Thinking Machines

Many in Silicon Valley—including Elon Musk and Facebook—see brain-computer interfaces (BCIs) as the next technology gold rush.

marco lopes

3 … 2… 1… Go!”

On cue, a collection of small white drones rose into the air to kick off a very unusual race. Instead of using the usual remote controllers, the students in this Florida gymnasium were piloting their drones with their minds. Each wore a special headset designed to recognize electrical patterns in the brain associated with specific directions. They could focus and think “forward” to move their drone forward or think “left” or “right” to move their drone from side to side.

“We did the drone race to show that this is a viable technology with huge potential,” said Juan Gilbert, a professor of computer science at the University of Florida who organized the race for his students. “People didn’t believe it; they thought the race would be rigged and you can’t move things with your thoughts.”

Human beings might not yet be capable of “Carrie”-like feats of telekinesis, but technology is fast closing the gap with so-called brain-computer interfaces. Gilbert envisions a future in which people sport brain-reading devices like they used to wear wristwatches, and he’s not alone. Many in Silicon Valley see brain-computer interfaces (BCIs) as the next technology gold rush.

University of Florida professor Juan Gilbert (center) working with students and their brain-controlled drone.
University of Florida professor Juan Gilbert (center) working with students and their brain-controlled drone.

Tech entrepreneur Bryan Johnson founded a company called Kernel in 2016 to develop implanted neural prosthetics, although initially the company’s main focus is on research into key brain functions. Last April, Regina Dugan, who then headed Facebook’s Building 8 advanced technology division, announced at a conference in San Jose that the group is developing a cap for external wear that is capable of translating the user’s brain patterns into typed messages. Soon, she maintained, we will be able to type 100 words per minute with just our thoughts.

Even visionary entrepreneur Elon Musk, the man behind Tesla and SpaceX, is getting into the game via his latest startup venture, Neuralink. Musk famously fears that artificial intelligence eventually will outstrip human capability. His response: connect humans to computers with an Internet-enabled brain implant so they can upload and download thoughts. It’s a bit like “The Matrix,” in which human beings are plugged directly into the movie series’ titular network via implanted ports at the base of the skull — except instead of being exploited by a hostile AI, we work in tandem with it.

All of this is welcome news to Kevin Warwick, a British engineer at Coventry University and author of “I, Cyborg,” who has long warned of the threat of AI and advocated such augmentation, to the point of turning himself into a human guinea pig. In 1998, Warwick implanted a radio transmitter into his arm and used it to control doors, lights and other devices in his lab with just a snap of his fingers. Four years later, he had a neural electrode array surgically connected to his central nervous system for several months and was able to use it to compel a robotic arm to mimic his own arm movements. “I am pleased that the topic is on the agenda, that yes, AI is a potential danger, but we can potentially turn it our advantage by linking up directly with it,” he says.

Kevin Warwick, a British engineer at Coventry University, holds a small transponder that can be implanted in a human. Warwick has a similar chip in his arm that allows him to control appliances in his office.
Kevin Warwick, a British engineer at Coventry University, holds a small transponder that can be implanted in a human. Warwick has a similar chip in his arm that allows him to control appliances in his office.

Fears of artificial intelligence aside, there are very good reasons to consider integrating human brains with computers. Human beings cannot store large amounts of information and recall it perfectly again and again like a computer can. And a computer can rapidly transfer all that stored information to another device, whereas it takes years for a student to learn enough to equal or surpass a teacher.

Where humans excel — for now, anyway — is in imagination and improvisation. We can spot patterns and pick up new skills relatively quickly compared to a computer. Systems like those envisioned by Musk and other BCI entrepreneurs could successfully merge the best of both worlds, turning human beings into “cobots.”

Making that futuristic world a reality requires a biologically compatible system that interacts with brain cells, capable of accurately reading the patterns of the electrical activity of neurons and translating it into action. But it is no easy feat to build such a system. It requires a means of recording neuron activity (typically electrode sensors), a small wireless chipset to transmit the signals, and specially designed algorithms to translate those signals into actions. It could be external, placing electrodes onto the scalp or forehead via a wearable cap, similar to standard medical EEGs.

Alternatively, the electrodes could be implanted directly into the brain, as with the BrainGate device, a successor to the array Warwick experimented with back in 2002. External brain-computer interfaces are non-invasive, but less accurate. There is more external noise interfering with the signals, making them more difficult to interpret. Brain implants, usually called brain-machine interfaces (BMIs), have greater accuracy but require invasive, risky brain surgery to put the electrodes in place.


Elon Musk’s approach seems to focus on implants. Musk has been cagey about specifics of Neuralink’s nascent core technology, dubbed “neural lace.” So far, the company’s bare-bones website only lists a handful of job openings. Most of the public details about Neuralink appeared in a massive 36,000-word illustrated blog post by Tim Urban, a blogger/cartoonist and Musk confidante.

We do know the concept is based on work by co-founder Dongjin (DJ) Seo of the University of California, who has suggested that sprinkling tiny sensors the size of dust particles throughout the brain’s cortex could be the basis for an entirely new kind of implanted brain-machine interface. The sensors would be fabricated on a lace-like wire array, which could then be dipped into the brain to be embedded. Urban colorfully described this neural lace as a “wizard hat for the brain.”

To date, researchers have focused largely on surgically implanted devices for medical conditions. BrainGate II is now in clinical trials to treat patients with spinal cord injuries or Lou Gehrig’s disease, with promising initial results. A 2012 study published in Nature reported that two paralyzed subjects successfully controlled a robotic arm with their implants. One of them managed to drink unaided from a bottle for the first time in 15 years. Medtronic has implanted electrodes for deep-brain stimulation into the brains of many Parkinson’s patients to control the shaking and impaired movement that often accompany the disease. Much like a pacemaker regulates heart rhythm, these implants serve as a pacemaker for the brain.

Still, Neuralink has a long road ahead to create an implantable device for general use. The company must demonstrate that its procedure can be done repeatedly, that the system can operate with high accuracy, and that it will last for a sufficient period of time. Next there must be a clinical trial to gain FDA approval. As any pharmaceutical or medical company knows, FDA trials are expensive and time-consuming. NeuroPace first began developing a brain implant to control epileptic seizures in 1997, but it wasn’t approved for clinical use until 2013. It would take at least that long for a commercial brain implant to gain approval.


Then there is what is known as the “squick” factor, a term that merges “squeamish” and “ick.” The benefits of brain implants for those who are paralyzed or suffer from debilitating seizures easily outweigh the risks. But most healthy people aren’t likely to be willing to shave their heads and undergo risky invasive surgery just to get a brain boost. This is borne out by a recent Pew Research Center study, which found that people are far more wary than enthusiastic about brain implants by a 66-32 margin.

They have good reason to feel that way. The risks are not trivial and include coma, bleeding inside the brain, seizures and infection. Placing something like Neuralink’s neural lace in the brain would require careful surgery, increasing risk of infection. Should a device malfunction, more surgery would be needed to repair or remove it. Case in point: In 2014, neurologist Phil Kennedy, one of the early pioneers of BMIs, paid a surgeon in Belize $30,000 to implant electrodes into the motor cortex of his brain. His brain swelled from an unexpected spike in blood pressure during the procedure, and Kennedy was temporarily paralyzed and unable to speak. The incision in his skull never really closed, forcing Kennedy to have the implants removed after just a few weeks.

That’s why David Eagleman, a Stanford neuroscientist and adviser to Kernel, told Wired that he thinks implanting electrodes directly into healthy brains is “doomed from the start.” Gilbert agrees. “I think implants are necessary and valuable in the medical context,” he says. “I just think when it comes to popularizing these into everyday use, people are not going to go for an implant.”

Krishna Shenoy, a pioneer of neural prosthetics at Stanford University, working with a robotic arm that uses a proximity detector to help it pick up a wooden block.
Krishna Shenoy, a pioneer of neural prosthetics at Stanford University, working with a robotic arm that uses a proximity detector to help it pick up a wooden block.

Krishna Shenoy, a professor of electrical engineering at Stanford University, isn’t so pessimistic, pointing out that reconstructive cosmetic surgery became commercial fairly quickly. While he personally doesn’t advocate for it, if the brain-implant procedure could be shown to be safe and effective, perhaps attitudes would shift over time. There are already hundreds of thousands of people suffering from Parkinson’s walking around with four-inch electrodes surgically placed deep in the brain. “The deep brain stimulator showed that people can have stuff implanted in their brain and walk around and be completely accepted in society,” Shenoy says. “That was a huge step.”


Many of the companies seeking to commercialize external BCIs are focusing on EEG-based devices that target the gaming community, although they can be used for other purposes. San Francisco-based Emotiv’s Epoc headset — which has two arms snaking across the forehead and is anchored behind the ear — can replace standard joysticks in video games and control motorized skateboards and drones. That’s what Juan Gilbert’s students used for their brain-controlled drone race. Technically, the device doesn’t exactly “read” your thoughts. Rather, users “train” the device to recognize their brain patterns by thinking of the same objects or motions repeatedly and assigning those thought patterns to specific tasks.

San Diego-based Neuroverse is developing the BrainStation, described as a “Fitbit for the brain.” It monitors your brain waves while you go about your daily activities. The device is sleek, small and elegant, attaches to your forehead like the proverbial third eye, and operates wirelessly via Bluetooth to connect to a smartphone or tablet app. The system offers cognitive tests, games, brain wave tracking, social networking and control capabilities. (You can take a picture with your iPhone just by blinking.) The prototype is being used in a number of clinical studies, including measuring brain activity in patients suffering from Parkinson’s, schizophrenia and PTSD.

Facebook is taking a different tack, using an optical technique that beams white light through the skull and detects what bounces back. The brain-to-text device would operate much like Apple’s voice-responsive virtual assistant Siri — except you’ll be able to dictate with your mind. Like Emotiv’s headset, Facebook’s device won’t specifically read your thoughts. According to chief scientist Mark Chevillet,, the focus is on detecting and transmitting the neural signals that occur “right before you would say [a word] out loud.”

Facebook hopes to have a working prototype within a year, a timeline many view as unrealistic. It’s certainly possible to measure how neural cells reflect light as a clue to deciphering brain activity. “If you use the right wavelength of light, you could actually see changes in blood flow and potential neural activity directly,” Stanford’s Shenoy says. The question is whether the technique can be scaled up to achieve a typing rate of 100 words per minute.

Shenoy is using the BrainGate to enable paralyzed volunteers to send text messages with their minds. Using an implant with 100 electrodes, Shenoy has achieved a best of eight words per minute. He is skeptical that what he calls “a hat with lights in it” could perform 10 times better than the BrainGate within two years. “I deeply want it to work non-invasively,” he said. “But I think it will take time to understand if there is enough signal there that can be scaled up.”

At this point, the biggest challenge to widespread adoption of wearable BCIs might be aesthetics. Neuroverse’s BrainStation, for instance, is relatively small and fits neatly on the forehead, yet users still balk at the thought of wearing it in public. “You have to put it on your face, and that is valuable real estate for people,” says CEO Ricardo Gil-da-Costa. “How can we convince people that this is interesting and useful enough for you to wear it?” A possible solution: Make the device a fashion statement, regardless of its ultimate function — perhaps helped along by well-placed celebrity endorsements.

Gil-da-Costa envisions a day when BCIs are just another wearable electronic device — not just for gaming applications, but as much a part of daily life as a smartphone or a Fitbit tracker. While external BCIs have reached the point where they can be commercialized, he believes implanted versions aren’t quite there yet. But perhaps widespread adoption of devices such as BrainStation will help people be more accepting of full implants when that technology matures.

That day might come sooner than we think. “The thing that people, I think, don’t appreciate right now is that they are already a cyborg,” Musk told Urban for his graphic manifesto. “You’re already a different creature than you would have been 20 years ago, or even 10 years ago. If you leave your phone behind, it’s like missing-limb syndrome.”


One of the main impetuses for work on brain-computer interfaces is that it is seen by Musk and others as a defensive strategy against pure artificial intelligence. Musk hasn’t minced words about what he perceives as an impending threat from AI, calling it “a fundamental risk to the future of human civilization” in a speech last July. “I keep sounding the alarm bell, but until people see robots going down the street killing people, they don’t know how to react because it seems so ethereal.” Physicists Stephen Hawking and Max Tegmark also have voiced concerns, along with Apple co-founder Steve Wozniak.

But others dismiss this as fear-mongering. “AI will not kill us. That’s science fiction,” Max Versace, CEO of Neurala, a Boston-based neural network software company, told CNBC last September.

The truth may fall somewhere in between. It’s not necessarily paranoid to be wary of the repercussions of an AI that eventually surpasses human intelligence, given the rapid advances in that field. Gil-da-Costa doesn’t feel the same sense of urgency as Musk, but he agrees that the notion isn’t entirely absurd. Computers keep getting smarter, and our world is becoming increasingly automated. “The truth is, our dependencies are pretty simple and basic,” he says. A rudimentary AI hack that took out the electrical grid or shut down water supplies could cripple modern civilization.

Kevin Warwick, for his part, has been sounding the alarm for the past 20 years. “If AI really does become truly intelligent, and we defer to it and allow it to have physical capabilities, and it can learn and decide what to do, then it is potentially dangerous,” he says, pronouncing himself gratified that leaders of Musk’s stature are joining the BCI cause. “Hopefully, now people will take it seriously.”

Keep reading: The Roots of Sexism In Silicon Valley

Science journalist Jennifer Ouellette is a senior writer at Ars Technica and former science editor for Gizmodo.
Advertisement - Continue Reading Below
More From Premium