When the joystick is replaced by a petri dish, who is really playing whom? Prepare to rethink what learns, what computes, and what wins.
After teaching living neurons to play Pong, Australia’s Cortical Labs has pushed its bio-computer into far darker corridors: Doom. Within a week, the CL1, built on roughly 200,000 human neurons atop a multi-electrode array, began aiming at enemies as the game’s visuals were piped in as electrical impulses. Working with independent researcher Sean Cole, the team hints at advantages for neuron-based processing over traditional algorithms, even if the gameplay remains novice-level. With an open API, they are inviting others to refine learning rules, encodings and rewards to see how far this curious player can go.
In March 2026, Cortical Labs shared a remarkable update: a bio-computer built from human neurons learned to play Doom. The feat anchors a new chapter where living tissue meets computation with purpose. It also extends a line of work begun with DishBrain (first shown in 2022 with Pong). The tone is cautious yet confident, and the ambition is clear: make neurons compute in useful, testable ways.
A breakthrough in neuron-based computing
DishBrain proved that neurons could adapt to a simple feedback loop, batting a virtual ball with surprising speed. By 2025, the team moved to the commercial CL1 system, offered as hardware or through the cloud. Doom, released in 1993, is a tougher proving ground. It demands perception, decision-making, and timing—an enduring benchmark for AI systems and hackers alike.
From Pong to Doom: how we’ve come
To bridge biology and the 3D maze, Cortical Labs partnered with researcher Sean Cole. Visual scenes were translated into electrical stimuli, and neuronal responses were converted into commands. In 1 week, the culture showed novice play: finding targets and firing with basic competence. Notably, the setup ran on only 200,000 neurons (200,000 neurons on a multi-electrode array), highlighting efficient adaptation through tightly tuned feedback.
Rethinking how computers learn
The CL1 suggests that neuron-based processors could complement—or even rival—traditional algorithms in some tasks. Today, Doom performance remains basic, roughly akin to a first-time player. Yet the learning loop is real and improvable (via a public API). Cortical Labs is asking researchers to stress-test the stack and iterate on what works in living computation.
- Refine encoding schemes that better map game states to neural stimuli
- Design richer reward signals to stabilize and speed learning
- Establish fair benchmarks to compare bio-computers with silicon models
What the future holds
Beyond gaming, this hybrid approach invites new strategies for data analysis, robotics, and neuroscience. Neurons excel at plasticity and pattern discovery, traits that could complement brittle code paths in edge cases. According to Cortical Labs, these systems may fill roles where conventional AI stalls. So where does this lead next? To platforms where living intelligence and digital infrastructure learn side by side, step by measured step.
