‘Organoid intelligence’ could make human computers that are way smarter than AI
In a laboratory at Johns Hopkins University, there’s a small orb the size of a pinprick. But this speck of matter contains multitudes.
Within it, there are neurons firing across synapses and electric impulses zipping to and fro. Its microscopic universe holds the key to a world that even the brightest stars in the technology industry have yet to unlock. It can carry out tasks that stymie even the smartest artificial intelligence models, such as learning new information or remembering old information, and then commandeering that data in an independent, logical, decision-making process—something that has discombobulated even whiz-machines like OpenAI’s ChatGPT.
There is one entity on Earth that does this best (as far as we know): the human brain. Why has it not yet been recruited to power future computers?
That’s the question being asked by a team at Hopkins’s Bloomberg School of Public Health and Whiting School of Engineering. They’re working with what they call “brain organoids”—tiny, lab-grown human brains, sprouting from clumps of human skin cells that were reprogrammed into something akin to embryonic stem cells. And in a paper published in the journal Frontiers in Science on Tuesday, they say these specks of brain could one day, within our lifetime, be channeled into a powerful “biocomputer.”
It would be quite a feat for the artificial to mimic the natural. “The brain is still unmatched by modern computers,” Thomas Hartung, a professor of environmental health sciences at Hopkins and the lead researcher, said in a statement. “Frontier, the latest supercomputer in [Tennessee], is a $600 million, 6,800-square-feet installation. But only in June of last year, it exceeded for the first time the computational capacity of a single human brain—while using a million times more energy.”
He thinks biocomputing could be a Brave New World—or at least, a new frontier. “[It’s] an enormous effort of compacting computational power and increasing its efficiency to push past our current technological limits,” said Hartung.
Head scratcher
Switching to brainpower could solve the problem of the massive energy load demanded by today’s silicon-semiconductor-chip-powered supercomputers. And beyond that, while supercomputers still excel at, well, supercomputing—crunching billions of numbers at what feels like the speed of light—human brains are better trained at making more complex calls, such as differentiating a cat from a dog. They also have a built-in information storage capacity that’s estimated to be 2,500 terabytes, supported by a network of some 100 billion neurons (think of all the amazing things your mind just knows).
This work is being dubbed “organoid intelligence”—a sort of foil to the artificial intelligence mania that has gripped the tech world, with companies from Microsoft to Spotify to BuzzFeed rushing to fast-track AI projects ever since ChatGPT amazed the public in November. Meanwhile, tiny organs have been used in biology labs for nearly two decades, to conduct kidney and lung research free of real human or animal testing. But the gray-area matter of probing a human brain with its own “intelligence” (even if lab grown) required oversight from a diverse consortium of scientists and bioethicists.
Still, it was fertile territory. “You can start manipulating the system, doing things you cannot ethically do with human brains,” Hartung said. Separately, he cited the possibility of ethical drug testing for neurodegeneration, such as Alzheimer’s disease.
But they’ve got a long way to go. Each of the Hopkins team’s brain organoids is made of roughly 50,000 cells, with a nervous system the size of a fruit fly’s. It might take decades to scale this up into a brain system that’s as smart as even a house mouse (that is, smart but not smart enough—until you need to trap one, and then suddenly it’s a genius). For starters, to power a computer or host memory similar to a human brain, the team would need a mass of at least 10 million cells. Then, how exactly the scaling up happens is unclear—but it could involve a data-based training regimen similar to those used by AI algorithms. And if it were to happen, more implications abound: Would a computer made of human cells be more likely to develop a consciousness? Could we morally keep it in our servitude?
Still, there are promising results in the field already. In October, scientists from Australia’s Cortical Labs said they had grown 800,000 brain cells in a dish that successfully learned to play the 1970s arcade game, Pong.
Are organoid intelligence-powered biocomputers a distant future or a sci-fi fantasy? Nobody knows yet. But if we just look within ourselves, we might find our answers.
(24)
