Since the late 1990s, researchers have been growing neurons on microelectrode arrays (MEAs) and studying how these living neural cells form networks and respond to electrical stimuli. It was, at the time, just basic science, without any practical computational applications.
During the 2000s, cells in MEAs were already responding to stimuli and forming patterns of activity, which led some researchers to “condition” these cultured neural networks to perform certain behaviors through repeated stimulation.
But the big leap came in 2021, when the Australian company Cortical Labs carried out the DishBrain experiment. The goal was for neurons to not only respond to a fixed stimulus, . The vehicle chosen was the 1972 video game “Pong”.
Recently, in a test of strength that substantially increased the credibility of the field, the same Cortical created the CL-1 — a hybrid biological computer. Using around 200,000 live human brain cells grown on a microchipthe system was able to play the classic first-person shooter (FPS) game “Doom”, a cultural phenomenon from 1993.
From “Pong” to the doom of “Doom”: burning neurons
The evolution of “Pong”, one of the simplest games in computational terms, to “Doom” represents a genuinely complex cognitive challenge. The three-dimensional environments and constant unpredictable encounters with enemies pose an enormous challenge for 200,000 neurons grown on a plate, without eyes, without a nervous system and without a body.
To make this confrontation possible without any evolutionary context for brain cells, researchers had to translate the digital world of the video game into the natural language of biology: electricity. The cell culture was positioned on a plate with multiple electrodes.
The dynamics are impressive. When the Imp creature appears on the left side of the screen, the electrodes stimulate that specific region of the neural culture, causing the neurons to react, firing electrical signals back. If the shooting pattern is recognized by the system, the character shoots or moves quickly.
What is most fascinating about this entire operation is that no one explicitly programmed the neurons to associate a specific pattern with shooting or moving the character. The cells themselves learned to make this association through feedback, spontaneously developing useful functional patterns to interact with the game environment.
Even though it behaved like a functional noob (an inexperienced player, but can perform its role) and lost matches, the system reached its current level of performance much faster than many traditional silicon-based artificial learning systems. AIs trained to play games, for example, generally need millions of simulated games to achieve similar performance.
Learning: a way out of Hell?
Making real neurons play “Doom” is an experience that goes beyond mere entertainment or scientific curiosity. This is real proof that the hybrid organic technology present in the CL-1 is viable, managing to combine the power of the human brain — energy efficiency, plasticity, fast learning — with the best of silicon: processing speed and precision.
Just like the protagonist of Doom, who needs to survive and find a way out, going through the chaos of hell, . If conventional computing goes through its own “hell” of energy efficiency and learning speed, wetware (fusion of brain tissue with hardware) can be an escape portal.
Passing the level here does not mean combat skill, but demonstrating the evolution of the cognitive process, which, in this case, is real-time adaptive learning aimed at concrete goals. In other words: it is an elegant and effective way of understanding how biological systems react and reorganize themselves under stimulus.
The system works like a continuous, fast-paced conversation between the game and the cells — the game speaks, the cells respond, the system interprets this response as action, the action changes the game, and the cycle begins again. With each round, neurons progressively reorganize to respond more efficiently — what we call learning.
