When Cortical Labs unveiled its biological computer playing Doom this week, the tech world took notice for the obvious reason: yes, living neurons can run a video game on silicon. But the real story is not the novelty. It is what the breakthrough reveals about a fundamental shift in how we think about computing infrastructure.
The Melbourne-based company demonstrated its CL1 biological computer using around 200,000 living human neurons on a microchip to control the 1993 shooter. The neurons sit on a microelectrode array, kept alive in nutrient solution, connected to electrodes that both stimulate them and record their electrical responses. Software translates game data into electrical signals that the cultured cells can process.
This is not a gimmick.The CL1 is marketed as the first biological computer enabling medical and research labs to test how real neurons process information, offering an ethically superior alternative to animal testing while delivering more relevant human data. The Doom demo serves a purpose: it proves that a biological neural network can learn to navigate a complex, three-dimensional environment and adapt behaviour in response to consequences.
The Road to Doom
Cortical Labs achieved a landmark in "wetware" computing in 2021 by training a biological chip to play Pong, with neurons learning to receive sensory feedback and send motor commands to control the game's paddles. Moving from a bouncing square to a 3D maze filled with enemies was exponentially harder.The primary obstacle was translating Doom's visual data into electrical patterns that eyeless neurons could interpret; independent developer Sean Cole solved this challenge in just one week, teaching the neurons to navigate the game's 3D environment.
From a fiscal perspective, the efficiency gains are striking.While large AI models consume megawatts in massive data centres, the CL1 runs on less than a thousand watts per year. For data centre operators and organisations running resource-intensive neural networks, this matters. Energy costs are not peripheral to AI infrastructure; they are central to economics.
Where This Goes
Cortical Labs is not pitching this as a replacement for your laptop.In experiments, biological networks have often outperformed deep reinforcement learning algorithms in sample efficiency and learning improvement, suggesting niche applications where small, efficient learning systems matter more than raw processing speed.
In medicine, the CL1 could serve as a testing platform for neurological drugs, simulating the behaviour of real human networks without needing lab animals, and could aid in studying degenerative diseases such as Alzheimer's and Parkinson's by allowing researchers to observe drug impacts directly on human neurons in real time. This is where the value concentration lies: high-cost, high-stakes research where conventional silicon either fails to replicate human biology accurately or consumes too many resources to justify the expense.
Yet scepticism is warranted.The lifespan of neural cultures is around six months, after which the neurons lose functionality and must be replaced, and there is still no effective way to transfer memory from one generation of neurons to another. For any commercial venture betting on biocomputing, that is a serious constraint. Each new CL1 must retrain from scratch.
A Pragmatic Assessment
Biocomputing sits at the intersection of genuine innovation and speculative hype. Cortical Labs has demonstrated something real: cultured neurons can learn. They can adapt. They can be programmed to perform useful tasks.Biocomputing combines real neurons with hardware to create processing systems that use less energy and learn from smaller datasets than conventional computers, unlike traditional AI which requires minimal energy and training data to master complex tasks.
But the technology is not competing with GPUs. It is solving a different class of problems.Future computing is likely to become heterogeneous, mixing silicon, specialised accelerators, and, in narrowly defined domains, biological components. Doom running on neurons is proof of concept. The real application is drug discovery, disease modelling, and robotics where adaptability and energy efficiency create tangible economic value.
For investors and researchers watching the field, the takeaway is clear: biocomputing is not coming to replace conventional computing. It is coming to fill specific niches where biology's properties outperform silicon's constraints. That is less explosive than the headlines suggest, but more durable as a business case.