Skip Navigation

What Neuralink is missing (it turns out that connecting brains with computers is the easy part)

www.theatlantic.com What Neuralink Is Missing

It turns out that connecting brains with computers is the easy part.

What Neuralink Is Missing
42

You're viewing a single thread.

42 comments
  • Could you post the full text? The article seems to be paywalled

    • Until recently, in all of human history, the number of true cyborgs stood at about 70. Ian Burkhart has kept a count because he was one of them—a person whose brain has been connected directly to a computer.

      Burkhart had become quadriplegic in a swimming accident after a wave ran him into a sandbar and injured his spine. He was later able to receive an implant from a research study, which allowed him to temporarily regain some movement in one hand. For seven and a half years, he lived with this device—an electrode array nestled into his motor cortex that transmitted signals to a computer, which then activated electrodes wrapped around his arm. Burkhart now heads the BCI Pioneers Coalition, an organization for the small cohort of other disabled people who have volunteered their brain to push the boundaries of brain-computer-interface technology, or BCI.

      Last month, Burkhart, along with perhaps millions of other people, watched the debut of the newest cyborg. In a video posted on X, the first human subject for Elon Musk’s BCI company, Neuralink, appeared to control a laptop via brain implant. Neuralink has not published its research and did not respond to a request for comment, but the device presumably works this way: The subject, a paralyzed 29-year-old named Noland Arbaugh, generates a pattern of neural activity by thinking about something specific, like moving the cursor on his computer screen or moving his hand. The implant then transmits that pattern of neural signals to the computer, where an AI algorithm interprets it as a command that moves the cursor. Because the implant purportedly allows a user to control a computer with their thoughts, more or less, Musk named the device Telepathy.

      Read: Demon mode activated

      Burkhart watched Arbaugh play hands-free computer chess with a mix of approval and frustration at how clearly the demo was created for investors and Musk fans, not for disabled people like him. It’s no secret that Musk’s real goal is to create a BCI device for general consumers, and not just so we can move a cursor around; he envisions a future in which humans can access knowledge directly from computers to “achieve a symbiosis with artificial intelligence.” That dream is ethically fraught—privacy, for instance, is tricky when your thoughts are augmented by proprietary algorithms—but it is also a long way from being realized. Researchers have sort of managed two-way information transfer with rats, but no one is sure how the rats felt about it, or whether it’s an experience they’d be willing to pay for at a mall kiosk.

      Yet a more modest vision for a safe, workable neuro-prosthesis that would allow disabled people to use a computer with ease is realizable. The question is whether our social structures are ready to keep pace with our advanced science.

      It’s taken decades for BCI tech to get to this point—decades of scientists building prototypes by hand and of volunteers who could neither move nor speak struggling to control them. The most basic challenge in mating a brain and a computer is an incompatibility of materials. Though computers are made of silicon and copper, brains are not. They have a consistency not unlike tapioca pudding; they wobble. The brain also constantly changes as it learns, and it tends to build scar tissue around intrusions. You can’t just stick a wire into it.

      Different developers have tried different solutions to this problem. Neuralink is working on flexible filaments that thread inconspicuously—they hope—through the brain tissue. Precision Neuroscience, founded in part by former Neuralink scientists, is trying out a kind of electrode-covered Saran Wrap that clings to the surface of the brain or slips into its folds. Then there’s the Utah Array, a widely used model that looks a little like a hairbrush with its bristly pad of silicone spikes. That’s what Burkhart had in his head until 2021, when the study he was part of lost funding and he decided to have the implant taken out. He was worried surgeons might have to “remove some chunks of brain” along with it. Luckily, he told me, it came out “without too much of a fight.”

      Once an implant is in place, the tiny signals of individual neurons—measurable in microvolts—have to be amplified, digitized, and transmitted, preferably by a unit that’s both wireless and inconspicuous. That’s problem number two. Problem three is decoding those signals. We have no real idea of how the brain talks to itself, so a machine-learning algorithm has to use a brute-force approach, finding patterns in neural activity and learning to correlate them with whatever the person with the implant is trying to make the computer do.

      None of these problems is trivial, but they’ve been substantially tackled over the past 30 years of BCI research. At least six different companies are now testing applications such as desktop interfaces (like the one that helped Arbaugh play chess), drivers for robotic limbs and exoskeletons, and even speech prostheses that give voice to thought. Proof-of-concept devices exist for all of these by now.

      But that only brings us to problem number four—which has nothing to do with engineering and might be harder to solve than all the others. This problem is what Ben Rapoport, the chief science officer at Precision, described to me as “the productization of science.” It’s where engineering successes run into political and economic obstacles. To roll out even a basic point-and-click medical BCI interface, developers would have to win approval not just from the FDA but also from “payers”: Medicare, Medicaid, and private insurance companies. This is make-or-break: Medical devices, even ingenious ones, won’t get to consumers if insurance won’t cover them. Few people can afford such expenses out of pocket, which means too small a pool of potential consumers to make production profitable.

      Read: I’m disabled. Please help me.

      Other devices have cleared this hurdle—cochlear implants, deep-brain stimulation devices, pacemakers—and it’s not unlikely that BCI implants could join that list if insurers decide they’re worth the expense. On the one hand, insurance companies might argue that BCI devices aren’t strictly medically necessary—they’re “life-enhancing,” not “life-sustaining,” as Burkhart put it—but on the other hand, insurers are likely to see them as cost-efficient if their implementation can save money on other, more expensive kinds of support.

      Even so, there’s a limit to what brain implants can do and what they can replace. The people who would benefit most from BCI devices, people with major motor impairments like Arbaugh and Burkhart, would still depend on human labor for many things, such as getting in and out of bed, bathing, dressing, and eating. That labor can easily cost as much as six figures a year and isn’t typically reimbursed by private health-insurance companies. For most people, the only insurer that covers this kind of care is Medicaid, which in most states comes with stringent restrictions on recipients’ income and assets.

      In Ohio, where Burkhart lives, Medicaid recipients can’t keep more than $2,000 in assets or make more than $943 a month without losing coverage. (A waiver program raises the monthly income cap for some to $2,829.) The salary they’d have to make to cover both expenses and in-home care out of pocket, though, is much more than most jobs pay. “A lot of people don’t have the opportunity to make such a giant leap,” Burkhart said. “The system is set up to force you to live in poverty.”

      In addition to his work with the BCI Pioneers Coalition, Burkhart also leads a nonprofit foundation that fundraises to help people with disabilities cover some of the expenses insurance won’t pay for. But these expenses would be “nowhere near the size that would pay to get a BCI or anything like that,” he told me. “We do a lot of shower chairs. Or hand controls for a vehicle.”

      Starting in the late 20th century, simple switch devices began to enable people with severe motor disabilities to access computers. As a result, many people who would previously have been institutionalized—those who can’t speak, for example, or move most of their body—are able to communicate and use the internet. BCI has the potential to be much more powerful than switch access, which is slow and janky by comparison. Yet the people who receive the first generation of medical implants may find themselves in the same position as those who use switch technology now: functionally required to stay unemployed, poor, or even single as a condition of accessing the services keeping them alive.

      Musk may be right that we’re quickly approaching a time when BCI tech is practical and even ubiquitous. But right now, we don’t have a social consensus on how to apportion resources such as health care, and many disabled people still lack the basic supports necessary to access society. Those are problems that technology alone will not—and cannot—solve.

      S.I. Rosenbaum is a journalist based in Providence, Rhode Island, who plays the musical saw and has written for The New York Times and Slate.

42 comments