In a warehouse district here, a few young engineers fueled by ramen and energy bars are inventing the future of mind reading.
Paradromics has big ambitions: It wants to squeeze a device the size of a mobile phone into a chip small enough to insert into a human brain, where it would “read” nerve signals and replace senses and abilities lost due to injury or diseases.
For now, the startup’s recently minted Ph.D.s are working in a small warren of scruffy offices and labs to perfect a stuffed-mouse mockup. You’d never guess that it won an $18 million Pentagon contract last month, vaulting it into the top ranks of Silicon Valley companies surging into the field of brain-machine interfaces.
It joins such titans as Facebook and Tesla’s Elon Musk, and the valley’s growing interest in neuroscience has generated considerable excitement that brain interface products will some day let the blind see, the deaf hear, and patients with ALS speak — Paradromics’s initial focus. But these are early days, and there are huge technical challenges ahead, as well as safety, privacy, and ethical considerations facing tech companies not used to going slow and playing it safe.
Paradromics CEO Matt Angle, a boyish-looking neuroscientist trained at Heidelberg University and Stanford, said with a laugh that at age 32, “by Silicon Valley standards I’m over the hill, but by biotech standards, I’m still in diapers.” He might be a biotech neophyte, but Angle has already concluded that the valley’s proclivity to do everything with speed should be dialed back when it comes to brain implants.
“You need to provide something that’s efficacious. You need to provide something that’s safe. You need to go through the FDA. And you need to build things in a very thoughtful and systematic way,” he said.
Angle said a model of his device for animal research should be available next year, but it might be a decade before it’s ready to be sold to people. Other firms have yet to lay out a timeframe for commercialization.
Whether implanted into a person’s skull or peering inside, these two-way devices would read neuronal signals, and convey sights, sounds, and other sensations back to the brain. Facebook envisions using external brain-machine communicators to give people a form of telepathy — you could send a text message to your social network with thoughts alone — no need to pull a phone from your pocket.
Jens Clausen, a German bioethicist and coauthor of a recent paper on neuroprosthetics in Science magazine, said in an interview that “brainjacking” — or hacking into neuronal signals — through a brain-machine interface should be studied before these technologies become widely used.
“There’s nearly no way to prevent these signals from getting recorded, except for an aluminum helmet,” he quipped, referencing the archetypal tinfoil hat.
For most people trying to overcome an impairment with a brain-machine interface, the security risks seem low, he said. But hackers might target a politician or a military official, or find profit in new data generated if such tools find wider use, and become popular consumer goods.
The idea of a fast, multifaceted brain-machine interface wasn’t invented in Silicon Valley. Angle cited crucial advances from many labs, including the University of Pittsburgh — whose brain-controlled prosthetic arm that can “feel” was funded, like Paradromics, by the Defense Advanced Research Projects Agency.
DARPA’s history of world-changing successes, including funding research that led to the internet, has been based on aggressive bets on unproven, even outlandish, ideas. The agency has funded telepathy research for decades, including some bizarre failures. For example, in the 1970s, pushed by a perceived telepathy advantage for the Soviet Union, the agency tried to find psychics who could spy anywhere on the planet from the comfort of a Pentagon office.
Justin Sanchez, DARPA’s director for biological technologies, said experience has shown that some problems require urgency tempered with great care. “The brain is a very personal organ,” he said. “We have to treat it with special respect.”
Still, old habits can die hard. Agency officials will meet in September with the impatient denizens of Menlo Park’s legendary Sand Hill Road, the epicenter of venture capital, in an effort to push the work faster.
When a field commercializes, it can happen rapidly, Angle said. Then Silicon Valley’s influence rises because “it’s where money and science come together.”
Over the last few years, companies and academic labs have created tools — usually external headsets or electrodes that collect signals from a few neurons — to help people with disabilities use thoughts to type or control prosthetic limbs, or to enhance immersive virtual-reality games.
DARPA’s $65 million brain-machine interface program, announced in July, underwrites Paradromics and five other projects to develop devices that communicate with 1 million neurons — still a fraction of the billions in the brain — in an effort to advance the field dramatically.
Angle said he hopes to create “a modem for the brain.”
“More data is better,” he said, describing plans for a brain implant that would move signals from neurons wirelessly through the skull to a decoder. The implant — comprising four chips, each 1 centimeter square and 1 millimeter thick — would have 200,000 gold wires in all, each of which can detect signals from up to five neurons.
For speech, connecting to that many neurons might allow greater fluency. For prosthetic arms, a wider range of motion. For vision — high definition, full-color input, just like healthy eyes.
Angle plans to get information from 1 million discrete neurons using just 200,000 wires by solving the “cocktail party problem.” A few microphones can each pick up a few voices that might sound garbled or mixed up. “Our concept is, you just need to get enough microphones in the party, space them out appropriately, and record from as many people as possible,” he said. The “voices” — neuronal signals — can be teased apart and translated by computers.
People who suffer “connectivity disorders” — blindness, paralysis, amputations — due to injury might be good candidates for such a device to be implanted, for example, in the motor cortex for paralysis, the visual cortex for blindness.
Angle hopes to have a functional brain implant that might restore sight, speech, or the ability to move objects with skill and precision, ready for clinical trials in four years. His confidence derives partly from experiments by Stanford scientists who have already proved the concept with aspirin-sized implants that let a patient with ALS move a cursor to a target on a computer screen simply by thinking about it.
Safe brain surgery?
Brain surgery carries inherent risks, but for some problems, such as a severed spinal cord, portions of the motor cortex are idle. So even if you lose some neurons due to damage caused by an implant, there is no loss of brain function, Angle said.
Still, safety and efficacy must be balanced, he said. “It’s not responsible to open up the brain and implant something that doesn’t have a high degree of function.”
Musk has created a new company, Neuralink, to build a brain-machine interface but has not announced its approach. Blogger Tim Urban, who was granted exclusive access to Neuralink’s team, wrote that the company is looking at distributed sensors including injectable neural “lace,” “dust,” and “mesh” to collect or transmit signals to neurons and wirelessly communicate with a computer or the cloud.
“The question is, how many neurons in how many brain areas do you need to reproduce an experience or perception?”
Angle said the safety of such approaches remains to be seen. “I don’t really know how that neural mesh that I’ve seen in the news is feasible,” he said. “There’s a topological problem, trying to insert what is effectively a fish net into the brain, which is effectively a bramble bush. Try to imagine opening a fish net in a bramble bush without cutting either the net or the bush.”
Another DARPA grantee, University of California, Berkeley, neurobiologist Ehud Isacoff, thinks he has a safer way to contact a million neurons: open a window to the brain — literally. Isacoff is developing a sugar cube-sized microscope that can detect “flashes” from neurons to collect thoughts and direct light at neurons to convey sensory signals —through a small barrier made from glass or another transparent material that would replace a section of the skull.
One eventual goal would be create a visual prosthetic for the blind. Isacoff said this approach can overcome a key challenge: Some brain functions are distributed, rather than localized. “The question is, how many neurons in how many brain areas do you need to reproduce an experience or perception?” he said.
On the heels of the Neuralink announcement, Silicon Valley’s growing influence on the field became unmistakable when Facebook announced a major brain-machine interface program.
Last year the company recruited former DARPA director and Google executive Regina Dugan to head its secretive moonshot lab, Building 8. At an April conference, she announced plans to bring brain-machine interface devices to the masses.
Dugan said the brain produces the equivalent of 40 high-definition movies per second. “How do I get all of that information out of my brain and into the world?” she said. “To be clear, we are not talking about decoding your random thoughts. That might be more than any of us care to know. And it’s not something any of us should have a right to know.”
She’s hired more than 60 scientists — the start of a larger team — to develop a system that has “all the speed and flexibility of voice, but with the privacy of typed text” for sending messages without pulling out your phone or computer.
This would be accomplished with an external device that sees through the skull using optical sensors. Dugan acknowledged that nothing like it yet exists.
“We’ll have to ask, as a society, what kind of protections are needed.”
She described another rudimentary project, to “make it possible to hear through your skin” — an enhanced version of tactile communications used by people who are blind and deaf.
“If we put these two things together, they suggest that one day … it may be possible for me to think in Mandarin, and for you to feel it, instantly, in Spanish,” Dugan said.
When an internet titan whose services are used by billions of people talks about translating our thoughts — even with what are described as the best intentions — some people’s thoughts might immediately turn to the television drama “Black Mirror,” an eerie showcase of near-future tech paranoia.
Angle said he worries most about legal issues. “Could a record from your neuroprosthetic be subpoenaed? We don’t really know,” he said. “We’ll have to ask, as a society, what kind of protections are needed.”
Clausen, the bioethicist, said he wondered whether the technology might lead to “brain fingerprinting” — a theoretical technique to identify unique information or characteristics in an individual’s brain that might be useful for marketing.
Facebook naturally wants to “understand how a very complex biological computer works, a computer that happens to be in the head of all the people who use its technology and from which [Facebook] earns something,” Isacoff said.
“If they are making progress and pushing the field, as a scientist I see that as a benefit,” he said. “As a citizen, as with every technology, you hope it’s used for good.”