real time through electrical pulses on a chip no larger than a postage stamp. This is not a stunt — it is a quiet demonstration that biological matter, given the right conditions, can learn, adjust, and improve in ways that silicon alone has never matched. The researchers at Cortical Labs are asking an old question in a new way: what if the most efficient computer we will ever build is the one that already lives inside us?
- Two hundred thousand lab-grown human neurons are playing Doom — stumbling at first, then learning to target enemies with increasing accuracy across repeated sessions.
- The tension is not in the game itself but in what it reveals: biological cells on a chip are demonstrating real-time neural plasticity, the same adaptive learning that underlies human cognition.
- As AI data centers consume electricity at industrial scale, a human brain running on twenty watts poses an urgent and uncomfortable question about the future of computing.
- The CL1 chip is being positioned not as a rival to artificial intelligence but as a complement — capable of drug screening, disease modeling, and personalized medicine in ways silicon cannot replicate.
- The technology remains fragile: cells live only six months, results are inconsistent, and the science is early — but credible researchers say the progress is real and the implications are vast.
Inside a chip no larger than a postage stamp, something quietly remarkable is unfolding in an Australian laboratory. Two hundred thousand human brain cells, grown from stem cells derived from blood donations, have learned to play Doom — the legendary 1990s shooter — adapting in real time to a digital world translated into electrical pulses they can process and respond to.
The team at Cortical Labs built a specialized device called the CL1, whose surface is lined with thousands of electrodes that both stimulate the neurons and read their activity. When an enemy appears in the game, specific electrodes fire; the neurons react; different patterns of activity produce different outputs — move, turn, shoot. The cells began their education with Pong, then graduated to Doom's chaotic three-dimensional world, where they initially walked into walls and fired at nothing. Over time, something shifted. They began recognizing patterns, targeting enemies more consistently, learning the relationship between action and outcome.
What matters here is not the score but the mechanism: real neural plasticity, happening on a chip. The researchers converted digital environments into biological signals, monitored the responses, and adjusted inputs — training a living system the way one might train an artificial neural network. The cells adapted. They improved.
Brett Kagan, Cortical Labs' chief scientific officer, frames this as a proof of concept for something far larger. The CL1 can be reprogrammed for drug screening, disease modeling, robotics, and personalized medicine. More fundamentally, the human brain's twenty-watt power efficiency dwarfs anything silicon computing has achieved — a fact that grows more urgent as AI data centers consume electricity at industrial scale. Biological computing, Kagan argues, is not meant to replace AI but to offer capabilities we have never had before.
The cells currently live only six months and produce inconsistent results. The technology is young. But credible voices in the semiconductor world are paying attention, seeing beneath the spectacle a genuine scientific advance — one that asks what becomes possible when intelligence is no longer assumed to require silicon, and biology itself is allowed to teach us something about computation.
In a laboratory in Australia, something quietly remarkable is happening inside a silicon chip no larger than a postage stamp. Two hundred thousand human brain cells, grown from stem cells harvested out of blood donations, have learned to play Doom—the legendary 1990s shooter that defined a generation of gamers. They are not perfect at it. They walk into walls. They fire at nothing. But they are learning, adapting in real time to a digital world that has been translated into patterns of electrical pulses they can understand. And the researchers behind this work say they have barely begun to explore what these neurons might become.
The team at Cortical Labs, a biotech company working at the intersection of neuroscience and computing, built a specialized chip called the CL1 to make this possible. The neurons live on the chip's surface, connected to thousands of electrodes that can both stimulate the cells and read their electrical activity. When an enemy appears in the game, specific electrodes fire, triggering the neurons to react. Different patterns of neural activity produce different outputs—move left, move right, fire the weapon. A computer screen displays the neurons' activity as thousands of tiny dots, a visual representation of thought happening in a petri dish.
The journey to Doom began elsewhere. The cells first mastered Pong, that simplest of digital games, where a paddle moves up and down to volley a ball across a screen. When they graduated to Doom's chaotic three-dimensional world, they arrived as beginners in the truest sense. Alon Loeffler, a senior application scientist at Cortical Labs, watched them stumble through their first attempts. "They were walking into walls a lot, shooting the walls, turning around, doing funny things like that," he recalled. But something shifted. The neurons began to recognize patterns. They started targeting enemies more regularly, more correctly. A demon might take several shots from multiple directions before falling, but the cells were learning the relationship between action and outcome—the fundamental basis of goal-directed behavior.
What makes this work significant is not that the cells play a video game well, but that they demonstrate something deeper: real-time learning and adaptation. The researchers converted the digital environment into electrical signals the neurons could process. They monitored the cells' responses and adjusted their inputs accordingly, essentially training a biological system the way you might train an artificial neural network. The cells adapted. They learned. They improved. This is not simulation or metaphor. This is actual neural plasticity happening on a chip.
Brett Kagan, the chief scientific and operations officer at Cortical Labs, frames the work as a proof of concept for something far larger. The CL1 chip is not limited to games. It can be reprogrammed for drug screening, disease modeling, robotics, machine learning tasks that mimic artificial intelligence, even personalized medicine. The applications stretch across healthcare and beyond. But there is a more fundamental reason to pursue this line of research, one rooted in physics and efficiency. The human brain operates on approximately twenty watts of power—a level of energy efficiency that silicon computing and artificial intelligence have not come close to matching. As data centers and AI systems consume ever more electricity, the question of how to do more with less power becomes urgent. Biological computing, if it can be made reliable and scalable, offers a different path.
Kagan is careful not to position this as a replacement for artificial intelligence. "It's not aimed to replace what AI is doing," he said. Rather, it is meant to "give us abilities that we've never had before." The cells currently have a lifespan of six months and do not yet produce consistent, programmable results. The technology is young. But William Keating, CEO of the semiconductor research firm Ingenuity, sees the substance beneath the spectacle. "This isn't wacky science or some bunch of scammers," he said. "This is real science and it's making real progress." The neurons playing Doom are not the endpoint. They are a signal of what becomes possible when you stop thinking of intelligence as something that must be silicon-based, and start asking what biology itself might teach us about computation, learning, and efficiency.
Notable Quotes
We are just scratching the surface of what these neural cultures can achieve when integrated in systems like our CL1.— Brett Kagan, chief scientific and operations officer, Cortical Labs
This is real science and it's making real progress.— William Keating, CEO of Ingenuity semiconductor research firm
The Hearth Conversation Another angle on the story
Why does it matter that brain cells can play a video game? Isn't that just a parlor trick?
It's not about the game itself. It's that the cells learned. They adapted their behavior in response to feedback in real time. That's the hard part—not the playing, but the learning.
But we have artificial intelligence that learns far better than cells in a dish.
True, but AI systems running on data centers consume megawatts of power. Your brain runs on twenty watts. If we could harness that efficiency, we'd solve one of computing's biggest problems.
So this is really about power consumption?
Partly. But it's also about discovering whether there are kinds of problems that biological systems solve differently than silicon does. We don't know yet what we don't know.
What happens to the cells after six months?
They die. That's the current limitation. We can't keep them alive indefinitely, and we can't yet make them produce reliable, repeatable results. But the fact that they work at all is the breakthrough.
What would make you believe this is actually going somewhere?
When we can keep the cells alive longer, when we can make their responses predictable, when we can scale it beyond a single chip. Right now we're still in the proof-of-concept phase. But the proof is real.