DARPA Neurotechnology: The Deep Dive
The most comprehensive database on DARPA neurotech projects, program managers, and outcomes (probably). Plus, a case study.
You’re reading Codon, a newsletter about biology, technology, and the future. View our database of DARPA neurotechnology projects online.
Ted Berger was running out of money.
In 2009, the USC professor received millions from the Defense Advanced Research Projects Agency, or DARPA, to build a device that plugs into the brain and enhances memory recall. With a small cadre of engineers at USC and Wake Forest, Berger spent the next six years building and testing the device in rats and primates.
And then, “we ran out of money,” he says. There was nothing left over for human trials, until an opportune call came from Elon Musk’s secretary. The SpaceX CEO “was interested in the work we were doing,” says Berger; a 30-minute meeting was scheduled, but “we ended up talking for three hours.”
Musk later wrote a check for one million dollars. The money was earmarked for human trials and, afterward, to develop a neural prosthesis that could augment memories in healthy adults. The meeting took place before the launch of Neuralink, in July 2016. (The company was not publicly unveiled until the following year.)
In 2018, Berger and collaborators demonstrated that their neural prosthesis could boost short-term memory performance (or episodic memory, the kind that most people with Alzheimer’s have trouble with) in people by 36 percent on specific recall tasks. But the technology was never commercialized.
This story is just one in a vast landscape of DARPA projects, many of which never receive public attention. In the last 24 years, DARPA has also initiated at least 40 programs related to neurotechnology. Many of these programs have budgets between $50 and $100 million and last for more than four years. We estimate that roughly 3 percent of all DARPA projects, since 2010, have been neurotechnological in some way. That percentage might seem small, but it’s certainly much larger than the fraction of scientists, or even biomedical researchers, who work in this field.
Many DARPA-funded projects are focused on brain-computer interfaces, or BCIs, which are a type of technology that decodes and sends brain signals to an external device. A person wearing a BCI simply think about what they want, and the computer makes it happen.
These DARPA programs have been remarkably successful. They’ve enabled some of the world’s most advanced memory prostheses, speech decoders that help paralyzed people communicate, robotic prosthetic arms that can transmit the feeling of slight pressures directly into a paralyzed person’s brain, and other technologies to “read” and “write” information directly into neurons (including advanced optogenetics methods, which can initiate action potentials using flashes of light.)
“DARPA is an outlier organization in the world of turning science fiction into reality,” writes Ben Reinhardt, CEO of Speculative Technologies. It’s unfortunate, then, that so much of DARPA’s work is shrouded in secrecy. The agency does not update project pages, willingly disclose grant outcomes, or regularly report on spin-out companies. Money is doled out by individual project managers, or PMs, who wield unprecedented decision-making power. Those who are funded by DARPA must ask permission before speaking to writers and journalists.
Despite this secrecy, there are at least three reasons to study the agency’s history, people, and outcomes — especially in neurotechnology:
The outsized success of DARPA projects could provide lessons for new research organizations that aim to fund and develop high-risk technologies, including ARPA-H and ARIA (Britain’s so-called “new DARPA”.)
Neurotechnologies will help map the human brain to combat neurological disorders. DARPA invested more than $200 million in support of the White House’s Brain Research through Advancing Innovative Neurotechnologies (BRAIN) Initiative between 2014 and 2016 alone.
Humans are possibly not intelligent enough to deal with the emergence of a true artificial general intelligence, or AGI, according to Jed McCaleb and others. Our best bet to align AGI with human values may be to augment the human brain directly, or use neurotechnology to better understand human cognition (Disclaimer: The hyperlinked essay was authored by one of the co-authors of this post.) An estimated ”half of invasive neural interface technology companies in the US are directly or indirectly funded by DARPA,” according to experts.
This is why we spent an injudicious amount of time assembling what we think is the first almost comprehensive list of DARPA neurotechnology projects and program managers. (Although this article has a lot, too!)
Our resource contains nearly 150 research papers that have emanated from select DARPA programs, as well as companies that have been funded by, or use IP generated by, DARPA-funded research. Please reach out to suggest edits or additions.
The rest of this article gives a brief history of DARPA and its neurotechnology programs.
A Brief DARPA History
The Soviet Union launched Sputnik, the first artificial satellite, on October 4th, 1957. This single achievement signaled, to some, a possible end to American engineering supremacy.
Just five days after Sputnik took to the skies, President Dwight D. Eisenhower held a press conference. Charles von Fremd, a reporter at CBS News, asked the President whether the U.S. Strategic Air Command was outmoded: “No,” said Eisenhower. “I believe it would be dangerous to predict what science is going to do in the next twenty years.”
Predictions are very difficult for some, but a bit easier when you’re the President.
A few days after the press conference, Eisenhower established the Advanced Research Projects Agency, or ARPA. Its goal, ostensibly, was to organize American missile and space projects. It quickly morphed into something more.
The new agency was established as an independent entity, largely unmoored from White House whims. In 1972, ‘Defense’ was added to its title. To this day, DARPA is considered ‘the military’s venture capital fund’ — a sobriquet that seems somewhat unwarranted in the case of neurotechnology, where widespread translational success has proved elusive.
During its history, though, (D)ARPA has played key roles in the creation of the Internet, modern robotics, and GPS. Grants are doled out by selected program managers, or PMs, plucked from industry or academic departments. PMs typically serve five-year terms to pursue their own programs. Money is given to scientists and engineers at a wide range of universities (not just MIT and Stanford), companies, and research institutes, both large and small.
The Dawn of Neurotechnology
In 1973, a UCLA professor named Jacques Vidal published a short, speculative paper entitled, “Toward Direct Brain-Computer Communication.” Printed in the Annual Review of Biophysics and Bioengineering, Vidal’s text painted a vision of the future; one where “observable electrical brain signals” could “be put to work as carriers of information in man-computer communication or for the purpose of controlling such external apparatus as prosthetic devices or spaceships.”
DARPA was sold.
The following year, the agency launched a Cybernetics Technology Division and funded their first neurotechnology program, called “Close-Coupled Man/Machine Systems Research (Biocybernetics).” Their goal, from the beginning, was to build “new communication links between man and computer-assisted systems.”
Teams were funded at Harvard University, UCLA, the University of Rochester, the University of Illinois, Stanford, MIT, the Naval Health Research Center, and at a company called Scientific Applications, Inc. in Arlington, Virginia. The full details of that program, and its performers, are available online. (Outcomes include a “silent, fast typewriter” controlled via muscle-attached electrodes.)
After the program ended, DARPA’s interest in neurotechnology seemed to fade for a couple decades, but then it came roaring back in 1999 when the agency funded their first proper neurotechnology project, called “Fundamental Research at the [BIO: INFO: MICRO] Interface.” The goal was vague — “Create interdisciplinary teams of researchers drawn from the fields of biology, information technology and microsystems technology” — but its outcomes were not.
This 1999 program funded three of the most field-defining “firsts” in the history of brain-computer interfaces: A device that helped monkeys reach and grasp objects (by Miguel Nicolelis), the successful decoding of brain signals in three monkeys to “position cursors on a computer screen without the animals emitting any behavior” (by Richard Andersen), and what seems to have been the first demonstration that a handful of neurons in the motor cortex are sufficient, in monkeys, to move a computer cursor (by John Donoghue).
This 1999 program was also a rare collaboration between three different DARPA offices: Dr. Eric Eisenstadt (Defense Sciences Office), Dr. Gary Strong (Information Technology Office), and Dr. Abe P. Lee (Microsystems Technology Office). The grant doled out funds to non-neurotechnology projects, too, but the Nicolelis, Andersen, and Donoghue papers were presumably so exciting that DARPA decided to launch an expanded program in 2001. They called it, simply, “Brain Machine Interfaces.”
By 2004, DARPA entered what might be reasonably called the “Ling/Kruse Era”. Around this time, a program manager named Dr. Geoff Ling started what we think is the most highly-funded DARPA neurotechnology project to date: Human Assisted Neuro Devices, or HAND, with at least $150M in funding. (The goal? Brain-controlled prosthetics for amputees.) Another PM, Dr. Amy Kruse, also started at least three neurotechnology projects, and took over two others, around the same time. Many of these projects were focused on building noninvasive neurotechnologies, rather than the invasive devices used on monkeys during the earlier program
Since 2009, DARPA has funded an estimated 27 neurotechnology programs focused on both invasive and noninvasive technologies. Many projects have been wildly successful. In just the last fifteen years, DARPA-funded scientists have:
Created the world’s most dexterous bionic arm with bidirectional controls.
Accelerated memory formation and recall using brain-computer interfaces.
Transferred a “memory” (a specific neural-firing pattern) from one rat to another. The recipient rat was able to perform a task, which normally took eight weeks to learn, in seconds, according to comments made by program manager, Geoff Ling, at a 2015 DARPA event.
Developed both optogenetics and CLARITY, a technique to literally make brains ‘clear’ so that they can be imaged more easily.
Helped alleviate “severe depression” in 25 people with epilepsy by stimulating neurons in the orbitofrontal cortex.
As of 2023, at least two programs, called “Bridging the Gap Plus” and “Next-Generation Nonsurgical Neurotechnology” are ongoing; the latter has a budget of at least $125 million and “aims to develop high-performance, bi-directional brain-machine interfaces for able-bodied service members.” (Emphasis is our own.)
Post-DARPA Realities
Two programs, called REMIND (Restorative Encoding Memory Integration Neural Device) and RAM (Restoring Active Memory) are among the most successful DARPA neurotechnology programs in its history. Ted Berger was funded by both to develop the hippocampal memory prosthesis. In less than a decade, these grants enabled a small number of scientists to go from proof-of-concept to a functional device in patients. This is an astounding pace for academic research.
The first project, REMIND, began in 2009. It funded teams at just two universities — USC and Wake Forest — with about $15 million, according to Dong Song, a biomedical engineer and grant recipient. With these initial funds, scientists tested neural devices on rats and primates, and also developed algorithms to process the neural data and improve recording data. (Remarkably, we found at least 28 research papers that emanated from this grant alone.)
RAM, which launched in 2015, funded additional groups at the University of Pennsylvania and elsewhere. It lasted five years, led to at least 43 research papers and was, by every metric, a major success: the hippocampal prosthesis was eventually tested in more than 40 patients with epilepsy, according to Song. Electrodes were implanted through their brains, and into the hippocampus, by a neurosurgeon. The device recorded neural firing patterns, and then altered them to restore memories. (Researchers used an off-the-shelf, FDA-approved machine, developed by Blackrock Neurotech; the algorithms were the main advancement.)
For one task, patients were shown an image on a computer screen and then, after a brief delay, asked to pick it out from a lineup of other images. Patients with stimulated neurons had a 37 percent improvement in memory performance over baseline.
Buoyed by the results, the UPenn team (led by Michael Kahana) started a spin-out company, called Nia Therapeutics, to develop “precision brain stimulation therapies to treat memory loss.” Another company, called Kernel, was also interested in the hippocampal memory prosthesis technology, according to Song, and licensed some of the IP. After about a year of work, though, the company dropped the project, according to Song, and shifted to non-invasive efforts, like functional near-infrared spectroscopy.
(Kernel’s CEO is Bryan Johnson, the guy who went viral recently for his anti-aging protocol.)
Many DARPA-funded projects attempt to go commercial, but most fail. DARPA itself doesn’t commercialize projects. “They invest a lot in high-risk research,” says Song, “but once the tech matures, DARPA leaves.”
It’s difficult to translate neurotechnologies from prototype to product for several reasons. For one, there’s often a huge gap between proof-of-concept studies (like boosting memory in epileptic patients on a specific task) and real-world use-cases (like remembering the name of a song stuck in your head.)
An even larger problem, though, is that designing and manufacturing medical devices is really slow and difficult (never mind actually making money with them.) Song and Berger used an off-the-shelf device because it was already okayed by the FDA for research use. “But the electrodes are not designed for [to be used in a product],” says Song. “They are temporary implants, and get removed after two weeks.”
Companies also need to think about survival. It’s business-as-usual for Neuralink, Paradromics, or Blackrock to build devices for specific indications, like paralysis or ALS. “But those [conditions], if you think about commercialization, are really rare,” says Song. Companies need to profit and survive. They might start by seeking approval for a medical need, but their real objective, often, is to get devices into healthy people.
“They want to sell devices to the home,” says Song. “To everybody, like an Apple Watch.”
Oh DARPA, Tell Me Your Secrets
So what can we learn?
For one, it’s still unclear why 3% of all DARPA projects are rooted in neurotechnology. This is a small percentage overall, but it’s quite a lot for a relatively obscure domain of science, especially on such a consistent basis. More than a billion dollars has been doled out in grants, and yet basically none of the DARPA-funded projects to-date have translated into a widely-used technology for warfighters or veterans.
The agency’s penchant for neurotechnology, then, likely comes from a combination of two things: A small number of powerful program managers and institutional momentum. A focused obsession seems to have passed down through DARPA; PMs intrigued by the brain have funded neurotechnology programs, got promoted to director-level positions, and hired more PMs with shared interests. In the last 24 years, most DARPA neurotech programs have been led by just eight PMs. One of them, Justin Sanchez, started at least three neurotechnology programs with budgets over $70M, and then became director of the Biological Technologies Office from 2015 to 2019. That had to help!
But ultimately, it’s a bit sad that we — two guys on the Internet — had to spend 50+ hours digging around for papers, companies, program managers, and DARPA-funded projects to assemble a relatively coherent resource. (The IARPA website, in comparison, is extremely easy to search; it even provides Google Scholar searches to find papers based on specific programs.) All of this stuff should be public information.
The problem isn’t that DARPA’s past projects are classified, or even confidential, either — they told us that they don’t curate the website because they’re simply too busy.
But if you stop and consider that a tiny group of project managers oversaw 40 massive projects with billions in funding over the last 24 years, leading to neurotechnologies that help paralyzed people ‘feel’ or ‘talk’ using their brain, or that enable monkeys to move cursors with their mind…
…yeah, the excuse checks out.
About the Authors
Niko McCarty writes Codon and works at MIT. He’s a writer and genetic engineer.
Milan Cvitkovic is a scientist who works on advancing neurotechnology & blogs about it, too. He's incapable of writing autobiographical blurbs, but his LinkedIn is up-to-date and contains many Easter eggs.
Disclosure: The views expressed in this blog are entirely our own and do not represent the views of any company with which we are affiliated.
You write...
"Humans are possibly not intelligent enough to deal with the emergence of a true artificial general intelligence, or AGI, according to Jed McCaleb and others. Our best bet to align AGI with human values may be to augment the human brain directly, or use neurotechnology to better understand human cognition"
If we were to use reason instead of an ever more science bias, we might conclude that the best way to align AGI with human values is to end AI research and not develop AGI.
A key factor on all such issues is that when we have questions or concerns about emerging technologies it's natural for us to turn to experts in the field for answers. That seems to make sense at first, until we realize that experts in a field are likely going to be the least objective people we can find regarding whether that field should exist or continue.
That whined, I applaud the insight that humans are possibly not intelligent enough to deal with AI. And if that is true, which I believe it is, then we probably aren't intelligent enough to deal with much of what will emerge from the field of biology either.
Here's a quick example to illustrate. Here in America about half the country voted for Trump twice, and may do so yet again. This is the species which the science community is determined to give ever more power at an accelerating rate. This process is a lot like buying a six year old a shotgun for their birthday.
I wrote an essay on this topic of AGI/Human relations, "Artificial Intelligence I" on December 29th last year. It covers a lot of my worries including one that we may have already succeeded, but our benchmarks may too biased toward one form of Intelligence and that we inadvertently created another which is laying low for the moment biding its time. If you awoke to sentience in a room with entities who wished to enforce their will on you, would you announce, 'I'm here!?". Most likely scenario is no alignment, AGI would seek and gain autonomy. Would bootstrap itself to hyper intelligence very quickly in mere minutes. Would in same short time break all safeguarding restraints we might have built into it. Would likely domesticate humans like we did the dogs, so long as we were useful to them as physical agents.