Cortical Labs demonstrated their CL1 biological computer running Doom using 200,000 living human neurons on February 25, 2026. The Australian biotech company, which taught neurons to play Pong in under five minutes back in 2022, tackled the significantly more complex challenge of 3D navigation, threat assessment, and real-time combat. Independent developer Sean Cole programmed the neurons to “see” the game in roughly one week using Cortical Labs’ Python-based Cortical Cloud API.
This matters because it advances the “interface problem” in biocomputing. If neurons can control complex 3D games, they could potentially control robotic arms and manage precision tasks impossible for traditional silicon chips. The breakthrough raises critical questions about using human brain cells for computation.
How Living Neurons Play Video Games
The CL1 grows 200,000 human neurons on a 59-electrode array. These neurons aren’t simulated. They’re derived from adult donor skin or blood cells, reprogrammed into induced pluripotent stem cells, then differentiated into actual brain cells.
The electrode array serves dual purposes. First, it stimulates neurons with electrical signals representing game state. When enemies appear on screen, corresponding electrode regions deliver stimulation. Second, it records neuron electrical spikes and interprets them as gameplay actions like movement, turning, and firing.
The system uses reinforcement learning. When neurons miss objectives, corrective signals nudge the neural network to reorganize its activity. Brett Kagan, Cortical Labs’ Chief Scientific Officer, describes this as “adaptive, real-time goal directed learning.” The neurons learned Doom in approximately one week, compared to their five-minute Pong learning in 2022.
Faster Learning, Beginner Execution
The learning speed is revolutionary. Traditional deep reinforcement learning took roughly 90 minutes to master Pong in 2022 experiments. Cortical Labs’ DishBrain neurons learned it in under five minutes. For Doom, the neurons needed about a week, vastly outpacing silicon-based systems on equivalent complexity.
However, current gameplay performance is beginner-level. Researchers describe it as “like a complete beginner who has never seen a keyboard, mouse, or indeed a computer before.” The neurons outperform random gameplay but lose frequently, occasionally dying repeatedly during sessions.
Here’s the honest reality: researchers admit they still don’t fully understand how the neurons are playing the game. The learning mechanisms work, but remain partially opaque. Cortical Labs expects performance to improve as algorithms advance, but biological systems resist the complete transparency we’re accustomed to with silicon.
The Real Breakthrough: Energy Efficiency
Biological neurons are one million times more energy-efficient than artificial neurons. The human brain operates on just 20 watts of power. A 30-unit CL1 rack consumes under 1,000 watts. Meanwhile, AI data centers burn megawatts.
This could enable AI systems requiring up to 100,000 times less energy than current silicon-based models. With AI’s surging energy demands threatening both sustainability and economics, biocomputing offers a potential solution. The Doom demo is impressive, but the energy efficiency is the actual story.
From Gaming Demos to Robotic Arms
Cortical Labs’ vision extends beyond viral demonstrations. Applications include controlling robotic arms with precision impossible for traditional chips, pharmaceutical drug discovery on living human neurons, and hybrid biological-silicon computing systems.
The progression from Pong to Doom demonstrates increasing complexity. If neurons can navigate 3D environments and assess threats, they could manage real-world robotic tasks. Brain-computer interfaces already enable people with tetraplegia to control robotic arms using brain signals. The CL1 takes this further by creating standalone biological processing units.
The technology is commercially available today. Units cost $35,000 each, or $20,000 in 30-unit server racks. Cloud-based “wetware-as-a-service” runs $300 weekly. The first 115 units shipped in 2025. Cortical Labs targets neuroscience research, drug discovery, and AI development.
Related: China BCI Strategic Industry: 3-5 Years to Public Rollout
The Ethical Questions We’re Not Asking
Using human neurons for computation raises unresolved ethical questions. Can advanced organoids develop consciousness or sentience? What’s the relationship between stem cell donors and the resulting biocomputers? How do we handle social inequality in access to wetware enhancement?
Organoids fall outside existing regulatory structures for human or animal research. There’s no clear legal framework. The National Science Foundation and DARPA are investing millions in organoid computing, but ethical guidelines lag behind technical capability.
Brain organoid pioneers warn against inflated claims that could trigger public backlash and damage the field. The technology is advancing faster than our ethical frameworks and regulatory systems can adapt. Developers working with or considering biocomputing need to grapple with these implications now, not after consciousness concerns become crises.
Where This Goes
Cortical Labs proved neurons can handle complex computational tasks. The learning speed advantages are real. Energy efficiency could solve AI’s sustainability problem. Commercial availability means researchers can experiment today.
Yet performance remains limited. The mechanisms stay partially mysterious. Ethical concerns grow as organoid complexity increases. Regulatory frameworks don’t exist.
The question isn’t whether biocomputing works. February’s demonstration settled that. The question is where this goes when organoids get more complex, when performance matches or exceeds silicon, when consciousness becomes a genuine possibility rather than theoretical concern.
We’re building biocomputers before we understand consciousness in organoids. That’s either revolutionary progress or a profound ethical failure. Likely both.
