In a room at the Life Sciences Institute in the National University of Singapore, a technician feeds sugar and salt to 20 boxes of human neurons every three days to keep them alive. This is the second “bio data centre” developed by data centre operator DayOne and biological computing start-up Cortical Labs. The first was built in Melbourne last year. If all goes to plan, a third centre with as many as 1,000 boxes will soon be built.
From Pong to Doom
In 2022, Cortical Labs trained a dish of lab-grown human brain cells to play Pong, the 1970s arcade classic. This February, they connected a petri dish containing 200,000 cells to Doom, the 1990s first-person shooter. Software translated the game into tiny electrical pulses delivered to the cells via a chip. The cells’ own electrical responses were then translated back into actions in the game — move left, right, shoot. To train them, cells were sent tiny electric zaps: a predictable pattern when they did something desired and a chaotic pattern when they didn’t. Like other living things, even in their primitive state, cells prefer order over chaos.
The technology behind CL1 boxes
Cortical Labs was founded by medical-doctor-turned-entrepreneur Dr Hon Weng Chong, after Demis Hassabis, chief executive of DeepMind, argued that “AI should go back to neuroscience”. Chong volunteered his own cells, which were gradually reprogrammed into neurons and multiplied in a petri dish. Once they are neurons, they stop multiplying and link up into a network.
In March last year, the company began creating and commercialising modular brain cell boxes called CL1s, with shipping beginning that summer. The size of a shoe-box and set at a temperature of 37C, CL1s contain a chip no bigger than a stamp, with a grid of electrodes that harness about 800,000 neurons sitting on top as a fleshy film. Most of the box is taken up with its life-support loop, because brains need brain food. It’s called “media” and is the nutrient-rich salt and sugar liquid.
The neurons, which live for between six months and a year, cost $35,000 each, or $20,000 each for a rack of 30. Alternatively, you can rent one remotely via the cloud for $300 a week. This interface between digital hardware and lab-grown neurons is called biOS (Biological Intelligence Operating System). Unlike iCloud, lab-grown neurons will not store your photos. Instead, you can train them to do things in the same way they were trained to play video games. To buy a box, you need a proper laboratory. Cortical’s buyers so far are mainly labs and researchers.
Applications and impact
One buyer is Nii Osae, who runs Mindbeam, a company working on speeding up AI models. “We need a different computational power, one that is very energy efficient, they [neurons] will take things to the next level,” he says.
The boxes are also proving popular for researching disease. “Normally if you wanted to study a disease, you end up having to give it to an animal,” says Dr Brett Kagan, chief executive of Cortical Labs. “So they spend their time suffering until you kill them. CL1 neurons don’t suffer or have pain like animals, and they show how diseases affect living human cells in a way that animal cells often fail to.”
Last year, Cortical Labs stacked 120 CL1s into its first “bio data centre” in Melbourne. Customers who purchase “biOS cloud space” can log in and access their box remotely. This summer, the company’s Singapore centre in the Life Sciences Institute marked the next phase in its scaling. If all goes to plan, Cortical’s 1,000 units would equate to 800 million neurons. A human brain contains about 86 billion neurons, so those 1,000 boxes would achieve only one per cent of that. They would be 1,000 separate units, each dying in months. As Kagan puts it: “What you have is not a brain in a jar, but tightly controlled neural circuitry.”
The sentience question
It raises an important question: what, if anything, do these cells experience? A few years ago, Cortical Labs published a scientific paper stating that its neurons had displayed sentience because they responded to zaps adaptively. Critics, including researcher Fuat Balci and 30 other co-authors, pushed back, pointing out that adaptive firing is not proof of the cells having a lived experience. Kagan has since clarified that the marker they measure is not consciousness. The Nuffield Council on Bioethics, the UK independent advisory body, takes a similar line: today’s lab-grown neurons are not considered sentient. But as these systems become more complex, the question becomes harder to dismiss.
For now, Kagan says, “there’s no evidence that these things can suffer”. Yet he admits that, as they experiment, they are steering into unknown territory. The deeper problem is that science still has only a limited understanding of consciousness and sentience — making it difficult to draw clear ethical or regulatory boundaries. The Human Tissue Act, for instance, wasn’t designed to cover cells which are grown for months outside a human body. Nor is there a dedicated regulator. This leaves companies like Cortical Labs to bear the responsibility. According to Kagan, it’s something they take seriously. “We don’t have certainty, but we can have humility and say here’s what we know and what we don’t… and here’s how we can go about finding out before we end up in a bad position.”
