On Friday, Elon Musk gave an update on what’s probably his third-most prominent company: Neuralink. Neuralink had been pretty low profile (especially in comparison to Tesla and SpaceX) prior to this time last year, which is when Musk first went into detail about the company’s goals and progress. And the goals were striking: a mass-market brain implant that could be installed by a robot via same-day surgery.
With this year’s update, little has changed about the overall plan, but plenty of little details have been tweaked in the intervening 12 months. And progress has been made, in that Musk introduced his audience to a group of pigs who were already carrying what he suggested was version 0.9 of his implants, with human testing set to follow shortly.
Designs on the brain
One of the big differences between this year and last is the overall design of the implant and its supporting hardware. The original goal had been to keep the surgery simple in part by minimizing the size of the hole that needed to be made in the skull. This meant a small-diameter implant that wouldn’t necessarily be placed near the neurons it interacted with and would require a connection to separate hardware placed behind the ear. All of this added to the level of complication and would necessarily require running some wires across the surface of the brain.
Most of that’s been simplified away. For now, rather than attempting to target anything deep in the brain, implants will only target things near the surface of the cerebral cortex, the outermost layer at the front of the brain. A neurosurgeon employed by Neuralink was on hand (in scrubs, as he was last year, just so you knew he was a surgeon) and said that there are lots of motor and sensory processes that route through the cortex and thus can be targeted by these implants. Musk followed up on that by saying that we can “solve” blindness and deafness solely by focusing on the cortex.
As in the old design, roughly 1,000 electrodes will be inserted into the targeted collection of neurons, and those will connect to an implant above the surface of the brain. But in this case, they’ll do so by the shortest route possible, eliminating the need to run the wires across the surface of the brain. The behind-ear hardware is gone as well. Instead, there will be a single implant that spans the skull, essentially replacing part of it.
Musk showed off one of the implants, and it represents a major rethinking. He said it was 23mm across (though he didn’t specific radius or diameter) and about 8mm thick—the latter chosen as a close match to the thickness of human skulls. The device looks like a very thick coin or miniature hockey puck, and it contains all the hardware needed to keep the implant functional. This includes a battery large enough for all-day operation and the hardware needed for wireless inductive charging. There are also support chips, mostly derived from wearables, that control the charging and enable wireless communication via Bluetooth.
But the central feature is still Neuralink’s custom chip designed to identify and transmit patterns of neural activity. Individual neurons, which the electrodes will be listening in on, communicate by firing off a series of what are called “spikes”—short bursts of electrical activity that stand out from the background noise. Musk said that Neuralink’s chip comes programmed with a set of spike templates that match the usual range of behaviors seen in actual neurons. The chip will take the analog electrical activity recorded by the electrodes, convert it to digital data, identify any spikes of activity, and then find the template spike that matches the activity best.
That allows it to transmit a code that identifies the template, making for a huge compression compared to the complicated, noisy neural activity. It’s absolutely necessary for a device that will be communicating via a low-bandwidth interface like Bluetooth.
The chip will also allow the electrodes to be used to stimulate neurons, although Musk didn’t go into the details on this. It presumably takes a bit more power than passively reading their activity, which may limit how much this can be used.
Putting it in place
The team is still counting on a robot to do the hardest parts of the implant, using a microscope to identify and avoid blood vessels when threading electrodes into the brain. The prototype robot that Neuralink showed off last year, which could easily have been mistaken for an interrogation droid, is no more. In its place is a clean, white, very medical-looking design—one with all the tools and pointy bits necessary for putting a hole in someone’s skull kept from view while not in use. While the team is still working on adapting the robot to reach deeper areas in the brain, the initial focus on the cortex makes the robot’s task somewhat easier for now.
Musk promised that the surgery would only take about an hour, wouldn’t require general anesthesia, and the recipient can go home with their implant the same day.
Musk then introduced the first recipients of the current implant design: a group of pigs. Naturally, there was a control animal, who happily waddled out before the camera and ate from its handler’s hand. That was followed by a second pig that had previously had an implant but had since had the device removed. This, Musk emphasized, was important because people might want to upgrade their implants, especially as the company plans on continuing to improve its hardware.
That should have been followed by a pig with a functional implant, but Gertrude was being shy, which left an animal technician exasperated and Musk joking about the perils of live demos. Eventually, she did come out, and Neuralink was able to show that it could listen in on neurons that received signals from sensory cells in the animal’s snout. Yet another pig was brought out that carries two implants; one device was listening in on the animal’s proprioception system, and the team was able to reasonably accurately figure out the position of the animal’s leg.
Next stop: humans?
As he did last year, Musk said that the goal of the presentation was to recruit good people to work at Neuralink, which is currently at roughly 100 employees. But he definitely dropped some hints that things are getting closer to when he would like to start on the marketing. To begin with, he mentioned that Neuralink had gotten a Breakthrough Device designation from the Food and Drug Agency, which handles medical-implant approval. That enables the company to engage in an ongoing dialog with the FDA, which will help it identify the sorts of data it will need to gather in order if Neuralink wants to make sure approval is ultimately granted.
Musk also said that it plans to do initial testing with tetraplegics. That’s following in the footsteps of some earlier work, including people who have successfully used a brain implant to control a robot arm. Getting approval for human experiments with tetraplegics is relatively easy, as there are no other treatments available to them. Musk didn’t indicate when those experiments might start.
So for now, all we have are a few bits of data recorded from freely roaming pigs and not many technical details about how those were obtained. To an extent, that doesn’t represent a lot of obvious progress from last year’s presentation or a lot of progress compared to where the research community already is.
But that really doesn’t capture the situation well. While the field has been able to put a large number of implants into animal’s brains over the years, I don’t recall any that left an animal that was visibly indistinguishable from its non-implanted peers, as these pigs were. And Neuralink’s hardware is designed to be manufactured at scale, unlike the typical production of an academic research lab.
Just as critically, the company has shown plenty of creativity and flexibility. It’s got a solid solution to the problem of compressing the complicated data that its hardware will gather, and the team made some major revisions to its designs compared to a year ago, all of which seem to simplify the implant and the surgery needed to put it in place. And that, within the span of a year, represents significant progress.
That isn’t to say that Neuralink is set for an easy glide to success. There are still an abundance of hard problems and years of safety and efficacy testing in its future. But the update does provide some very real reasons for optimism.