Why a startup is teaching human brain cells to play “Doom”
They could herald a new type of computing

Mar 30th 2026

In February Cortical Labs, an Australian startup, announced that a programmer had taught one of its “biological computers”—made of 200,000 human brain cells mounted on a silicon chip—to play “Doom”, a classic first-person shooter game. The firm had previously taught a collection of brain cells to play the much simpler “Pong”. Its ambitions are much bigger than video games, however. It hopes that neurons, packaged into super-efficient “biological computers” and slotted into racks at conventional data centres, might one day take their place alongside the transistor-packed chips of silicon that have defined conventional computing for the past half-century.

At the heart of Cortical’s system is an array of thousands of tiny electrodes, upon which sit neurons grown from stem cells taken from a human donor. The array allows a conventional computer to both pick up the electrical activity generated by those neurons and to stimulate them with electrical activity of its own. The neurons are kept alive for up to six months by tubes and pumps that supply oxygen and nutrients, and remove cellular waste products like carbon dioxide. The whole thing is packaged into a box designed to fit in the standard server racks used in commercial data centres.

Neurons offer several possible advantages over electronics when it comes to computing, says Hon Weng Chong, Cortical’s boss. Efficiency is one. Modern artificial-intelligence models gulp power by the millions of watts. That demand for energy has become one of the biggest barriers to the industry’s growth. Neurons, by contrast, sip power: a typical human brain, made up of almost 90bn of them, consumes something in the region of 20 watts.

Sophistication is another. The transistors from which electronic computers are built are tiny switches that can be in one of two states: on or off. Neurons are more complicated. Their behaviour depends on all sorts of variables, including the voltage across a cell’s membrane and how long it has been since they last received signals from other neurons. Existing computer architectures also store information far from where the actual processing happens. Micron, a big maker of memory chips, estimates that up to half the energy budget of a conventional AI processor is spent shifting data around. It also causes traffic jams as data is shuttled back and forth. Brains mix data and processing side by side, minimising such logistical issues.

Brett Kagan, a neuroscientist and Cortical’s chief scientific officer, speculates that all this may make neurons better suited than electronics to some types of computational work, especially those involving the interpretation of the sort of messy, analogue signals common in the real world. He cites the example of “Moravec’s paradox”, a long-standing and counter-intuitive observation in AI research that suggests that abstract reasoning—playing high-level chess or multiplying huge numbers—is computationally easier, in some deep and fundamental way, than the trivial-seeming motor skills needed to navigate the physical world. Dr Kagan gives the example that, while he cannot do maths like a calculator, modern AI models cannot do something simple like making a cup of tea.

That, at least, is the pitch. Realising it will be tricky. Cortical is still experimenting with how best to translate signals between electronic computers and living cells. And it is swimming against a powerful tide. Big tech firms and AI labs are betting hundreds of billions of dollars that the future of computing involves doubling down on standard electronics.

In an attempt to build momentum and to see what might be possible, the firm has decided to open its technology to anyone who fancies playing with it. The “Doom” demonstration came out of a hackathon for students at Stanford University, says Dr Chong. Sean Cole, the coder whose efforts were featured in the video documenting the event, whipped up his program “in about a week”.

Cortical has also connected some of its computers to the internet, allowing anyone to experiment; around 5,500 people have already done so, says Dr Chong. On March 10th it announced a deal with DayOne, a Singapore data-centre developer, that will see the firm install 20 of its bio-computers at the National University of Singapore.

And it will have allies in high places, too. On March 3rd DARPA, an agency of the American government that funds speculative technologies, announced a research funding programme into biological computing, with the hope of producing “biological processing units” that might use a fraction of the energy of conventional silicon chips, and which might one day prove useful for tasks like autonomously flying drones—and tackle much more of the messy real world besides. ■

https://www.economist.com/science-and-technology/2026/03/30/why-a-startup-is-teaching-human-brain-cells-to-play-doom