Elon Musk says he wants to merge your brain with a computer. But by his own admission, the idea of allowing people to control their smartphone or connect directly to the internet with their thoughts is many years away. “It will take a long time,” Musk said at an event for his brain-computer interface startup Neuralink Tuesday night in San Francisco.
Nonetheless, Musk claimed that Neuralink could be ready to test its technology in human beings who are paralyzed as early as next year. In a much-hyped presentation that was half neuroscience conference, half Silicon Valley spectacle, the San Francisco-based company, which Musk has invested $100 million in, revealed that it has developed tiny, thread-like implants far thinner than a human hair that are capable of recording neuron activity.
A brain-computer interface is meant to provide a direct pathway of communication between the brain and an external device. These interfaces use electrodes to “read” brain signals, plus a computer that translates these signals into commands, which carry out a desired action on a computer program or device.
Eventually, Musk sees a future in which anyone could opt in to getting one of these interfaces and achieve what he calls a “symbiosis with artificial intelligence.” Musk envisions an elective brain surgery that would be minimally invasive and take just a few hours, similar to a modern LASIK procedure. With such an interface, he says people will “have the option of merging with A.I.” — an area of particular interest for Musk, who has warned about the existential threat that ever more powerful A.I. could eventually pose to humanity.
Neuralink says that eventually its interface will allow people with disabilities to speak, hear, and see.
But the first use of the company’s technology will be far more practical: to allow people with complete paralysis from spinal cord injuries to gain “naturalistic control” of a computer, according to Neuralink researcher Philip Sabes. In a question and answer session after the presentation, Musk said the goal is to allow people with quadriplegia — paralysis of the limbs and torso — the ability to type at 40 words per minute. Current brain-computer interfaces can allow such patients to type at about 10 words per minute. Neuralink says that eventually its interface will allow people with disabilities to speak, hear, and see.
Even for use in the most severely affected patients, marketing approval from the U.S. Food and Drug Administration is likely years away. The FDA has already approved deep brain stimulation, a type of implant that produces electrical pulses in the brain, to treat Parkinson’s and a few other conditions. But implantable interfaces for direct control of devices are still only used for research purposes and are not commercially available.
Research on brain-computer interfaces began in the 1970s, and the first crude devices were implanted in humans in the 1990s. They’ve so far been used to restore a limited range of motion in severely paralyzed people, permitting them to control computer cursors or robotic arms directly with their thoughts. These devices have mostly been tested in research labs and are years away from being practical devices that people could use in their daily lives.
Current interfaces also involve invasive brain surgery, requiring neurosurgeons to drill through the skull to implant electrodes. And many interfaces for motor movement have used metal pedestals that connect the electrodes in the brain to a computer with a cable. External devices, like robotic limbs, also need to be connected to a computer.
Musk wants to get rid of all that clunkiness. He says the interface Neuralink is developing will be completely wireless (though his system wouldn’t be the first). A prototype of the company’s chip shows a small device, about the size of a hearing aid, that will sit behind the ear. Max Hodak, Neuralink’s president, said the whole interface will eventually be controlled by a smartphone app.
It’s unclear how much of what Neuralink is promising actually exists yet. What the company did reveal on Tuesday is that it has created arrays of small, flexible “threads” with thousands of electrodes on them. Current interfaces rely on rigid metal electrodes that are implanted in the brain. These electrodes can cause scar tissue to form in the brain, affecting their ability to record neural signals over time. By contrast, Neuralink’s polymer threads are about a tenth of the width of a human hair, according to Musk. The company claims these threads are 15 times better at reading information in the brain than current electrodes.
At Tuesday’s event, Musk also described a sewing machine-like robot that can precisely insert these flexible threads into the brain. The idea was first explained in a March paper on the preprint server bioRxiv by researchers at the University of California, San Francisco and the University of California, Berkeley. Two of those researchers, Sabes and Timothy Hanson, have since joined Neuralink.
While Musk focused on future use of the interface technology in human beings — and claims that the first human subjects could be part of a trial by the end of 2020 — the white paper the company released on Wednesday describes a small study in rats. In 19 surgeries, the robot was able to successfully place the threads 87% of the time in the rats’ brains. (Musk also said, almost in passing, that “a monkey has been able to control a computer with its brain,” but stopped short of providing details.)
There are drawbacks that come with using smaller electrodes. “The body is a harsh place. These materials can break down over time,” Vikash Gilja, an assistant professor of electrical and computer engineering at the University of California, San Diego who served on the technical team at Neuralink from July 2017 to June 2018, told OneZero. “If you shrink the materials down, they may be more susceptible to breakdown because you’re putting less of that material in or because you’re using a material that hasn’t been tested yet in those conditions.”
Assuming the company can develop an interface that can remain stable in the brain for years — few people would volunteer for annual brain surgery to maintain the machine — the success of Neuralink’s implant will come down to the data rate: how many neurons the interface can simultaneously record and stimulate.
“Having more data means that you can control a larger number of degrees of freedom,” said Matt Angle, CEO and founder of Paradromics, a company based in Austin, Texas, that’s also building high data rate brain-computer interfaces. In the case of prosthetics for conscious control of movement, a higher data rate would give a person more dexterity. In terms of communication, a bigger data rate would increase your typing speed. With a visual prosthesis, a higher data rate would translate to sharper resolution.
Neuralink says it has the ability to record from 1,000 neurons at once. Melody Jackson, a brain-computer interface researcher at the Georgia Institute of Technology, told OneZero the technology is “completely feasible.” She said reading 100,000 neurons would be even better — the human brain has about 100 billion neurons — but the problem researchers have run into is how many holes to drill in the brain and where to place the electrodes.
One question she has for Musk’s team is how it plans to interpret these neural signals. Usually that relies on machine-learning techniques. “That’s the key,” Jackson said. “Getting the brain signal is not hard. When you have the brain signal, what does it mean? That’s the hard part.”
These devices won’t be connecting our brains to the internet anytime soon.
Other companies, like Kernel and CTRL-Labs, are developing wearable brain-computer interfaces that are non-invasive.
These devices won’t be connecting our brains to the internet anytime soon, but Angle can imagine a broader use in a few years for treating psychiatric diseases. “As these [implantable] devices are shown to be safe, it will become thinkable for larger and larger groups of patients to have them,” he says.
The first human tests of Neuralink’s device will still require traditional brain surgery, Musk said, but his grand vision is to be able to robotically implant these interfaces in a few hours in an outpatient facility. Angle says that scenario is likely 10 to 15 years off, which gives us a time frame to start thinking through not just the technical questions posed by Neuralink and Musk’s inventions, but the societal and ethical ones of connecting our brains — the organ that makes us human — to a computer.