They were walking into walls, shooting walls, doing funny things
In a Melbourne laboratory, 200,000 human neurons grown from donated stem cells have learned to navigate a virtual world of enemies and corridors — playing Doom, and improving at it. The achievement by Cortical Labs is less about gaming than about what it reveals: that biological tissue can receive digital signals, adapt to feedback, and pursue goals in real time. It is an early chapter in a longer story about what intelligence is, where it can live, and how little power it might one day require.
- A chip smaller than a postage stamp now houses living human brain cells that began by walking into walls and ended by targeting enemies with growing accuracy.
- The tension is not just scientific — it sits at the intersection of biology and computation, raising urgent questions about the nature of learning itself.
- Researchers are actively reprogramming the same platform for drug screening, disease modeling, and AI-adjacent tasks, racing to move from proof-of-concept to practical deployment.
- A six-month cell lifespan and inconsistent reproducibility remain real obstacles standing between this breakthrough and any widespread application.
- Industry observers are taking notice: as AI infrastructure strains global power grids, a biological system that thinks on 20 watts represents a genuinely disruptive alternative.
Inside a Melbourne lab, a silicon chip no bigger than a postage stamp holds roughly 200,000 human brain cells grown from stem-cell-derived blood donations. Researchers at Cortical Labs have wired these neurons into the digital world of Doom — not by showing them a screen, but by translating the game's environment into electrical signal patterns the cells can interpret. Enemy appearances trigger specific electrode firings; the neurons' responses move the character, aim the weapon, pull the trigger.
At first, the cells performed like a first-time player — spinning, shooting walls, going nowhere. Then, gradually, something changed. The neurons began targeting enemies more consistently, adapting to stimuli and completing goal-directed tasks in real time. Senior scientist Alon Loeffler watched the fumbling give way to recognizable learning.
The implications reach well beyond gaming. The same platform, called the CL1, is being explored for drug screening, disease modeling, robotics, and personalized medicine. Chief scientific officer Brett Kagan frames it as a more sustainable form of intelligence — and sustainability is the point. The human brain runs on roughly 20 watts, a level of efficiency that silicon AI has never approached. As data centers consume ever more power, that gap is drawing serious industry attention. Semiconductor analyst William Keating calls the work real science making real progress.
Limitations remain honest and present: the cells live about six months, and consistent programmability at scale is still out of reach. But the fact that living neurons can be trained, monitored, and adjusted in real time — that they can learn a complex 3D game — suggests the surface of this technology has barely been touched.
In a laboratory in Melbourne, something unusual is happening inside a silicon chip no larger than a postage stamp. About 200,000 human brain cells, grown from stem cells harvested out of blood donations, are learning to navigate a three-dimensional world filled with demons and weapons. They are playing Doom, the 1990s shooter game, and they are getting better at it.
The researchers at Cortical Labs who built this system call it a biological computer. The cells themselves don't see pixels on a screen the way you do. Instead, the team converts the digital environment of the game into patterns of electrical signals—a language the neurons can understand. When an enemy appears in the game world, specific electrodes on the chip fire, stimulating the cells in precise ways. Different patterns of neural activity produce different outcomes: the gun fires, the character moves left, the character moves right. A computer monitors thousands of tiny electrical signatures flowing through the culture, represented as dots on a screen, and the researchers use that data to adjust their inputs and guide the learning process.
At first, the neurons performed like someone who had never held a controller. They walked into walls. They shot at walls. They spun in circles. Alon Loeffler, a senior application scientist at the lab, watched this fumbling unfold with the patience of someone witnessing genuine learning. Then something shifted. The cells began targeting enemies more regularly. They began hitting their marks. One demon might take several shots from multiple angles before falling, but the pattern was unmistakable: the neurons were adapting to stimuli in real time and completing goal-directed tasks.
This is not a parlor trick, and the researchers are careful to say so. The work proves that biological neural tissue can process digital information, learn from feedback, and improve performance—capabilities that suggest possibilities far beyond video games. The same chip, called the CL1, can be reprogrammed for drug screening, disease modeling, robotics, machine learning tasks that resemble artificial intelligence, and personalized medicine. Brett Kagan, the chief scientific and operations officer, describes the technology as "a more sustainable and more powerful form of intelligence."
That sustainability matters. The human brain operates on roughly 20 watts of power—a level of efficiency that silicon computing and current artificial intelligence systems have not managed to match. As data centers and AI infrastructure consume ever more electricity, the prospect of biological computing that achieves comparable or superior results while drawing a fraction of the power has caught the attention of industry analysts. William Keating, CEO of semiconductor research firm Ingenuity, calls it "real science" making "real progress," not speculation or hype.
There are limits worth naming. The cells have a lifespan of about six months. They do not yet produce results that are consistent and programmable enough for widespread deployment. The technology is not positioned as a replacement for artificial intelligence but as something complementary—a tool that might give us capabilities we have never had before. Still, the fact that lab-grown neurons can learn to play a complex 3D game, that they can be trained and monitored and adjusted in real time, suggests that the surface of what these cells might do has barely been scratched. The question now is what comes next.
Citas Notables
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 at Cortical Labs
This isn't wacky science or some bunch of scammers. This is real science and it's making real progress.— William Keating, CEO of semiconductor research company Ingenuity
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that brain cells can play a video game? Isn't that just a demonstration?
It's a proof of concept for something much larger. The game is the language we're using to teach the system. What matters is that biological tissue can process digital information, learn from feedback, and improve—all in real time. That's the capability.
But couldn't artificial intelligence do this faster and better?
Probably, yes. But AI systems consume enormous amounts of power. The human brain does comparable work on 20 watts. If we can harness that efficiency in a biological system, we're not replacing AI—we're opening a different door.
How long can these cells actually function?
About six months. They're not stable enough yet for consistent, programmable results. That's a real constraint. But the fact that they can learn at all in that window is significant.
What would you actually use this for beyond the novelty?
Drug screening, disease modeling, personalized medicine. You could grow cells from a patient's own tissue, test compounds on them, see how they respond. That's not a game. That's medicine.
So this is early, but the direction is clear?
Very early. But yes—the direction is toward a form of computing that's more efficient, more adaptable, and potentially more useful than what we have now.