When Scientists Taught Brain Cells in a Lab to Play Doom

It sounds like something pulled from science fiction: a clump of brain cells, sitting in a lab dish, learning how to play a video game. But in 2022, a group of researchers working on a system called DishBrain did exactly that—using living neurons to interact with a simplified version of the classic game Doom.

Not in the way a human plays, obviously. There’s no screen, no controller, no conscious decision-making. But at a basic level, these neurons were able to receive information, respond to it, and improve their performance over time. Which raises a slightly uncomfortable question: what does it actually mean to “learn”?

What They Were Working With

The experiment used living neurons grown in a lab—some derived from mouse cells, others from human stem cells. These neurons were placed on a device called a multi-electrode array, which can both stimulate the cells and record their activity.

Think of it as a two-way interface. The electrodes send signals into the neurons (input), and the neurons respond with electrical activity (output). Over time, patterns start to emerge.

This setup isn’t new by itself. Scientists have been studying how neurons behave in controlled environments for years. What made this experiment different was how they structured the interaction. Instead of just observing the neurons, they gave them something like a task.

Turning a Game Into Signals

The researchers took a very simplified version of Doom—basically a stripped-down environment where the only goal was to move a paddle left or right to hit a target. Then they translated that environment into electrical signals.

The position of the target was encoded into patterns of stimulation sent to the neurons. In response, the neurons produced their own electrical activity, which the system interpreted as movement—left or right.

So the neurons weren’t “seeing” the game, but they were receiving structured information about it and producing outputs that affected what happened next. That loop, input, response, feedback—is the core of learning systems.

How the Neurons “Learned”

Here’s where it gets interesting. The system was designed around a principle related to Free Energy Principle, which suggests that biological systems tend to minimize unpredictability in their environment. In simple terms: systems prefer stable, predictable patterns over chaotic ones.

So the researchers gave the neurons two types of feedback:

• Predictable, structured signals when they responded “correctly”
• Random, noisy signals when they didn’t

Over time, the neurons began to adjust their activity to avoid the randomness and maintain the predictable input. The result? Their performance improved. Not because they “understood” the game, but because their internal activity adapted in a way that produced better outcomes.

This Isn’t Consciousness

It’s important not to over-interpret what happened. These neurons weren’t thinking. They weren’t aware, and they didn’t know they were “playing” anything. What they were doing was adapting and changing their behavior based on feedback. That’s something even very simple biological systems can do.

Still, it blurs a line people tend to assume is very clear: the line between passive biological matter and something that can process information and respond in a goal-directed way.