RYAN EVANS
It’s back-to-school season, and the smell of defense reform is in the air. Depending on how much of a wonk you are, the mere mention of the U.S. defense acquisition system will either make you fall asleep (normal person) or tear your hair out (Hill staffer, Pentagon drone, etc.). The word most associated with the defense acquisition system is probably “broken.” Sen. John McCain, the chairman of the Senate Armed Services Committee, has described it as “a national security crisis.” And he is right. McCain has led the charge to reform the system, and while this cause is noble, it might also be lost. It could be that the defense acquisition system is so far gone — so inefficient, unwieldy, and misaligned with commercial incentives — that any hopes of reforming and repairing it on a reasonable timeframe are unrealistic.Steve Blank lecturing at the first H4D class at Stanford
So what, then, is the answer? When I was visiting Stanford University earlier this year, the state of the defense acquisition system came up in conversation with Steve Blank, a guru in the world of Silicon Valley entrepreneurialism and one of the leaders behind a new course, Hacking for Defense (H4D), that is changing the future of defense innovation. Blank is perhaps best known as the founder of the Lean Startup movement, which was popularized by his former student, Eric Ries, who wrote theeponymous book that has become a must-read in the entrepreneurial canon (Ries, also a successful businessman, is up to somepathbreaking work of his own by rethinking the modern stock exchange). But beyond Blank’s fame in the tech sector, he is a Vietnam-era Air Force veteran who cares a lot about defense. Blank is more familiar than most with the historically tight relationship between Silicon Valley and the Pentagon, which he documented in his “Secret History of Silicon Valley.”
In talking with Blank, I mentioned the word “reform.” He told me that defense reform was the wrong answer. He explained the only hope was to set up a parallel system or systems to compete with the existing system, which would either wither or evolve over time. H4D aims to do exactly that, connecting islands of innovation in the defense and intelligence communities with the Silicon Valley innovation mindset.
Another member of the H4D teaching team, Jackie Space, an Air Force veteran and entrepreneur, offered a bold but less revolutionary take. She explained:
The way I see it is that H4D draws on the already existing systems and maximizes their access in a short amount of time. We don’t have to wait for acquisition reform to become efficient in how we solve national security problems as we have more than enough resources and people. It was startling what was uncovered through the weekly interviews of government and private industry stakeholders and how the original problem hypothesis pivoted and matured. I’ve seen military technical requirements written from a much thinner foundation of market research than what these students compiled in 10 weeks.
Whether pushing for a revolution or a dramatic evolution, this key group of innovators in Silicon Valley did not wait on anyone in Washington to make it happen. Blank joined forces with three former military officers who have reputations for busting silos in the Pentagon — Pete Newell, Joe Felter, and Space (full disclosure: They are friends of mine) — as well as Tom Byers, a leading figure at Stanford. The result was H4D. Newell and Space are veterans of the Army’s Rapid Equipping Force (which Newell directed), and both are fellows at the Center for Technology and National Security Policy at the National Defense University, which is supporting the project. Felter, on faculty at Stanford, is a former Special Forces officer who ran the Counterinsurgency Advisory and Assistance Team in Afghanistan. Newell explained:
Steve and I were talking in front of a dry-erase board and we realized that the problem curation methods I had built in uniform at the REF and my later efforts with Joe to recruit tech talent to solve battlefield problems were virtually the same as Steve’s methods from his decade-old Lean LaunchPad class. H4D is the result of that conversation.
When you get a teaching team like this together, you aren’t going to get just another a classroom exercise with lectures and notes. They were going to hack the defense acquisitions process and rewire its code with innovative viruses. And they were going to do it in ten weeks.
The concept behind the class is simple. The teaching team solicited a number of unclassified “problems” from the defense and intelligence communities. These included a poor ability to track vital signs among the Navy’s deep sea divers, how to protect soldiers from off-the-shelf drones conducting intelligence missions or even attacking them, how best to counter social media usage of adversaries such as the Islamic State, and more. Students at Stanford, from undergrads to MBA candidates to doctoral students, applied to join the class. Applicants attended information sessions and mixers where they met other applicants and formed teams. They then pitched ideas in a competitive process; only the best pitches and most impressive teams got into the class. This up-front commitment of time that didn’t even guarantee admittance ensured that only the most dedicated students joined the class.
Since this is Stanford you won’t be surprised to hear that these are some very impressive young people. They included software and mechanical engineers, future business executives, and even some veterans with skin in the game. Some teams boasted members from other countries — since the problems they would try to solve are unclassified, it made sense to loop in talented people no matter where they were from.
These teams received a lot of help: advisors, mentors, and military liaisons already based full-time at Stanford. They could also reach out to their sponsors in different organizations in the Department of Defense and intelligence community. Former Secretary of Defense Bill Perry, who played a major role in what is now known as the U.S. military’s second offset strategy, and is a himself an emeritus professor from Stanford, also offered his support.
The class embarked on a 10-week journey, drawing upon the considerable expertise of the teaching team, Blank’s Lean LaunchPad curriculum and mission model canvas to find and build solutions to these problems, many of which have life-saving and battle-winning implications. The teams “got out of the building,” as Blank likes to phrase it, and met repeatedly with sponsors and other stakeholders. They had to build and rebuild minimal viable products (MVPs) — prototypes used to elicit feedback from sponsors that would then inform a redesign of the MVP. This process refined the MVPs over and over again. In each weekly class meeting, every team presented on its progress and subjected itself to tough love from the teaching team. I sat in on two classes and watched the final class (now available for anyone to watch) via livestream. After the first class, I pulled out my phone and wrote several senior contacts back in Washington to tell them to get on a plane and come watch this for themselves.
All eight teams made impressive progress, but two stood out to me.
What is Happening in Seas Patrolled by the 7th Fleet?
The 7th Fleet is responsible for a massive chunk of the Earth that encompasses more than 124 million square kilometers. From east to west, it runs from the International Date Line to the border between India and Pakistan. From north to south, it runs from the Kuril Islands to the Antarctic. Its area of operations includes countries with the five largest militaries in the world after the United States, along with five U.S. treaty allies.
The problem is, the 7th Fleet cannot keep track of all activity in this vast area. They want what’s called a common operating picture. In service of that aim, the 7th Fleet told H4D that they wanted a distributed and disposable intelligence, surveillance, and reconnaissance (ISR) sensor network.
Enter the H4D Team Sentinel.
Darren Hau, a junior majoring in electrical engineering at Stanford, told me about his team’s experience over the phone. This is Hau’s second time as a Stanford undergrad. He took two years off to start a solar power startup called Dragonfly Systems that was eventually purchased by SunPower, one of the largest solar companies in the United States, which itself was started out of Stanford in the 1980s. Hau’s teammates included Jared Dunnmon, a Ph.D. candidate in mechanical engineering with an M.B.A.; Atsu Kobashi, who is getting an M.S. in electrical engineering; and Rachel Moore, joint M.B.A. and environment and resources M.S. candidate.
Team Sentinel agreed that this distributed and disposable ISR tech was a potential solution, but they tried to bring the issue back to its original problem. They went to the sponsor and asked why they thought they wanted this sensor network. Did they want the network there all the time, focused on a specific area, or quickly deployable during a crisis in a hotspot? They also spoke to other intelligence personnel, operators, as well as academics at the nearby Naval Postgraduate School.
They discovered that, as it currently stands, intelligence personnel at 7th Fleet headquarters are perpetually aggregating and trying to integrate data from multiple sources, often manually. They are dealing with two major hurdles. First, the data is incomplete. The second issue is data latency, meaning that there are different time lags associated with the various sources of data, which makes an up-to-date common operating picture impossible with their current technology. Team Sentinel learned that sensors would be useful, but wouldn’t necessarily solve the upstream problem of integrating that data effectively and efficiently.
So the team converged on a software solution rather than the hardware of a sensor network and set about building an MVP. They wanted to create, as Hau described it, “command-and-control for the future.” They wanted something that would suck in and integrate multiple data feeds in such a way that would allow for analysis to provide a common operating picture. They sought to build something useful for everyone, from analysts to staff officers to decision-makers.
As they worked on this MVP, they kept running up against the classification wall. Not all the stakeholders could share useful information with them because much of it was secret. They had to get a better sense of what tools, data sets, and programs were already out there and get feedback on their MVP, but classification restrictions were getting in the way. To get around this wall, the team members looked for similar problems faced by organizations that don’t classify their information and settled on illegal fishing. This might sound strange, but it is actually a useful analogue problem.
Estimates hold that illegal fishing nets an amount of fish equal to somewhere between 14 and 33 percent of the world’s legal catch. Illegal fishing can involve vessels from one country fishing without permission in the territorial waters of another country or just simply vessels violating their own country’s fishing laws. Given the shocking scale of fisheries depletion (reviewed by Claude Berube in a War on the Rocks article), this is a serious problem with grave consequences to livelihoods, national economies, and ecosystems around the globe. It is also a wicked problem that involves activities taking place by a multitude of actors moving amid a plethora of jurisdictions across vast expanses of ocean.
Stopping illegal fishing is a tough proposition that starts with being aware of where and when it is happening and ends with some form of interdiction. It presents many of the same problems the 7th Fleet is trying to solve by building a common operational picture: data latency, information sharing, and integrating multiple data streams within governments and between allies or partners.
Hau found out about the illegal fishing problem from friends in the U.S. Coast Guard. The Coast Guard works closely with the 7th Fleet on this issue — a useful reminder that our maritime forces are not just about defense and warfighting, but also about protecting the global commons and maintaining legal access to them. The military liaisons on the course, who were all resident at the Hoover Institution, included Coast Guard officers who helped the team explore this problem, introduce them to stakeholders, and refine their MVP.
Team Sentinel now possesses a wireframe MVP — a basic idea of what the product would look like — that has buy-in from many of its stakeholders. The course is over, but the members of Team Sentinel aren’t done. They are working on the technical side to build a higher-level MVP. They don’t necessarily plan on starting a company for the time being, choosing instead to keep it as an educational project through which they can work to engage a wide variety of stakeholders. It is too early for them to know if this product could be the basis for a viable company, but they are looking for organizational support —grants, other sources of funding, and partner organizations, including organizations working the illegal fishing problem.
Dedicated Drones for Infantry
For Kevin Mott, his team’s assigned problem was particularly personal. Mott is a U.S. Army officer who deployed to Iraq once and Afghanistan twice. On his first tour in Afghanistan, he was shot on the second month of his deployment. He rehabilitated quickly enough to return to his unit before their tour was over. Mott is at Stanford as an international policy studies grad student thanks to the Downing Scholarship, which is run by the Combatting Terrorism Center at West Point, formerly run by Joe Felter, one of the H4D instructors.
The other members of this team, dubbed Skynet, included Sam Gussman, a symbolic systems engineer; Olga Musayev, a law student with a background in data science; and Alvin Goh, an embedded systems engineer. The problem they were assigned, sponsored by Special Operations Command (SOCOM), was called “autonomous UAV control and threat detection.” Basically, SOCOM wanted to know how their units could use drones to increase situational awareness of threats while keeping operators holding their guns and focused.
By the first class, Team Skynet had carried out ten interviews to help define the problem more clearly. Mott was able to draw on his Army network to set up many of these interviews, which were carried out by the other members of the team so that they could have a better understanding of what combat was like. Their first idea was a haptic feedback system by which the drone would tell operators which direction to look in by basically poking them in the arm or providing an auditory signal. Their interviewees almost uniformly disapproved of that idea, worried they might miss the signal or be distracted by a poke.
Team Skynet took a step back and conducted more interviews. SOCOM said it wanted this to be a commercially available, off-the-shelf solution rather than some new hardware built from scratch. In some ways, this made the task even harder. To help, SOCOM sent them a drone, the DJI Phantom 3, which is sold at Best Buy. SOCOM wanted Team Skynet to use this as the base platform.
To understand the problem and the rigors of combat better, Mott arranged a visit to Moffet Field where they completed a mega-cognitive obstacle course — basically a series of puzzles and brain games to solve while also doing rigorous physical activities, wearing body armor, and carrying a rifle. It replicates the confusion, elevated heart rate, and physical rigor of combat without the danger. This helped the team better understand all the noise that someone in combat deals with.
They set about designing a “Call of Duty” style map, on which friendlies are blue and the enemy is red. The commercial drone would identify who was on the ground, categorize them, and beam the data to the soldiers below. The below video demonstrates their MVP.
But what would receive the data? The good news is that the military already fields something that could serve as the platform for this, obviating the need to build a whole new software platform. It is called the Android Tactical Assault Kit (ATAK). It is like Google Maps for the military. Soldiers carry around a Harris radio hooked up to a tablet or phone. When the two are linked, the location of everyone carrying these appears on a map on the tablet — just like in “Call of Duty.” But it is more than that. It is a mobile intranet with communications tools, overlays, and more. The Rangers have been using ATAK since 2013.
This was the perfect solution for Team Skynet. It didn’t require anyone to carry more gear and increase the already heavy load soldiers lug around. ATAK already exists within the Army acquisition system. The team shared its idea with sponsors and other stakeholders and received uniformly positive feedback.
All Skynet needed to build was a way for the drone to recognize enemy personnel on the ground. This is where the next problem came in: image detection. How can a drone look down at the ground and tell whether or not someone is a bad guy and exactly where on the ground he is?
Team Skynet reached out to a professor at the University of Durham working on a similar problem for the U.K. Ministry of Defence to learn more. Much of this part of the story is very technical, but the solution could be found in a combination of image detection and Euclidean geometry. They had to modify the rate at which the Phantom drone pumped out images and build algorithms to detect people images and then determine if identified personnel are hostile. To train the algorithms, they needed to feed in thousands of tagged computer images of people of all different shapes and sizes. And these images needed to be shot from above. Finding these images was a challenge, but DARPA and the Air Force Research Laboratory that build ATAK were helpful in this regard. DARPA also provided some key insights on coding for ATAK and geolocation through Euclidean geometry. Normal Stanford students asking DARPA for help probably will not receive a reply, but the infrastructure built into this course ensured their calls would get answered.
By the seventh week, Team Skynet had a demo of what the product would look like and tested it in a park near Stanford, with the Phantom flying overhead. They shared the video with SOCOM, their sponsors (which you can watch above). SOCOM was so impressed that they asked for a full demo over the summer. Team Skynet went to work building a full prototype. They have refined the user interface and explored partnerships. Mott returns to the Army next year, but the team members are sticking with their product.
Why Hacking for Defense Matters
At first glance, you might be wondering what the real value added here is, particularly if you haven’t had the pleasure of serving in the bowels of the U.S. defense bureaucracy. Did the teams always solve the problems they were presented with? No. So why is this important?
Fixing the Real Problems
While conducting their research and meeting with everyone they could find that touched the problem they were addressing, teams often found deeper problems at work and pivoted to find the deeper solutions. Jackie Space told me, “It’s not about building prototypes. The military has plenty of prototypes. It was about getting a deeper understanding of the problem.”H4D Teaching Team (left to right): Jackie Space, Tom Byers, Pete Newell, Joe Felter, and Steve Blank
To entrepreneurs, this notion of failing fast and pivoting is a normal way of doing business. But to companies traditionally working on building defense systems, such a way of approaching a problem is practically unheard of beyond a precious few islands of innovation. The traditional way is slow. Yes, it provides solutions, but they are expensive, take a great deal of time, and often do not really address the underlying problem.
Having a group of motivated outsiders spend just ten weeks talking to various stakeholders that touch upon the same problem has helped everyone, including the original sponsors of the problem, better understand the puzzle they were trying to solve. Bureaucracies are often so divided internally that one office doesn’t know what the other office is doing, even though they might be grappling with the same issues. This doesn’t mean that Defense Department and defense industry workers are lazy, ill-informed, or bad at what they do. That couldn’t be further from the truth. One of the things that was constantly reinforced was that these teams could not have been successful without the mentorship, assistance, and buy-in of these public and private organizations. But these workers are in a system that, as it currently exists, is not setting them up to succeed on many key technology issues. Jared Dunnmon, a student in the class, told me over email:
[T]here is a legitimate appetite within the armed forces for engaging productive technological change. While current organizational structures are not necessarily optimized for rapid uptake of new technology (and occasionally for very good reason), conversations with many operators, analysts, and strategic decision makers over the past three months have emphasized just how important it is that technology and national security remain closely tied together.
This sort of exercise helps government bureaucracies better know their counterparts, and it will foster deeper future cooperation. Sometimes, defense personnel and industry workers are so caught up in their perspective of their problem that they don’t take the time (often because they don’t feel they have the time) to talk to folks looking at the same problem from another perch. The H4D model has proven itself as a forcing mechanism solve this stove-piping of problem-solving.
People
One class at Stanford cannot change the way the Department of Defense develops new technology and equipment. That is why this class is just the beginning — H4D is spreading across the country. In this academic year, it will be taught at a total of 12 universities, including UC San Diego, the University of Pittsburgh, University of Southern California, Georgia Tech, and Georgetown University. Fifteen more schools are in the process of adopting the class. That means nearly a hundred problems will soon be addressed every semester by talented and driven teams of young people. And later this month, Stanford will launch a sister course, Hacking for Diplomacy, sponsored by the State Department to address foreign policy challenges.
One of the key problems our defense leaders grapple with is recruiting and keeping talented people with niche (often tech-related) skills. The Department of Defense’s moribund personnel system makes this hard, to put it lightly. Secretary of Defense Ash Carter and former Undersecretary of Defense for Personnel and Readiness Brad Carson tried to address these issues through their Force of the Future reforms, but they would be the first to admit this isn’t enough. So why does H4D matter for this issue? Most of the H4D teams at Stanford will keep working on national security projects now that the class is over. These are talented and intelligent people with valuable technical skillsets who otherwise might have never worked on defense issues. As Jackie Space explained to me, “It was an experiment to see if we could get students interested in national security. They dedicated way more hours to this course than any other course.”
Sam Gussman of Team Skynet told me over email that, as he faced the end of his college education, he worried he might be destined to just do web development for the rest of his life. In his words, “H4D took me out of that Silicon Valley tunnel-vision, and showed me that both public service and working with the Department of Defense are viable career-paths.”
Fixing the System from the Outside with the Inside
Blank’s revolutionary vision may or may not one day prevail. But if H4D and its animating ethos spreads, there will be major positive evolutionary change. By exposing the stovepipes in the acquisition system and then talking to all of them at once, the students formed networks beyond what government sponsors are tapping into. These networks within and between the public and private sectors are enduring the end of these H4D projects, which will have profound effects well beyond the accomplishments of the H4D teams. And hopefully, it will get the system itself to think differently about how to solve our most pressing defense problems. I suspect you’ll be hearing a lot more about H4D in the future. There’s even a book on it by Blank, Felter, and Newell, coming out next year.
No comments:
Post a Comment