Elon Musk is known for trumpeting bold and sometimes brash plans. So it was no surprise last week when the Tesla founder made an announcement—in front of a live audience and streamed online, with a video trailer and thematic music—that his new company Neuralink plans to sync our brains with artificial intelligence. (Don’t worry, he assured the audience, “this is not a mandatory thing.”)
What was surprising was the breathless coverage in most media, which lacked context or appreciation for the two decades of research on which Neuralink’s work stands.
Neuralink, based in San Francisco, has been in stealth mode since it was founded by Musk a couple of years ago. The company has reportedly raised over $150 million, including $100 million from Musk.
It was known that Musk founded Neuralink to produce an implantable brain-computer-interface (BCI), but details have been scanty. Last week’s live presentation was the first time Musk revealed some of the concrete progress the company has made.
Musk wants to develop a brain implant that enables people to control computers, prosthetics, and other medical devices using only thoughts. No typing or speaking necessary.
Creating this kind of interface involves recording electrical brain activity, decoding it, and translating it into computer commands. Some systems also input sensory stimulation into the brain, which is helpful in, for example, controlling a prosthetic. Each step of this chain of technologies can demand its own area of study, and lots of scientists have devoted their careers to one or more components.
Neuralink appears to have moved the needle forward in the field of BCI hardware. According to Musk, the company’s scientists have developed an array of 3000 electrodes on flexible threads that can be implanted into the brain of a rat. In previous designs from other researchers, the number of electrodes that could be safely implanted into one brain has numbered in the hundreds.
If Neuralink’s announcement is true (and by the way, Neuralink has not published its findings in a peer-reviewed journal), then this is a nice advance for the field of BCI. But reading the media coverage, one would think Neuralink invented the concept of decoding brain signals and was poised to revolutionize human processing capability.
“The media coverage was essentially riding on top of the company’s pitch,” says Rajesh Rao, a professor at the University of Washington and author of the book Brain-Computer Interfacing, who spoke with IEEE Spectrum in an interview. Neuralink appears to have made some interesting advancements, he says, but “there was a lack of acknowledgment of what people in the field have accomplished over the last two decades or more.”
That gaping hole came despite the fact that Neuralink presenters emphasized the history on which they stand, and the long road ahead of them. “The work we’re doing doesn’t come out of thin air,” said Philip Sabes, senior scientist at Neuralink. “We’re building on over a century of neuroscience research and decades of neural engineering research.”
Work in BCI began in earnest in the 1990s, and progressed to successful human experiments using both implanted and non-invasive recording electrodes. The BrainGate consortium, for example, uses BCI to restore communication and mobility to people with neurologic diseases and limb injuries. This team of scientists has enabled people to operate tablets, type eight words per minute, and control prostheses, in a limited way, using only their thoughts.
Rao and other researchers have taken the BCI concept a step further with experiments that connect two people’s brains via computer—a technology called brain-to-brain interface. In a recent set of experiments, Rao and his colleagues enabled groups of three people to collaboratively play a video game using only their thoughts. The electrodes recorded brain waves non-invasively from the scalp.
Scientists have also made great strides in treating Parkinson’s disease, epilepsy, and psychiatric disorders using deep brain stimulation (DBS). In that method, electrodes are surgically implanted in the brain in key areas, and deliver electrical stimulation in an effort to control symptoms and seizures.
In a white paper released after the presentation, Neuralink reports that its technology enables the implantation of over 3,000 electrodes contained in ultra-thin, flexible polymer probes, or threads.
“This is an advance over current technologies, such as the Utah Array, which is not flexible and might record from close to a hundred locations,” says Rao. “But if you think about the context of the brain, 3000 locations is just a drop in the bucket compared to the number of neurons in the brain, which run into the billions,” he says.
Musk’s company announced it had built a neurosurgical robot capable of implanting its custom threads into the brain at 192 electrodes per minute. It also developed a low-power chip to process the signals in real time.
Neuralink presenters say the technology has been demonstrated in rats. Musk also mentioned—seemingly off-the-cuff during the Q&A portion of the presentation—that the company has experimented in monkeys.
Musk estimates Neuralink’s electrodes will be recording human brain activity by the end of next year. That’s ambitious, considering how hard it is to get regulatory approval from the U.S. Food and Drug Administration for new electrode technologies in humans.
And so far, the company has publicly demonstrated only half the puzzle. Recording brain activity is great, but decoding it and turning into something meaningful is another matter. “What can we do now with 3000 neurons recorded? What kind of performance enhancement do we get?” says Rao. “That was not described” by Neuralink, he says.
A huge amount of math, coding, and human experimentation goes into developing systems that can reliably parse human thought based on the brain’s electrical activity, and turn it into something useful such as typing.
Plus, if Musk wants his BCI device to control prostheses or improve the mobility of injured limbs, his scientists will have to figure out a way to reliably provide sensory feedback to the brain in the form of electrical stimulation. Recording brain activity and stimulating at the same time has proven hard to do, since the signals can interfere with each other. Neuralink has not described how it will address this challenge.
So Musk’s latest endeavor isn’t the mind blowing advance some media outlets made it out to be. But Neuralink’s new tool, if it can been validated in the peer review process, does provide an incremental advance for the field of BCI. And that’s exactly the way science is supposed to work.