Summary written for researchers on The Bond Lab. Not affiliated with Cortical Labs. See Cortical Labs’ official video and linked reporting for primary material.
Dish of neurons learn to play Doom
There is a rule that all the nerds know. If something has some processing power, however small and some sort of screen, someone will eventually try to make it run the 1993 video game Doom. We’ve seen it run on digital pregnancy tests, lawnmowers, and smart refrigerators and microwave ovens. However, an Australian biotech startup called Cortical Labs just took a different approach. They got Doom to run on a dish of cultured neurons, or more accurately, they got the dish of neurons to play Doom.
Yes, really. About 200,000 neurons, sitting in a nutrient-rich culture medium inside a petri dish, have been spending their week wandering around digital corridors of hell trying to blast demons.
Before anyone panics about a disembodied, sci-fi super-intelligence plotting our downfall, let’s dial it back. The dish isn’t conscious. It doesn’t have eyes, it doesn’t know what a shotgun is, and it isn’t feeling a rush of 1990’s gaming nostalgia.
From “DishBrain” to the Demon Wastes
Cortical Labs isn’t some rogue outfit that suddenly decided to do this on a whim. Led by CEO Hon Weng Chong and Chief Scientific Officer Brett Kagan, the Melbourne-based team has been laying the groundwork for “biological computing” (often called wetware) for years.
You might remember them from headlines back in 2022. Their original system, dubbed “DishBrain,” made waves globally when they successfully trained a larger cluster of about 800,000 neurons to play the 1970s arcade classic, Pong.
But Pong is just a single vertical axis, moving a paddle up and down to hit a bouncing ball. Doom is a lot more complex. It requires navigating a semi three-dimensional environment, spotting threats, and prioritizing whether to move, turn, or shoot all at the same time, as well as trying not to get killed. The jump from Pong to Doom represents a massive leap in how much parallel information these cells can handle.
How You Program a dish of neurons?
The backstory of where these cells came from is just as fascinating as the story itself. The neurons didn’t come from someone’s brain; they actually started as a blood donation from the company’s own CEO, Hon Weng Chong. The lab took white blood cells from his sample and used modern cellular reprogramming to reverse their biological clocks back into stem cells. The induced pluripotent stem cells were then coaxed to differentiate into functioning human neurons and then layered onto a specialized piece of hardware called the CL1 platform, a glass microchip about the size of a coin covered in tiny electrodes. Because electricity is the native language of both silicon computers and human nervous systems, the chip acts as a literal translator, an interface between the neurons and the computer running Doom.
The Method: How a Petri Dish Learns to Shoot
To get the cells to interact with a 90s shooter, Cortical Labs built a closed feedback loop. The neurons can’t see the game environment as we can, so the video feed of Doom is mapped into electrical signals. If a demon jumps out on the left side of the screen, the chip sends an electrical signal to the left side of the neuron culture. The neurons react by firing off their own tiny electrical spikes. The computer monitors these signals and translates distinct firing patterns into character actions, like “turn right” or “fire weapon.” The cells actually learn how to get better at the game and how to avoid taking damage from the hell demons based on a rule called the Free Energy Principle. Essentially, biological cells hate chaos and prefer predictable environments. When the neurons make a correct move, the system feeds them structured, predictable electrical signals. When they make a mistake or die, the system jolts them with completely random, chaotic static. To escape the annoying static, the neurons naturally reorganize their connections to play the game better.
You can watch the cells in action and hear the researchers explain the setup in Cortical Labs’ official release video, Living Human Brain Cells Play DOOM on a CL1 on YouTube.
So, How Good Are They?
To be blunt, they are not going to win any Doom deathmatches.
“Right now, the cells play a lot like a beginner who’s never seen a computer and, in all fairness, they haven’t,” noted Brett Kagan. The neurons spend a lot of time running face-first into grey walls, spinning aimlessly in circles, and getting taken down by low-level enemies. A bit like when your parents had a go at playing Doom.
However, they are objectively learning. When left to fire randomly, a player does nothing. These cells are actively seeking out targets, turning towards stimuli, and firing. They’ve even caught the attention of independent developers; a data scientist recently managed to hook up a basic Python script to the system to help the neurons streamline their targeting, allowing them to actively hunt in-game objects within a week of training.
Why Do This?
It’s easy to look at this as an elaborate, tech-nerd stunt, but the real-world implications are huge.
Traditional silicon AI hardware requires massive data centers and melts through an incredible amount of electricity. The human brain, on the other hand, runs on about the same amount of energy as a dim incandescent light bulb (around 20 watts) while processing complex tasks instantly.
By proving that living cells can be integrated onto programmable chips via cloud access, Cortical Labs hopes to pioneer ultra-low-energy hybrid computers. Beyond raw computing, these chips allow medical researchers to test how new drugs affect living human brain tissue in real-time, potentially revolutionizing how we study neurological diseases or design advanced biological control systems for robotic prosthetics.
For now, though, the cells are just grinding away in the lab. And true to internet form, the moment the news broke, people were already asking the researchers the inevitable next question: When are they going to try playing Fortnite?