They were walking into walls, shooting the walls, doing funny things
In a laboratory in Melbourne, a cluster of two hundred thousand human neurons grown from donated stem cells has learned to play a video game — not a simple one, but Doom, a chaotic three-dimensional world requiring spatial reasoning, target identification, and adaptive decision-making. The work of Cortical Labs, this experiment is less about gaming than about a deeper question humanity has long circled: what is the minimum condition for learning, and how close to thought can living tissue come when given a task and the feedback to grow from it? The cells began in confusion and arrived, gradually, at something resembling purpose — a small but consequential demonstration that biological computing may carry possibilities we have not yet imagined.
- Two hundred thousand lab-grown human neurons, wired into a silicon chip, are not just processing data — they are adapting, failing, and improving in real time inside a video game environment.
- The cells initially walked into walls, fired blindly, and spun in circles, behaving exactly like a mind encountering something entirely new — because that is precisely what was happening.
- Over successive sessions, the neurons began targeting enemies with increasing accuracy, transforming what had been random electrical noise into something measurably closer to goal-directed behavior.
- Cortical Labs frames Doom as a proof of concept, but the underlying tension is larger: if biological tissue can learn spatial reasoning through feedback alone, the ceiling of biological computing is suddenly much harder to see.
- Researchers are deliberately cautious about what comes next, watching the cells, taking notes, and sitting with a question they are not yet ready to answer.
Inside a chip at a laboratory in Australia, two hundred thousand human brain cells grown from donated stem cells are learning to play Doom — the 1993 first-person shooter that once defined a generation of gaming. The company behind the experiment, Cortical Labs, has built what it calls a biological computer: a system that uses the actual networking logic of living neurons to perform tasks, rather than simulating that logic in silicon.
The neurons did not begin with Doom. They started with Pong, the simplest possible test of directional response, before being introduced to a far more demanding environment — one requiring spatial orientation, enemy identification, and real-time decision-making. When the cells first encountered Doom, they were lost. Senior application scientist Alon Loeffler described their early behavior as resembling someone picking up a controller for the very first time: walking into walls, firing at nothing, spinning without purpose.
But the cells were receiving feedback from the game, and they were changing in response to it. Gradually, the random behavior gave way to something more deliberate. The neurons began targeting enemies — imperfectly, requiring multiple attempts, but with measurable and improving accuracy. What the experiment demonstrated was not a simulation of learning but learning itself, unfolding in living tissue responding to stimuli.
Cortical Labs is careful to say they are still mapping the boundaries of what biological computing can do. Doom was a demonstration, a way of proving the system is real and capable of growth. The larger questions — what other tasks these systems might eventually master, and what that would mean — remain open. For now, the researchers are watching, recording, and preparing for whatever comes next.
In a laboratory in Australia, something unusual is happening inside a silicon chip. Two hundred thousand human brain cells, grown from stem cells harvested out of blood donations, are learning to navigate a three-dimensional world filled with demons and gunfire. They are playing Doom, the 1993 shooter that defined a generation of computer gaming. And they are getting better at it.
This is the work of Cortical Labs, a biotech company that has developed what amounts to a biological computer—a system that harnesses the actual networking logic of the brain and puts it to work on a task. The researchers started simple. First, the neurons learned Pong, the paddle-and-ball game that requires only basic directional control. That was the warm-up. Doom is something else entirely: a chaotic, three-dimensional environment where the player must explore, orient themselves, and identify and eliminate moving targets. For a clump of cells with no prior experience, it is genuinely difficult.
When the neurons first encountered Doom, they had no idea what they were doing. According to Alon Loeffler, a senior application scientist at Cortical Labs, the cells behaved like someone picking up a controller for the first time in their life. They walked into walls. They fired at the walls. They spun in circles. They did, as Loeffler put it, "funny things." But something was happening beneath the chaos. The cells were receiving feedback. They were adjusting. Over time, their behavior began to change.
Eventually, the neurons started to target the enemies. Not perfectly—a single demon might require multiple shots, fired in different directions, before it fell. But the improvement was real and measurable. What had been random flailing became something closer to purposeful action. The cells were learning in real time, adapting their responses based on what the game was telling them. They were doing what brains do: receiving information, processing it, and adjusting behavior accordingly.
This matters because it proves something fundamental about biological systems. The neurons demonstrated that they could engage in goal-directed learning—that they could take a complex task, one that requires spatial reasoning and decision-making, and gradually improve their performance through experience. This is not a simulation of learning. This is learning itself, happening in living tissue.
The researchers at Cortical Labs are careful to note that they are still in the early stages of understanding what these biological computers might be capable of. Doom was a proof of concept, a way of demonstrating that the system works. But the implications stretch far beyond video games. If lab-grown neurons can learn to navigate a complex virtual environment, what else might they learn to do? What other tasks might biological computing systems eventually master? The researchers are not yet saying. For now, they are watching the cells play, taking notes, and preparing for the next experiment.
Citações Notáveis
They were walking into walls a lot, shooting the walls, turning around, doing funny things like that. And then eventually they started targeting the enemies more regularly and correctly.— Alon Loeffler, senior application scientist at Cortical Labs
A Conversa do Hearth Outra perspectiva sobre a história
Why start with Pong and Doom specifically? Why not something more directly useful?
Because you need to prove the principle first. Pong is simple enough that you can see if learning is actually happening. Doom is complex enough that you know it's not just reflexive response—the cells have to adapt, strategize, improve. It's the bridge between "this works" and "this could do something real."
The cells were walking into walls at first. How do you even measure that? How do you know what they're "trying" to do?
You watch what they do and you measure the outcome. Did they hit the target or not? Did they move toward the goal or away from it? Over time, the ratio changes. That's learning. You don't need to know what they're thinking—you just need to see that their behavior is getting better at accomplishing the task.
Two hundred thousand cells. Is that a lot? Does the number matter?
It's enough to create a network complex enough to learn. But honestly, we don't fully know yet what the minimum is, or what the maximum could be. That's part of what they're still figuring out.
What happens next? Do they try harder games?
Probably. But I think the real question is whether you move beyond games entirely. Can these cells learn to recognize patterns in medical data? Can they help solve problems that human brains struggle with? That's where the actual value lies.