Mar 31, 2026
3 min read
The systems use about 200,000 neurons grown from human stem cells, placed in an array of thousands of electrodes.
AI startup develops living computer by training human brain cells to play 'Doom'
Their system uses about 200,000 neurons grown from human stem cells, mounted on thousands of electrodes.
Australian start-up Cortical Labs is developing a new type of computer that combines human brain cells with silicon chips, with the aim of creating so-called biological computers.
The system uses about 200,000 neurons grown from human stem cells, embedded in a series of thousands of electrodes.These electrodes allow standard computers to monitor and stimulate the electrical activity of neurons, creating what the company describes as a biological processing unit that can be integrated into common data server racks, says the Economist.
To keep the organisms active, the systems include equipment that provides oxygen and nutrients while removing waste, allowing the bacteria to remain viable for up to six months.
Interest in biological computing is driven in part by potential gains in energy efficiency.Hong Weng Chong, CEO of Cortical Labs, said: "Neurons, by contrast, sip power: a typical human brain, made up of nearly 90 billion of them, consumes something up to 20 watts."
Researchers also show the complexity of neural behavior.Unlike binary transistors, neurons produce a variety of electrical currents and signal timing, allowing complex types of calculations.Additionally, biological systems integrate the storage and transmission of data, reducing the need to transfer information between individual components—a process that conventional computers can do without.
Researchers say these systems may be better suited to processing real-world input.Brett Kagan, a neuroscientist and chief scientist at Cortical Labs, said neurons are adapted to interpret “messy analog signals that we commonly see in the real world.”He added that current artificial intelligence still lags behind in basic physical tasks.“He can't do math like a calculator can, but modern AI models can't do something as simple as making tea,” he pointed out.
This ties in with Moravec's paradox, which sees that while artificial intelligence performs well in abstract tasks, it often struggles with basic physical and sensory tasks.
Researchers have suggested that biological computing could eventually support applications such as drone exploration or other tasks that require real-time interaction with complex environments.
However, this technology faces challenges in converting signals between biological and electronic systems, as well as competition from large-scale investments in traditional semiconductor technology.
To expand development, Cortical Labs has made its platform available to researchers and the public.Demonstrations include training neural systems to play the video game Doom.Sean Cole, the experiment's programmer, said he created the program "in about a week."According to Hong Wen Chong, the demonstration "came out of a student hackathon at Stanford University."
The sector is also attracting institutional support.The Defense Advanced Research Projects Agency has announced funding for biological computing research, and Cortical Labs has partnered with organizations such as DayOne to deploy biological computing at the National University of Singapore.
