Scientists Taught Brain Organoids to Play Doom

Two hundred thousand human brain cells, grown on a microchip, brain organoids, have been trained to play the original Doom.

These brain organoids are moving, aiming, and shooting in a feedback loop driven entirely by biological neural activity. The research comes from Cortical Labs, a Melbourne-based team working at the intersection of synthetic biology and machine learning, and news of the experiments began circulating publicly in early 2026.

This is a significant step forward from where biocomputing was even two years ago, and if you’ve been following the broader push to replace or augment silicon with living neural tissue, this is a proof-of-concept that’s hard to ignore.

This Isn’t the First Time Neurons Played Video Games, But the Jump Matters

Cortical Labs isn’t new to this. In 2022, their “DishBrain” system made headlines for teaching a small cluster of neurons to play Pong. That experiment demonstrated that living cells could respond to basic feedback stimuli, but Pong is only a single back-and-forth motion and a bouncing ball. Doom is a three-dimensional environment with spatial navigation, threat detection, and prioritized decision-making across multiple simultaneous inputs.

The jump from Pong to Doom is a huge press forward. It demonstrates the complexity of what brain organoids are being asked to process.

It’s also worth noting what this is not: the paper behind this research has not yet completed full peer review as of this writing. Cortical Labs has released research updates, but independent scientific replication, the actual bar for established findings, hasn’t happened yet. That context matters before anyone declares biology has cracked computing.

How You Actually Teach Brain Cells to Shoot Demons

The setup is more elegant than it may initially seem. Researchers cultivate human neurons from stem cells and grow them on a microelectrode array, which is a chip embedded with tiny electrodes that can both stimulate the cells and record their electrical responses. This system operates as a closed feedback loop: the game’s state is translated into electrical signals sent to the neurons, which respond by firing in specific patterns. These patterns are then decoded into in-game actions.

For example, one firing pattern causes the character to move forward, while another triggers a shot. Over time, the synaptic connections that lead to useful actions strengthen, while those that do not are weakened. No one programmed these responses; they emerged organically.

This process does not represent artificial intelligence in any traditional sense. There are no transformer architectures, no backpropagation, and no gradient descent involved. Instead, what occurs is biological plasticity, which is the same mechanism that allows the human brain to learn and form memories, happening in a controlled environment linked to a first-person shooter game from 1993.

The choice of “Doom” as the game is intentional. It is computationally lightweight and structurally simple, consisting of basic actions like moving, aiming, and shooting, making it a suitable environment for testing. At the same time, it is complex enough to provide meaningful insights when a brain organoid successfully navigates its challenges.

Where This Fits in the Biocomputing Developments

Cortical Labs is not operating in a vacuum. The push to move beyond silicon has multiple fronts running simultaneously. Intel’s Loihi neuromorphic chip and IBM’s NorthPole processor are both attempts to build silicon that mimics the brain’s architecture, sparse, event-driven, low-power, without using any actual biology. DARPA has funded neuromorphic research for years under its broader electronics resiliency programs. These are serious efforts with serious funding behind them.

Wetware, the term for using actual biological tissue as a computing substrate, is a different bet entirely. It doesn’t try to replicate how neurons work. It just uses them.

Last year, I covered Final Spark, a Swiss startup that has built a remote biocomputing platform connecting 16 brain organoids through multi-electrode arrays: accessible to researchers via API, 24/7, from anywhere in the world. Their core argument: the human brain runs on roughly 20 watts while performing feats of parallel processing that no silicon chip can match at comparable energy costs. As co-founder, Dr. Fred Jordan put it, simulating human-level neural networks digitally would require the output of a small nuclear plant. The biology, by comparison, is essentially free.

Cortical Labs and Final Spark are approaching the same underlying idea from different directions. Final Spark is building infrastructure. Cortical Labs keeps stress-testing what the infrastructure can actually do.

The Ethical Question Is Real, and It’s Getting Louder

Every time one of these stories breaks, the same question surfaces: Are these systems conscious?

The current scientific consensus says no. Organoids of this size and structure don’t possess the large-scale cortical organization, sensory systems, or integrative complexity that neuroscientists associate with consciousness or subjective experience. They are not suffering. They do not have anything resembling awareness.

But the question doesn’t stay answered just because today’s answer is reassuring. The brain organoid demonstrations keep escalating. First, Pong, then progressively more complex tasks, now playing Doom, and each step makes the ethical conversation more urgent to have before the technology outpaces it. Bioethicists working in this space have already begun calling for formal frameworks around organoid research, and the European Commission’s Human Brain Project has flagged consciousness thresholds as an open policy question.

Cortical Labs has been public about engaging with these concerns directly rather than deflecting them. That’s the right posture. It doesn’t resolve anything, but it’s the difference between a field that governs itself and one that gets governed.

Biology Just Passed the Doom Test. What Comes Next Is the Real Question.

Can it run Doom? A half-joking benchmark for whether any new computing system was real.

Biology just answered it.

Two trajectories are now converging in real time: AI racing forward through software and silicon, and synthetic biology growing living neural networks on chips. Those arcs are visibly overlapping.

Silicon replaced vacuum tubes. Nobody called that transition obviously inevitable until it already was.

The limitations of wetware are real. Brain organoids are fragile, variable, difficult to scale, and short-lived. Whether this technology follows the same arc as silicon is genuinely unknown. What’s no longer theoretical is that 200,000 human brain cells just navigated a hellscape and learned to shoot back.

The line between artificial and organic intelligence is starting to blur, and this one you can watch happen in real time, in a lab in Melbourne.

Earlier coverage: Final Spark’s Biocomputers, How a Swiss Startup Is Building the World’s First Living Computers