Neuralink’s Wildly Anticipated New Brain Implant: the Hype vs. the Science

Neuralink

Neuralink’s wildly anticipated demo last Friday left me with more questions than answers. With a presentation teeming with promises and vision but scant on data, the event nevertheless lived up to its main goal as a memorable recruitment session to further the growth of the mysterious brain implant company.

Launched four years ago with the backing of Elon Musk, Neuralink has been working on futuristic neural interfaces that seamlessly listen in on the brain’s electrical signals, and at the same time, “write” into the brain with electrical pulses. Yet even by Silicon Valley standards, the company has kept a tight seal on its progress, conducting all manufacturing, research, and animal trials in-house.

A vision of marrying biological brains to artificial ones is hardly unique to Neuralink. The past decade has seen an explosion in brain-machine interfaces—some implanted into the brain, some into peripheral nerves, or some that sit outside the skull like a helmet. The main idea behind all these contraptions is simple: the brain mostly operates on electrical signals. If we can tap into these enigmatic “neural codes”—the brain’s internal language—we could potentially become the architects of our own minds.

Let people with paralysis walk again? Check and done. Control robotic limbs with their minds? Yup. Rewriting neural signals to battle depression? In humans right now. “Recording” the electrical activity behind simple memories and playing it back? Human trials ongoing. Linking up human minds into a BrainNet to collaborate on a Tetris-like game through the internet? Possible.

Given this backdrop, perhaps the most impressive part of the demonstration isn’t lofty predictions of what brain-machine interfaces could potentially do one day. In some sense, we’re already there. Rather, what stood out was the redesigned Link device itself.

A FitBit for the Brain

In Neuralink’s “coming out” party last year, the company envisioned a wireless neural implant with a sleek ivory processing unit worn at the back of the ear. The electrodes of the implant itself are “sewn” into the brain with automated robotic surgery, relying on brain imaging techniques to avoid blood vessels and reduce brain bleeding.

The problem with that design, Musk said, is that “it had multiple pieces and was complex. You still wouldn’t look totally normal because there’s a thing coming out of your ear.”

The prototype at last week’s event came in a vastly different physical shell. About the size of a large coin, the device replaces a small chunk of your skull and sits flush with the surrounding skull matter. The electrodes, implanted inside the brain, connect with this topical device. When covered by hair, the implant is invisible.

Musk envisions an outpatient therapy where a robot can simultaneously remove a piece of the skull, sew the electrodes in, and replace the missing skull piece with the device. According to the team, the Link has similar physical properties and thickness as the skull, making the replacement a sort of copy-and-paste. Once inserted, the Link is then sealed to the skull with “superglue.”

“I could have a Neuralink right now and you wouldn’t know it,” quipped Musk.

For a device that small, the team packed an admirable array of features into it. The “Link” device has over 1,000 channels, which can be individually activated. This is on par with Neuropixel, the crème de la crème of neural probes with 960 recording channels that’s currently used widely in research, including by the Allen Institute for Brain Science.

Compared to the Utah Array, a legendary implant system used for brain stimulation in humans with only 256 electrodes, the Link has an obvious edge up in terms of pure electrode density.

What’s perhaps most impressive, however, is its onboard processing for neural spikes—the electrical pattern generated by neurons when they fire. Electrical signals are fairly chaotic in the brain, and filtering spikes from noise, as well as separating trains of electrical activity into spikes, normally requires quite a bit of processing power. This is why in the lab, neural spikes are usually recorded offline and processed using computers, rather than with on-board electronics.

The problem gets even more complicated when considering wireless data transfer from the implanted device to an external smartphone. Without accurate and efficient compression of those neural data, the transfer could tremendously lag, drain battery life, or heat up the device itself—something you don’t want happening to a device stuck inside your skull.

To get around these problems, the team has been working on algorithms that use “characteristic shapes” of electrical patterns that look like spikes to efficiently identify individual neural firings. The data is processed on the chip inside the skull device. Recordings from each channel are filtered to root out obvious noise, and the spikes are then detected in real time. Because different types of neurons have their characteristic ways of spiking—that is, the “shape” of their spikes are diverse—the chip can also be configured to detect the particular spikes you’re looking for. This means that in theory the chip could be programmed to only capture the type of neuron activity you’re interested in—for example, to look at inhibitory neurons in the cortex and how they control neural information processing.

These processed spike data are then sent out to smartphones or other external devices through Bluetooth to enable wireless monitoring. Being able to do this efficiently has been a stumbling block in wireless brain implants—raw neural recordings are too massive for efficient transfer, and automated spike detection and compression of that data is difficult, but a necessary step to allow neural interfaces to finally “cut the wire.”

Link has other impressive features. For one, the battery life lasts all day, and the device can be charged at night using inductive charging. From my subsequent conversations with the team, it seems like there will be alignment lights to help track when the charger is aligned with the device. What’s more, the Link itself also has an internal temperature sensor to monitor for over-heating, and will automatically disconnect if the temperature rises above a certain threshold—a very necessary safety measure so it doesn’t overheat the surrounding skull tissue.

An Inherent Tension

From the get-go of the demonstration, there was an undercurrent of tension between what’s possible in neuroengineering versus what’s needed to understand the brain.

Since its founding, Neuralink has always been fascinated with electrode numbers: boosting channel numbers on its devices and increasing the number of neurons that can be recorded at the same time.

At the event, Musk said that his goal is to increase the number of recorded neurons by a factor of “100, then 1,000, then 10,000.”

But here’s the thing: as neuroscience is increasingly understanding the neural code behind our thought processes, it’s clear that more electrodes or more stimulated neurons isn’t always better. Most neural circuits employ what’s called “sparse coding,” in that only a handful of neurons, when stimulated in a way that mimics natural firing, can artificially trigger visual or olfactory sensations. With optogenetics—the technique of stimulating neurons with light—scientists now know that it’s possible to incept memories by targeting just a few key neurons in a circuit. Sticking a ton of wires into the brain, which inevitably causes scarring, and zapping hundreds of thousands of neurons isn’t necessarily going to help.

Unlike engineering, the solution to the brain isn’t more channels or more implants. Rather, it’s deciphering the neural code—knowing what to stimulate, in what order, to produce what behavior. It’s perhaps telling that despite claims of neural stimulation, the only data shown at the event were neurons firing from a section of a mouse brain—using two-photon microscopy to image neural activation—after zapping brain tissue with an electrode. What information, if any, is really being “written” into the brain? Without an idea of how neural circuits work and in what sequences, zapping the brain with electricity—no matter how cool the device itself is—is akin to banging on all the keys of a piano at once, rather than composing a beautiful melody.

Of course, the problem is far larger than Neuralink itself. It’s perhaps the next frontier in solving the brain’s mysteries. To their credit, the Neuralink team has looked at potential damage to the brain from electrode insertion. A main problem with current electrodes is that the brain will eventually activate non-neuronal cells to form an insulating sheath around the electrode, sealing it off from the neurons it needs to record from. According to some employees I talked to, so far, for at least two months, the scarring around electrodes is minimal, although in the long run there may be scar tissue buildup at the scalp. This may make electrode threads difficult to remove—something that still needs to be optimized.

However, two months is only a fraction of what Musk is proposing: a decade-long implant, with hardware that can be updated.

The team may also have an answer there. Rather than removing the entire implant, it could potentially be useful to leave the threads inside the brain and only remove the top cap—the Link device that contains the processing chip. The team is now trying the idea out, while exploring the possibility of a full-on removal and re-implant.

A Futuristic Vision

As a demonstration of feasibility, the team trotted out three adorable pigs: one without an implant, one with a Link, and one with the Link implanted and then removed. Gertrude, the pig currently with an implant in areas related to her snout, had her inner neural firings broadcasted as a series of electrical crackles as she roamed around her pen, sticking her snout into a variety of food and hay and bumping at her handler.

Pigs came as a surprise. Most reporters, myself included, were expecting non-human primates. However, pigs seem like a good choice. For one, their skulls have a similar density and thickness to human ones. For another, they’re smart cookies, meaning they can be trained to walk on a treadmill while the implant records from their motor cortex to predict the movement of each joint. It’s feasible that the pigs could be trained on more complicated tests and behaviors to show that the implant is affecting their movements, preferences, or judgment.

For now, the team doesn’t yet have publicly available data showing that targeted stimulation of the pigs’ cortex—say, motor cortex—can drive their muscles into action. (Part of this, I heard, is because of the higher stimulation intensity required, which is still being fine-tuned.)

Although pitched as a prototype, it’s clear that the Link remains experimental. The team is working closely with the FDA and was granted a breakthrough device designation in July, which could pave the way for a human trial for treating people with paraplegia and tetraplegia. Whether the trials will come by end of 2020, as Musk promised last year, however, remains to be seen.

Rather than other brain-machine interface companies, which generally focus on brain disorders, it’s clear that Musk envisions Link as something that can augment perfectly healthy humans. Given the need for surgical removal of part of your skull, it’s hard to say if it’s a convincing sell for the average person, even with Musk’s star power and his vision of augmenting natural sight, memory playback, or a “third artificial layer” of the brain that joins us with AI. And because the team only showed a highly condensed view of the pig’s neural firings—rather than actual spike traces—it’s difficult to accurately gauge how sensitive the electrodes actually are.

Finally, for now the electrodes can only record from the cortex—the outermost layer of the brain. This leaves deeper brain circuits and their functions, including memory, addiction, emotion, and many types of mental illnesses off the table. While the team is confident that the electrodes can be extended in length to reach those deeper brain regions, it’s work for the future.

Neuralink has a long way to go. All that said, having someone with Musk’s impact championing a rapidly-evolving neurotechnology that could help people is priceless. One of the lasting conversations I had after the broadcast was someone asking me what it’s like to drill through skulls and see a living brain during surgery. I shrugged and said it’s just bone and tissue. He replied wistfully “it would still be so cool to be able to see it though.”

It’s easy to forget the wonder that neuroscience brings to people when you’ve been in it for years or decades. It’s easy to roll my eyes at Neuralink’s data and think “well neuroscientists have been listening in on live neurons firing inside animals and even humans for over a decade.” As much as I’m still skeptical about how Link compares to state-of-the-art neural probes developed in academia, I’m impressed by how much a relatively small leadership team has accomplished in just the past year. Neuralink is only getting started, and aiming high. To quote Musk: “There’s a tremendous amount of work to be done to go from here to a device that is widely available and affordable and reliable.”

Image Credit: Neuralink