Scientists just taught living human brain cells to play Doom. Yes, that Doom the 1993 classic where you run around a pixelated hell blasting demons with a shotgun. But before you picture some horror-movie brain-in-a-jar scenario or rogue AI takeover, relax: this is a legitimate lab experiment from Australia, and it’s equal parts brilliant and bizarre.
At Cortical Labs in Melbourne, researchers grew roughly 200,000 human neurons from adult donor cells and hooked them up to a custom microchip. These living cells, firing electrical signals exactly like they do inside your skull, learned to control basic movements in the game moving, turning, shooting, dodging. No consciousness, no thoughts, just pure biological electricity being turned into in-game actions.
A tweet from X.
They showed it off in late February 2026, and the videos are mesmerizing (and a little chaotic). The controls are jerky, sure, but the cells clearly do better than random guessing. Cortical Labs’ Chief Scientific Officer, Dr. Brett Kagan, put it best: the whole point was to prove these neurons adapt incredibly fast sometimes faster than traditional AI on simple tasks.
This is the same team that made headlines in 2022 when they taught neurons to play Pong with about 800,000 cells. For Doom they scaled down to 200,000 but stepped up the difficulty big time: a full 3D world, enemies that move, walls to navigate, the works.
Here’s the clever part. The neurons live on a multi-electrode array a grid of tiny sensors that both talks to the cells and listens to them. When an imp pops up on the left side of the screen, the system converts that into a precise electrical zap aimed at specific spots in the culture. The cells fire back in patterns, and those patterns get mapped to controls: one spike train means “shoot,” another means “strafe right,” and so on.
Software engineer Sean Cole knocked out the interface in about a week using the company’s open Python API(it’s on GitHub if you want to play around). CTO David Hogan explained it simply:
“If the neurons fire in a specific pattern, Doomguy shoots. If they fire in another pattern, he moves right, and so on.”
Training is pure feedback. Nail a good move and the cells get a nice, steady reward signal. Screw up and they get hit with noisy chaos as punishment. Everything runs on their CL1 chip, which costs about $35,000 and keeps the neurons alive in nutrient solution for up to six months. There’s even cloud access so researchers can log in from anywhere and tinker remotely.
Performance-wise? Total beginner. The culture can track enemies and pull the trigger, but it’s slow, clumsy, and nowhere near human-gamer level. Dr. Alon Loeffler from the team summed it up:
“To bridge that gap, we needed to translate the digital world of Doom into the biological language of neurons, which is electricity.”
This isn’t just a nerdy party trick. It’s a serious peek at biological computing machines that use living cells instead of silicon and sip power like crazy compared to today’s data-center hogs. Think ultra-efficient AI, better robotics that actually handle the messy real world, and drug testing that uses human brain tissue instead of mice.
Of course, the ethics conversation is already heating up. These are human-derived cells, so questions about consent, boundaries, and “what if they ever got more complex?” are real. Bioethicists (including coverage in New Scientist) are pushing hard for guidelines before things scale up.
Cortical Labs has already started shipping CL1 units to the U.S. the first 115 arrived by early 2026. For American labs and startups, it’s a whole new toolbox: biology’s insane efficiency plus silicon’s speed and reliability. Sure, neurons are finicky and fragile compared to chips, but this Doom demo just proved they can learn fast. And if that’s what 200,000 cells can do today… imagine what comes next.
