A bio-computer made up of human neurons learns to play Doom and blurs the line between brain and machine

In a laboratory, human neurons cultivated on a chip have accomplished a surprising feat. After mastering Pong, this experimental bio-computer is now learning to play Doom, a cult classic from the 1990s. This fascinating experiment further blurs the line between biological brains and computer machines.

How Human Neurons Cultivated on a Chip Learn the Rules of a Complex Video Game

First, envision a small laboratory box. Inside, around 200,000 human neurons cultivated in a nutrient solution rest on a multi-electrode array. This interface allows the chip to send electrical impulses and listen to the responses of the cells, enabling direct communication with a tiny living neural network.

This system, developed by Cortical Labs and named CL1, forms what researchers call a bio-computer. The neurons receive signals and react. Their responses are then translated into digital commands. Gradually, through training, this biological network learns to adapt its behavior.

Why Doom Has Been a Favorite Testing Ground for Machines for Thirty Years

Since its release in 1993, Doom holds a special place in computer culture. This first-person shooter has become a notable technical challenge, with engineers eager to demonstrate that various devices can run it, whether it be a tractor, a calculator, or even an unexpected object.

However, in this new experiment, the goal was not just to display the game. The true challenge was to teach the biological system to actually play. Specifically, it needed to navigate a maze, identify enemies, and trigger actions like shooting.

This challenge once seemed unrealistic. During the original demonstration with Pong, it took almost 18 months of training. Yet, in this new setup, the neurons began grasping the mechanics of Doom in less than a week. This progress is particularly remarkable.

How Researchers Transform the Visual Universe of the Game into Readable Electrical Impulses

However, a major problem arises immediately. Neurons cultivated on a chip lack eyes or a visual system capable of interpreting an image. Thus, it is impossible to directly show them the game screen. Researchers had to translate each game scenario into electrical signals.

Specifically, information about the maze, the presence of an enemy, or the direction of movement is converted into impulse patterns. The electrical responses of the neurons are then interpreted by the computer, which translates these responses into in-game actions such as turning, moving forward, or shooting.

What the CL1 Experiment Reveals About the Promises and Questions of Biological Hybrid Intelligence

So far, the initial results remain modest. Researchers explain that the cells play like a beginner discovering a keyboard for the first time. Nonetheless, they are already able to identify enemies and trigger certain attacks, indicating that artificial biological learning is indeed underway.

Moreover, an intriguing fact emerges from this demonstration. The experiment utilized four times fewer neurons than the previous one with Pong, suggesting that some biological networks may learn faster than traditional algorithms operating on silicon processors.

Finally, for scientists, the objective goes far beyond simply playing a video game. This technology could pave the way for new forms of hybrid intelligence, blending biology and computing. Consequently, this fascinating advancement also raises numerous ethical questions regarding the use of human neurons in machines.

Scroll to Top