From Pong to Doom with Quarter the Cells

Cortical Labs has crossed a threshold that reframes what biological computing might become. The Australian startup taught 200,000 cultured human neurons to navigate Doom, the 1993 first-person shooter that defined a generation of gaming. The achievement matters less for its nostalgia factor than for its efficiency: the company used 75% fewer cells than the 800,000 required for their 2021 Pong demonstration.

The progression from Pong to Doom represents more than incremental improvement. Pong requires tracking a ball and positioning a paddle along a single axis. Doom demands spatial navigation, threat assessment, and decision-making across a three-dimensional environment. The neuron cultures, grown on microelectrode arrays in what Cortical Labs calls their DishBrain system, receive electrical stimulation corresponding to game state and respond with signals that control movement and actions.

Why Fewer Cells Matter More

The reduction in neuron count while increasing task complexity suggests these biological systems learn with a efficiency that silicon struggles to match. Where artificial neural networks require millions of parameters and extensive training data, cultured neurons appear to develop functional responses through mechanisms researchers are still working to fully characterize. The cells receive feedback through their electrical environment, experiencing something analogous to consequences when gameplay deteriorates.

This creates a testbed for understanding how biological intelligence emerges from cellular networks. The research has immediate applications beyond gaming: pharmaceutical companies could test drug effects on functional neural tissue, while BCI developers gain insights into how biological and electronic systems might integrate more seamlessly.

The Question of Scale

Cortical Labs hasn’t published the technical details of their Doom implementation, which means the complexity of gameplay and learning duration remain unclear. The company previously demonstrated that their neuron cultures could distinguish signal patterns within minutes, a speed that suggests rapid plasticity. Whether this translates to sophisticated gaming behavior or simpler pattern recognition matters for projecting where organoid intelligence fits in the broader computing landscape.

The work sits at the intersection of neuroscience, bioengineering, and computer science. As the neuron count required for complex tasks continues to drop, the economic and ethical questions around biological computing systems become more urgent. Cultured neurons don’t experience consciousness as we understand it, but the boundaries of that understanding keep shifting.