Human Brain Cells Taught to Play Doom. Seriously.

This is one of those genuine "no way" moments — like when the first electric lights flicked on in a small town, or your grandparents made their first long-distance telephone call. A technological leap that feels so implausible it borders on absurd.

Artificial intelligence already mystifies most of us. Now add this: living human brain cells trained to play Doom — a game I "wasted" countless hours playing (at work! Too bad, Pfizer. You can't fire me twice) when it exploded onto computer screens in the early 1990s.

Released in 1993, Doom was unlike anything before it. The first-person shooter game dropped players into claustrophobic corridors and poisonous lakes where increasingly horrifying creatures leapt out without warning and attempted to do very bad things to you. The objective was simple: survive, advance, repeat. It was wildly fun and sometimes pretty scary.

Speaking of Scary...

Now, perhaps as an unintended tribute to that milestone in gaming history, researchers at an Australian biotech company called Cortical Labs have grown clusters of human neurons on electronic chips and trained them to interact with the game. The cells were hardly championship material — no one will mistake them for professional gamers — but they performed well enough to demonstrate something that sounds suspiciously like the plot of a tacky science fiction film. 

Before anyone imagines a conscious mini-brain gleefully blasting digital demons, let's slow down. This is not a thinking organism. It's not aware. It doesn't "see" the screen. It doesn't understand what a monster is or what weapon to choose. [1] What it is, instead, is living neural tissue grown atop a microelectrode array — a device that can both stimulate the neurons with electrical signals and record their responses. You will not find a recipe for this in the Betty Crocker Cookbook.

The system works by translating elements of the game into electrical patterns delivered to the cells. The neurons respond with their own electrical activity. That output is then decoded back into in-game actions. Over time, the cells adapt their firing patterns based on feedback — a primitive form of learning.

No, this is not a neuronal network like you're imagining. It's the Spiderdemon, one of the most popular (and evil) Doom monsters. You don't want to get shot by it. But it does want to shoot you.

Cortical Labs previously demonstrated a similar setup playing Pong in 2021. Pong is a quintessential video game for lower life forms: two paddles, predictable motion, limited decisions. Doom is much more complex. It involves navigation, uncertainty, changing environments, and real-time decisions. The neuronal system performed better than random noise but far below a skilled human player, even a Pfizer ex-employee. 

Cortical Labs made a great YouTube video that explains this stuff far better than I can. You should watch it.

The jump from Pong to Doom isn't just fun and games. It suggests that these biological networks can handle more sophisticated feedback loops than previously shown. The authors point to applications such as controlling robotic arms — a task that, according to outside experts, may actually be more complex than navigating a simplified version of Doom.

"Playing Doom is like a simpler version of controlling a whole arm” 

— Yoshikatsu Hayashi, University of Reading, UK (not involved in the work)

The most significant advance isn't really the game. It's the interface. Cortical Labs has developed a system that allows these neuron-based chips to be programmed using Python [2]. An outside developer reportedly trained the neurons to interact with Doom in about a week. That's a dramatic reduction in technical barriers compared to earlier efforts that required years of specialized work, and something neither you nor I will ever understand.

Why Use Living Neurons?

1. It's very cool. 
2. Biology is extraordinarily energy efficient. 

The human brain runs on roughly 20 watts — less than many light bulbs — while modern AI systems require energy-hungry data centers packed with GPUs. Neurons are also inherently adaptable and tolerant of noisy, unpredictable inputs.

Researchers envision potential applications in adaptive control systems, such as robotic limbs operating in dynamic environments. Compared to that, a stripped-down version of Doom is just a convenient testbed.

Sleep Well

Neither the neurons nor the Spiderdemon are out to get you. They aren't thinking, dreaming, or choosing weapons. They're responding to electrical signals in ways that happen to move a character through Doom. Still, the fact that living tissue can be wired into a machine and trained to handle something this chaotic hints at a future where biology and electronics collaborate in ways they're only beginning to understand. Me? Forget it. I'm still confused by my toaster.

NOTES:

[1] Arguably, the most popular weapon was named the BFG. I'll leave it to you to figure this one out.

[2] Python is a widely used programming language valued for its simplicity and flexibility. It is commonly used in scientific research, data analysis, and artificial intelligence development.