Biological Computing Breakthrough: Human Neurons Learn to Play Doom

Australian biotechnology firm Cortical Labs has achieved a remarkable milestone in biological computing, successfully teaching approximately 200,000 living human neurons to play the 1993 first-person shooter game Doom. The research, conducted using the company's CL1 "biological computer," represents a significant advancement in understanding how biological neural networks can interact with digital environments.

The experiment involves growing human neurons on a microelectrode array, which is then immersed in a nutrient bath to maintain cell viability. Electrodes both stimulate the neurons and detect their electrical responses, creating a bidirectional communication channel between the biological and digital worlds. Software translates game events into electrical signals that the neurons can process, while the neurons' electrical activity is interpreted as gameplay actions.

"Pong was much simpler. There was a direct relationship. The ball went up, the paddle went up. It was a direct input-output relationship. Doom was much more complex," explained Alon Loeffler, scientist at Cortical Labs. The increased complexity of Doom, featuring 3D environments, exploration mechanics, and multiple enemies, required a more sophisticated interface between the game and the neural tissue.

The learning process operates on principles of reinforcement learning. When the neurons produce electrical activity that corresponds to beneficial actions—such as locating enemies or avoiding damage—they receive positive feedback. Conversely, detrimental actions result in negative feedback signals. Over time, the neural network adapts its activity patterns to improve performance, gradually developing basic gameplay skills.

While the performance is far from expert-level—researchers describe it as similar to a complete beginner who has never used a computer—the demonstration marks an important validation of earlier work from 2022, when the same team taught a cluster of lab-grown neurons called DishBrain to play Pong. That experiment demonstrated that cultured neurons could exhibit goal-directed learning when connected to a simulated environment.

The potential applications of this technology extend beyond novel gaming experiences. Researchers hope that studying how biological neurons learn and adapt in controlled digital environments could accelerate drug discovery for neurological conditions and inspire new approaches to computing architecture. Biological neural networks may offer advantages in energy efficiency and pattern recognition compared to traditional silicon-based processors.

From a compliance perspective, this research raises important questions about the ethical boundaries of biological computing. As neural interfaces become more sophisticated, regulatory frameworks will need to address issues of consciousness, autonomy, and the potential for creating hybrid biological-digital systems that blur traditional definitions of life and machine.

Cortical Labs continues to refine its biological computing platform, with the long-term goal of creating more sophisticated neural models that can perform increasingly complex computational tasks. The Doom experiment serves as both a technical demonstration and a testament to the remarkable adaptability of biological neural tissue when properly interfaced with digital systems.

The research joins a growing body of work exploring the intersection of neuroscience and computing, including previous experiments where Doom has been run on unconventional platforms including PCB design software, satellites, and even database queries. While these demonstrations often prioritize technical feasibility over practical application, they continue to push the boundaries of what constitutes a computing platform.