Computer run on human brain cells learned to play ‘Doom’
A biocomputer powered by lab-grown human brain cells has gone from Pong has Loss. While nowhere near ready to handle the video shooter’s toughest levels, researchers at Cortical Labs in Australia believe their neural chip is well on its way to powering a new generation of hybrid organic technologies.
“This is a major milestone because it demonstrated adaptive and goal-directed learning in real time,” Brett Kagan, chief scientific officer and chief operating officer of Cortical Labs, said in a recent video announcement.
It took years to cross the Loss reference. In 2021, Cortical Labs launched DishBrain, one of the first biocomputers using approximately 800,000 human nerve cells. These neurons were connected to a small processing chip that could interpret and direct electrical activity in the same way as a standard silicon-powered device.
Living human brain cells play DOOM on a CL1
To showcase DishBrain’s potential, engineers successfully trained their biocomputer to play Pong. Classic 2D gaming is often a test case for computational neuroscientists because it requires their system to navigate a dynamic information landscape in real time.
It took Cortical Labs over 18 months using its original hardware and software to achieve their Pong aim. DishBrain was eventually supplanted by CL1, which the company bills as the “world’s first code-deployable biological computer.”
But for a biocomputer to be truly useful, it will need to do much more than move a pixelated palette up and down on a screen. Enter Loss. For decades, big tech companies and DIYers alike have demonstrated how to run the video game on all types of devices, including calculators, tractors, and even ATMs. “Can it play Loss” is such a ubiquitous ask in the tech world that it wasn’t a question of “if” Cortical Labs would try it on neural chips, but “when.”
The major challenge for CL1 is to understand Loss is that we had to “see” what a human player sees when playing the game on a computer. Without any optical input, this meant engineers had to find a way to convert visual information into patterns of electrical stimulation recognizable by neurons.
The solution was not only feasible, but it was completed in about a week by Sean Cole, an independent developer with little experience in biological computing. The key lies in the CL1’s new interface, which allows anyone to program it using Python.
Don’t expect the biocomputer to win Loss tournaments though. It plays the game better than a system that just randomly shoots enemies, but it still loses a plot of the time. That said, Cortical Labs says it has reached its current level of performance faster than silicon-based machine learning systems and will likely improve as its algorithms improve.
Beyond destroying pixelated enemies, future generations of biocomputers could one day power robotic arms or help run complex digital programs. There’s a long way to go, but getting past rites of passage like playing Loss bodes well for technology.
The best new PopSci 2025 releases
