Of all Elon Musk’s exploits — the Tesla cars, the SpaceX rockets, the Twitter takeover, the plans to colonize Mars — his secretive brain chip company Neuralink may be the most dangerous.
What is Neuralink for? In the short term, it’s for helping people with paralysis. But that’s not the whole answer.
Launched in 2016, the company revealed in 2019 that it had created flexible “threads” that can be implanted into a brain, along with a sewing-machine-like robot to do the implanting. The idea is that these threads will read signals from a paralyzed patient’s brain and transmit that data to an iPhone or computer, enabling the patient to control it with just their thoughts — no need to tap or type or swipe.
So far, Neuralink has only done testing on animals. But in May, the company announced it had won FDA approval to run its first clinical trial in humans. Now, it’s recruiting paralyzed volunteers to study whether the implant enables them to control external devices. If the technology works in humans, it could improve quality of life for millions of people. Approximately 5.4 million people are living with paralysis in the US alone.
But helping paralyzed people is not Musk’s end goal. That’s just a step on the way to achieving a much wilder long-term ambition.
That ambition, in Musk’s own words, is “to achieve a symbiosis with artificial intelligence.” His goal is to develop a technology that helps humans “merg[e] with AI” so that we won’t be “left behind” as AI becomes more sophisticated.
This fantastical vision is not the sort of thing for which the FDA greenlights human trials. But work on helping people with paralysis? That can get a warmer reception. And so it has.
But it’s important to understand that this technology comes with staggering risks. Former Neuralink employees as well as experts in the field alleged that the company pushed for an unnecessarily invasive, potentially dangerous approach to the implants that can damage the brain (and apparently has done so in animal test subjects) to advance Musk’s goal of merging with AI.
Neuralink did not respond to a request for comment.
There are also ethical risks for society at large that go beyond just Neuralink. A number of companies are developing tech that plugs into human brains, which can decode what’s going on in our minds and has the potential to erode mental privacy and supercharge authoritarian surveillance. We have to prepare ourselves for what’s coming.
Why Elon Musk wants to merge human brains with AI
Neuralink is a response to one big fear: that AI will take over the world.
This is a fear that’s increasingly widespread among AI leaders, who worry that we may create machines that are smarter than humans and that have the ability to deceive us and ultimately seize control from us.
In March, many of them, including Musk, signed an open letter calling for a six-month pause on developing AI systems more powerful than OpenAI’s GPT-4. The letter warned that “AI systems with human-competitive intelligence can pose profound risks to society and humanity” and went on to ask: “Should we develop nonhuman minds that might eventually outnumber, outsmart, obsolete and replace us? Should we risk loss of control of our civilization?”
Although Musk is not alone in warning about “civilizational risk” posed by AI systems, where he differs from others is in his plan for warding off the risk. The plan is basically: If you can’t beat ’em, join ’em.
Musk foresees a world where AI systems that can communicate information at a trillion bits per second will look down their metaphorical noses at humans, who can only communicate at 39 bits per second. To the AI systems, we’d seem useless. Unless, perhaps, we became just like them.
A big part of that, in Musk’s view, is being able to think and communicate at the speed of AI. “It’s mostly about the bandwidth, the speed of the connection between your brain and the digital version of yourself, particularly output,” he said in 2017. “Some high bandwidth interface to the brain will be something that helps achieve a symbiosis between human and machine intelligence and maybe solves the control problem and the usefulness problem.”
Fast forward a half-dozen years, and you can see that Musk is still obsessed with this notion of bandwidth — the rate at which computers can read out information from your brain. It is, in fact, the idea that drives Neuralink.
The Neuralink device is a brain implant, outfitted with 1,024 electrodes, that can pick up signals from a whole lot of neurons. The more electrodes you’ve got, the more neurons you can listen in on, and the more data you’ll get. Plus, the closer you can get to those neurons, the higher quality your data will be.
And the Neuralink device gets very close to the neurons. The company’s procedure for implanting it requires drilling a hole in the skull and penetrating the brain.
But there are less extreme ways to go about this. Other companies are proving it. Let’s break down what they’re doing — and why Musk feels the need to do something different.
There are other ways to make a brain-computer interface. Why is Neuralink choosing the most extreme one?
Neuralink isn’t the only company exploring brain-computer interfaces (BCIs) for restoring people’s physical capabilities. Other companies like Synchron, Blackrock Neurotech, Paradromics, and Precision Neuroscience are also working in this space. So is the US military.
In recent years, a lot of the research that’s made headlines has focused on brain implants that would translate paralyzed people’s thoughts into speech. Mark Zuckerberg’s Meta, for example, is working on BCIs that could pick up thoughts directly from your neurons and translate them into words in real time. (In the long term, the company says it aims to give everyone the ability to control keyboards, augmented reality glasses, and more, using just their thoughts.)
Earlier success in the BCI field focused not on speech, but on movement. In 2006, Matthew Nagle, a man with spinal cord paralysis, received a brain implant that allowed him to control a computer cursor. Soon Nagle was playing Pong using only his mind.
Nagle’s brain implant, developed by the research consortium BrainGate, contained a “Utah” array, a cluster of 100 spiky electrodes that is surgically embedded into the brain. That’s only around one-tenth of the electrodes in Neuralink’s device. But it still enabled a paralyzed person to move a cursor, check email, adjust the volume or channel on a TV, and control a robotic limb. Since then, others with paralysis have achieved similar feats with BCI technology.
While early technologies like the Utah array protruded awkwardly from the skull, newer BCIs are invisible to the outside observer once they’re implanted, and some are much less invasive.
Synchron’s BCI, for example, builds on stent technology that’s been around since the 1980s. A stent is a metal scaffold that you can introduce into a blood vessel; it can be safely left there for decades (and has been in many cardiac patients, keeping their arteries open). Synchron uses a catheter to send a stent up into a blood vessel in the motor cortex of the brain. Once there, the stent unfurls like a flower, and sensors on it pick up signals from neurons. This has already enabled several paralyzed people to tweet and text with their thoughts.
No open brain surgery necessary. No drilling holes in the skull.
Musk himself has said that BCIs wouldn’t necessarily require open brain surgery, in a telling five-minute video at Recode’s Code Conference in 2016. “You could go through the veins and arteries, because that provides a complete roadway to all of your neurons,” he said. “You could insert something basically into the jugular and...”
After the audience laughed nervously, he added, “It doesn’t involve chopping your skull off or anything like that.”
In Neuralink’s early years, before the company had settled on its current approach — which does involve drilling into the skull — one of its research teams allegedly looked into the tamer intravascular approach, four former Neuralink employees told me. This team explored the option of delivering a device to the brain through an artery and demonstrated that it was feasible.
But by 2019, Neuralink had rejected this option, choosing instead to go with the more invasive surgical robot that implants threads directly into the brain.
Why? If the intravascular approach can restore key functioning to paralyzed patients, and also avoids some of the safety risks that come with crossing the blood-brain barrier, such as inflammation and scar tissue buildup in the brain, why opt for something more invasive than necessary?
The company isn’t saying. But according to Hirobumi Watanabe, who led Neuralink’s intravascular research team in 2018, the main reason was the company’s obsession with maximizing bandwidth.
“The goal of Neuralink is to go for more electrodes, more bandwidth,” Watanabe said, “so that this interface can do way more than what other technologies can do.”
After all, Musk has suggested that a seamless merge with machines could enable us to do everything from enhancing our memory to uploading our minds and living forever — staples of Silicon Valley’s transhumanist fantasies. Which perhaps helps make sense of the company’s dual mission: to “create a generalized brain interface to restore autonomy to those with unmet medical needs today and unlock human potential tomorrow.”
“Neuralink is explicitly aiming at producing general-purpose neural interfaces,” the Munich-based neuroethicist Marcello Ienca told me. “To my knowledge, they are the only company that is currently planning clinical trials for implantable medical neural interfaces while making public statements about future nonmedical applications of neural implants for cognitive enhancement. To create a general-purpose technology, you need to create a seamless interface between humans and computers, enabling enhanced cognitive and sensory abilities. Achieving this vision may indeed require more invasive methods to achieve higher bandwidth and precision.”
Watanabe believes Neuralink prioritized maximizing bandwidth because that serves Musk’s goal of creating a generalized BCI that lets us merge with AI and develop all sorts of new capacities. “That’s what Elon Musk is saying, so that’s what the company has to do,” he said.
The intravascular approach didn’t seem like it could deliver as much bandwidth as the invasive approach. Staying in the blood vessels may be safer, but the downside is that you don’t have access to as many neurons. “That’s the biggest reason they did not go for this approach,” Watanabe said. “It’s rather sad.” He added that he believed Neuralink was too quick to abandon the minimally invasive approach. “We could have pushed this project forward.”
For Tom Oxley, the CEO of Synchron, this raises a big question. “The question is, does a clash emerge between the short-term goal of patient-oriented clinical health outcomes and the long-term goal of AI symbiosis?” he told me. “I think the answer is probably yes.”
“It matters what you’re designing for and if you have a patient problem in mind,” Oxley added. Synchron could theoretically build toward increasing bandwidth by miniaturizing its tech and going into deeper branches of the blood vessels; research shows this is viable. “But,” he said, “we chose a point at which we think we have enough signal to solve a problem for a patient.”
Ben Rapoport, a neurosurgeon who left Neuralink to found Precision Neuroscience, emphasized that any time you’ve got electrodes penetrating the brain, you’re doing some damage to brain tissue. And that’s unnecessary if your goal is helping paralyzed patients.
“I don’t think that tradeoff is required for the kind of neuroprosthetic function that we need to restore speech and motor function to patients with stroke and spinal cord injury,” Rapoport told me. “One of our guiding philosophies is that building a high-fidelity brain-computer interface system can be accomplished without damaging the brain.”
To prove that you don’t need Muskian invasiveness to achieve high bandwidth, Precision has designed a thin film that coats the surface of the brain with 1,024 electrodes — the same number of electrodes in Neuralink’s implant — that deliver signals similar to Neuralink’s. The film has to be inserted through a slit in the skull, but the advantage is that it sits on the brain’s surface without penetrating it. Rapoport calls this the “Goldilocks solution,” and it’s already been implanted in a handful of patients, recording their brain activity at high resolution.
“It’s key to do a very, very safe procedure that doesn’t damage the brain and that is minimally invasive in nature,” Rapoport said. “And furthermore, that as we scale up the bandwidth of the system, the risk to the patient should not increase.”
This makes sense if your most cherished ambition is to help patients improve their lives as much as possible without courting undue risk. But Musk, we know, has other ambitions.
“What Neuralink doesn’t seem to be very interested in is that while a more invasive approach might offer advantages in terms of bandwidth, it raises greater ethical and safety concerns,” Ienca told me. “At least, I haven’t heard any public statement in which they indicate how they intend to address the greater privacy, safety, and mental integrity risks generated by their approach. This is strange because according to international research ethics guidelines it wouldn’t be ethical to use a more invasive technology if the same performance can be achieved using less invasive methods.”
More invasive methods, by their nature, can do real damage to the brain — as Neuralink’s experiments on animals have shown.
Ethical concerns about Neuralink, as illustrated by its animals
Some Neuralink employees have come forward to speak on behalf of the pigs and monkeys used in the company’s experiments, saying they suffered and died at higher rates than necessary because the company was rushing and botching surgeries. Musk, they alleged, was pushing the staff to get FDA approval quickly after he’d repeatedly predicted the company would soon start human trials.
One example of a grisly error: In 2021, Neuralink implanted 25 out of 60 pigs with devices that were the wrong size. Afterward, the company killed all the affected pigs. Staff told Reuters that the mistake could have been averted if they’d had more time to prepare.
Veterinary reports indicate that Neuralink’s monkeys also suffered gruesome fates. In one monkey, a bit of the device “broke off” during implantation in the brain. The monkey scratched and yanked until part of the device was dislodged, and infections took hold. Another monkey developed bleeding in her brain, with the implant leaving parts of her cortex “tattered.” Both animals were euthanized.
Last December, the US Department of Agriculture’s Office of Inspector General launched an investigation into possible animal welfare violations at Neuralink. The company is also facing a probe from the Department of Transportation over worries that implants removed from monkeys’ brains may have been packaged and moved unsafely, potentially exposing people to pathogens.
“Past animal experiments [at Neuralink] revealed serious safety concerns stemming from the product’s invasiveness and rushed, sloppy actions by company employees,” said the Physicians Committee for Responsible Medicine, a nonprofit that opposes animal testing, in a May statement. “As such, the public should continue to be skeptical of the safety and functionality of any device produced by Neuralink.”
Nevertheless, the FDA has cleared the company to begin human trials.
“The company has provided sufficient information to support the approval of its IDE [investigational device exemption] application to begin human trials under the criteria and requirements of the IDE approval,” the FDA said in a statement to Vox, adding, “The agency’s focus for determining approval of an IDE is based on assessing the safety profile for potential subjects, ensuring risks are appropriately minimized and communicated to subjects, and ensuring the potential for benefit, including the value of the knowledge to be gained, outweighs the risk.”
What if Neuralink’s approach works too well?
Beyond what the surgeries will mean for the individuals who get recruited for Neuralink’s trials, there are ethical concerns about what BCI technology means for society more broadly. If high-bandwidth implants of the type Musk is pursuing really do allow unprecedented access to what’s happening in people’s brains, that could make dystopian possibilities more likely. Some neuroethicists argue that the potential for misuse is so great that we need revamped human rights laws to protect us before we move forward.
For one thing, our brains are the final privacy frontier. They’re the seat of our personal identity and our most intimate thoughts. If those precious three pounds of goo in our craniums aren’t ours to control, what is?
In China, the government is already mining data from some workers’ brains by having them wear caps that scan their brainwaves for emotional states. In the US, the military is looking into neurotechnologies to make soldiers more fit for duty — more alert, for instance.
And some police departments around the world have been exploring “brain fingerprinting” technology, which analyzes automatic responses that occur in our brains when we encounter stimuli we recognize. (The idea is that this could enable police to interrogate a suspect’s brain; their brain responses would be more negative for faces or phrases they don’t recognize than for faces or phrases they do recognize.) Brain fingerprinting tech is scientifically questionable, yet India’s police have used it since 2003, Singapore’s police bought it in 2013, and the Florida state police signed a contract to use it in 2014.
Imagine a scenario where your government uses BCIs for surveillance or interrogations. The right to not self-incriminate — enshrined in the US Constitution — could become meaningless in a world where the authorities are empowered to eavesdrop on your mental state without your consent.
Experts also worry that devices like those being built by Neuralink may be vulnerable to hacking. What happens if you’re using one of them and a malicious actor intercepts the Bluetooth connection, changing the signals that go to your brain to make you more depressed, say, or more compliant?
Neuroethicists refer to that as brainjacking. “This is still hypothetical, but the possibility has been demonstrated in proof-of-concept studies,” Ienca told me in 2019. “A hack like this wouldn’t require that much technological sophistication.”
Finally, consider how your psychological continuity or fundamental sense of self could be disrupted by the imposition of a BCI — or by its removal. In one study, an epileptic woman who’d been given a BCI came to feel such a radical symbiosis with it that, she said, “It became me.” Then the company that implanted the device in her brain went bankrupt and she was forced to have it removed. She cried, saying, “I lost myself.”
To ward off the risk of a hypothetical all-powerful AI in the future, Musk wants to create a symbiosis between your brain and machines. But the symbiosis generates its own very real risks — and they are upon us now.