Lab-grown neurons learning to play a video game in real time
In a laboratory in Melbourne, living human neurons grown from donated blood stem cells have been wired onto silicon chips and trained to play a 1993 video game — a quiet but consequential moment in the long human effort to understand the mind by building with it. Cortical Labs has demonstrated that biological neural networks can receive input, adapt in real time, and perform measurable computational tasks, blurring the boundary between the living and the logical. This is early work, but it asks a question that will not easily be set aside: if biology can learn to play, what else might we ask it to do?
- 200,000 living human brain cells, grown from blood donations and mounted on silicon chips, have been successfully trained to control gameplay in Doom — in real time.
- The achievement unsettles a foundational assumption: that computation belongs to silicon, and that biology is too messy, too slow, too alive to be reliably harnessed.
- Cortical Labs is navigating the tension carefully, framing the work as exploratory while pointing toward future applications in drug testing, disease modeling, and energy-efficient computing.
- Unlike binary processors, these neural networks strengthen and weaken their own connections through experience — meaning they don't just execute instructions, they learn from them.
- The path forward remains open and uncertain, but the demonstration has cracked a door: biological computing is no longer theoretical, it is measurable and reproducible.
In an Australian laboratory, living human brain cells grown from blood-donation stem cells have been integrated onto silicon chips and taught to play Doom — the 1993 shooter that now, improbably, serves as a benchmark for biological computing.
Each hybrid system contains roughly 200,000 neurons. They are not conscious, and no one is claiming otherwise. But they are alive, networked, and adaptive in ways that traditional processors are not. The cells receive information about the game state, process it through their own electrochemical connections, and generate outputs that move and fire in real time. They learn. They improve.
The researchers at Cortical Labs are deliberate about framing this as early-stage work. But the implications are real. Where silicon operates on binary logic, biological neural networks run on something richer — feedback loops and adaptive connections that shift with experience. Such systems may prove better suited than conventional computers for pattern recognition, noisy-environment decision-making, and adaptive learning, potentially at a fraction of the energy cost.
The source material matters too. Blood donations are routine and ethically established, and the stem cells they yield can be reliably coaxed into neurons. The pipeline from donation to functional neural network is becoming reproducible.
No one expects lab-grown neurons to replace laptops. But the fact that they can be trained to perform a specific, measurable task — and that the boundary between biological and computational is proving far more permeable than assumed — changes what questions science is now permitted to ask.
In a laboratory in Australia, something that sounds like science fiction has become routine: living human brain cells, grown from the stem cells in blood donations, are wired onto silicon chips and taught to play video games. Specifically, Doom—the 1993 shooter that defined a generation of computer gaming and now, improbably, serves as a benchmark for what lab-grown neurons can do.
Each of these "biological computers" contains roughly 200,000 living brain cells. They are not brains. They are not conscious. But they are alive, networked, and responsive in ways that traditional silicon processors are not. The cells were grown from stem cells extracted from blood donations—a source both abundant and ethically straightforward. Once cultivated, they were integrated onto a silicon chip, creating a hybrid system where biological and computational elements work in tandem.
The researchers behind this work are at Cortical Labs, a biotech outfit focused on understanding and harnessing how neural networks actually function. What they've demonstrated is that these lab-grown neurons can learn to control a video game in real time. The cells receive input about the game state, process that information through their own networked connections, and generate outputs that move the player's character and fire weapons. It works. The neurons learn. They improve.
This is not a parlor trick. The implications ripple outward in multiple directions. Traditional computers operate on binary logic—ones and zeros, on and off. Biological neural networks operate on something far messier and far richer: electrochemical signals, feedback loops, adaptive connections that strengthen or weaken based on experience. A biological computer might solve certain kinds of problems—pattern recognition, adaptive learning, decision-making in noisy environments—in ways that silicon struggles with. It might do so while consuming far less energy. It might do so in ways we don't yet fully understand.
The researchers are careful to frame this as early work. They are, as they put it, scratching the surface. No one is suggesting that lab-grown neurons will replace traditional computers for everyday tasks. But the demonstration that they can be trained to perform a specific, measurable computational task—playing a video game—opens a door. If neurons can learn Doom, what else might they learn? What other kinds of problems might biological computing systems be suited to tackle?
The source material for these neurons matters too. Blood donations are routine, accessible, and ethically established. The stem cells derived from them can be coaxed into becoming neurons. This is not a scarce resource. It is not a morally fraught one. The path from donation to lab-grown neural network is becoming clearer and more reproducible.
What happens next is uncertain. Cortical Labs and other researchers in this space are exploring applications in drug testing, disease modeling, and fundamental neuroscience. But the fact that lab-grown neurons can now be trained to play a video game—that they can learn, adapt, and respond to real-time input—suggests that the boundary between biological and computational is far more permeable than we once thought. The neurons in that silicon chip are not thinking about Doom. But they are, in some measurable sense, playing it. And that changes what we might ask of biology in the future.
Notable Quotes
Researchers say they are just scratching the surface of what the neurons could be capable of doing— Cortical Labs researchers
The Hearth Conversation Another angle on the story
So these are actual human brain cells, grown in a lab, playing a video game. How does that even work mechanically?
The cells are networked together on a silicon chip. They receive input about the game state—where enemies are, where the player is—and that input triggers electrical activity across the network. The cells' connections strengthen or weaken based on what they experience. Over time, the network learns which patterns of activity lead to successful gameplay.
And they're learning? The cells themselves are learning?
Yes. That's what makes it different from a programmed algorithm. You're not writing code that says "if enemy appears, fire." You're letting the neurons figure out their own solutions through experience. They adapt.
Where do the cells come from?
Blood donations. Researchers extract stem cells from donated blood and coax them into becoming neurons. It's accessible, reproducible, and ethically straightforward. No one had to die or be harmed.
Why Doom, specifically?
It's a clear, measurable task. The game has simple rules, clear inputs and outputs. It's a way to test whether the neurons can learn a specific behavior. But it's also a proof of concept—if they can learn this, what else might they learn?
What's the actual application here? Are we replacing computers with brain cells?
Not for everyday computing. But biological systems might be better at certain problems—pattern recognition, learning from noisy data, adapting to new situations. And they use far less energy than silicon. The real question is what kinds of problems are worth solving with living neurons instead of traditional chips.
Does it feel strange to you, the idea of neurons playing a video game?
It does. But that strangeness is exactly the point. We're discovering that the boundary between biological and computational is far more fluid than we assumed. The neurons aren't conscious, aren't thinking about Doom. But they are, measurably, playing it. That changes what we might ask of biology.