(Ed. Note: The “Innovation Fact of the Week” appears as a regular feature in each edition of ITIF’s weekly email newsletter. Sign up today.)
For the military, capabilities in the field matter most, not R&D. So, when it comes to artificial intelligence, the Defense Department has been moving quickly by standing up a special team, like a startup enterprise. Its first pilot project, “Project Maven,” began as an intelligence application. Now the push is on to apply it in other areas. Rob and Jackie sat down with retired Lt. Gen. Jack Shanahan, the first director of the Defense Department’s Joint Artificial Intelligence Center (JAIC), to discuss how AI is being used in the defense world and the implications for the broader AI ecosystem.
- Daniel Castro, Michael McLaughlin, “Who Is Winning the AI Race: China, the EU, or the United States? — 2021 Update” (Center for Data Innovation, 2021).
- Rob Atkinson, Jackie Whisman, “Podcast: Innovating in the Defense Sector to Remain Competitive With China, Featuring Michael Brown” (ITIF, 2021).
- Event, “How to Deepen Transatlantic Cooperation in AI for Defense” (CDI, 2021).
- Rob Atkinson, “Emerging Defense Technologies Need Funding to Cross ‘The Valley of Death’” (RealClear Defense, 2020).
- ITIF, “ITIF Technology Explainer: What Is Artificial Intelligence?” (ITIF, 2018).
Rob Atkinson: Welcome to Innovation Files. I’m Rob Atkinson, founder and President of the Information Technology and Innovation Foundation. And we’re a D.C. based think tank that works on technology policy.
Jackie Whisman: And I’m Jackie Whisman. I handle outreach at ITIF, which I’m proud to say is the world’s top-ranked think tank for science and technology policy.
Rob Atkinson: And this podcast is about the kinds of issues we cover at ITIF from the broad economics of innovation to specific policy and regulatory questions about new technologies. Today, we’re going to talk about defense innovation and particularly about AI, Artificial Intelligence as it’s used in defense and the effect of that on the broader AI ecosystem.
Jackie Whisman: Our guest is Lieutenant General, Jack Shanahan, who retired in 2020 after a 36 year military career. His final assignment was as the inaugural director of the US Department of Defense Joint Artificial Intelligence Center. Welcome general. We’re excited to have you here.
Jack Shanahan: Well, thank you, Jackie and Rob. Thanks to both of you. Thanks for what you do for ITIF, but also I’m grateful for the opportunity to come and spend a little bit of time this afternoon, talking about AI.
Rob Atkinson: That’s great. One quick thing, always love when, I lecture over at National Defense University, The Eisenhower School, and I always love how the Defense Department has great acronyms. So the JAIC. They call it The JAIC, right?
Jack Shanahan: Well, that’s because I called it, The JAIC. Others wanted to say Jaike, or something but I was known growing up as couple of different names, Jack, of course, but Jake, what my brothers used to call me. So I thought it was fitting that JAIC would become known as The JAIC. And I think it’s a pretty good catchy name.
Rob Atkinson: Yeah. Absolutely.
Jackie Whisman: It stuck. And I read maybe 1/50 of your bio. I skipped over your 36 year career in the air force, but maybe you could kick us off by explaining how you came to be the lead for DOD AI policy.
Jack Shanahan: Well, to me, it’s an interesting story that I couldn’t have replicated. If you roll the dice a hundred times one time, one time only, it came up to the point where I closed out my 36 year career as the director of this new organization called The Joint AI Center, The JAIC. My background is very interesting, it’s a dog’s breakfast. So I started off in aviation, for those in your audience that might remember the F-4, most people don’t realize there was a thing called the F-4 fighter in the United States Air Force. So I started in the F-4 and then went to the F-15E Strike Eagle as a transition process. But after a while, I got involved in a lot of different things beyond fighter aviation. Things like command and control, intelligence surveillance and reconnaissance. I spent time in the policy staff in the Pentagon, in the Office of the Secretary of Defense.
I’ve commanded six different times at six different levels. And then finally ended up in 2015 under Secretary of Defense for Intelligence in the Pentagon. And we were presented with a very, very hard problem that we could not solve in traditional means. And the problem was, the information that was coming off of platforms and centers of what we call ISR, Intelligence Surveillance Reconnaissance. Like anybody in industry, like anybody in any organization dealing with volumes of information that are almost incomprehensible today, we needed a different way of solving the problem of analysts staring at screens for 12 hours at a time, trying to interpret what was on that screen. It’s mind-numbing work, and it was actually not what you want to use humans for. There are ways to that machines should be able to do that. So I gave that problem to a Marine Corp Colonel that worked for me at the time.
In fact, he’s still there, and we could not find anything in the Department of Defense available for immediate fielding. The department has done world class work in research and development on AI. What they didn’t have available to us at the time was something that we could put into fielding very rapidly. When I say rapidly, I’m talking six months to a year. For the Department of Defense, that’s extraordinarily fast, not fast by commercial industry terms, but still very fast. So we went out to commercial industry and of course, they told us, “This is what’s available. We can do this. It’s called artificial intelligence.” It’s called computer vision specifically for the problem that we were handing to them. And to make a long story much, much shorter, we stood up a project. We demonstrated sort of the art of the possible to the Deputy Secretary of Defense, a gentleman named Bob Work, who’s still very active in this space. He’s gone from the department, but writes, and talks a lot about AI. And we stood up a thing called Project Maven, also known as the Algorithmic Warfare Cross Functional Team.
Jackie Whisman: I like Project Maven.
Jack Shanahan: Project Maven’s a lot easier to say. But if you think of about Algorithm of Warfare, the reason Secretary Work likes that term, and we like that term, is because we envision a future of algorithms against algorithms. And we might come back to that in terms of where we see the future going. So we looked at Project Maven as a pilot project. Could we show enough success that then the department could move beyond. This was all for the intelligence enterprise, this was nothing beyond intelligence. At the end of a year and a half, there was a push, a very strong push to go well beyond intelligence, to all the other parts of the department of defense.
And again, I always come back to this, to focus on fielding capabilities, not research and development. And so, there was a decision and to stand up an organization, we came up with the name, The joint AI Center, or The JAIC. Congress pushed us very strongly in this direction, the Defense Innovation Board pushed us very strongly that we needed to move very quickly. We weren’t moving fast enough to look at what was happening in commercial industry around us, we were way, way behind. And then they sought a person to lead this organization. And sad to say a little bit that there really was no senior official. They were looking for a military person because they were going to work very closely with the general office and admirals in the other services. And because I had had a little bit of experience, two years of experience with Project Maven, they asked me if I would be the first director of the JAIC. And after a long, hard reflective process, I said, I would do it and served in that position until I retired in the summer of last year.
Rob Atkinson: That’s really, really interesting, Jack. I didn’t really know all of those pieces. Let me actually jump back to Maven, because I think a lot of people who follow tech policy in Washington when they hear man and they think Google. Because there was that little kerfuffle, but my understanding is Google has sort of resolved that as have a lot other companies are now happy to work with the national security establishment including DOD on developing better tools for our country. Can you just say a little bit about how that all worked out?
Jack Shanahan: Yeah, I will remain so proud of Maven for the rest of my life. And we made a lot of mistakes, as you would expect. And I say only a little tongue in cheek, but it actually is true to say I was the CEO of two AI startups in the Department of Defense, Project Maven and the JAIC. Like any startup, we made a lot of mistakes, but one thing I know that we really did well is the Marine Corps Colonel that we put in charge was a classic disruptor, who could not take no for an answer and would never take no for an answer. And he pulled off things that I don’t think anybody else could have done. And what do I mean by that? Well, you had to fight a bureaucracy that was not used to these concepts of agile development. And just the idea of innovation and an obsession on customer experience, user experience, user interface. How quickly can you get through acquisition and contracting?
So we were doing all that at the same time, we were just trying to figure out what is this thing called AI and computer vision? How quickly can we give it to people that are engaged in combat operations? So within four months standing up and getting the official seal of approval from the Secretary of Defense, first of all, we started with no money, which is a classic startup problem. So we had to work our way around, get several tens of millions of dollars. Congress wanted us to move quickly, so we had bipartisan support on the Hill.
And then with four months, we had four startup companies on contract from Silicon Valley, and then another two months behind that we had Google. Nobody had ever expected Project Maven to have Google as a partner. It was a stunning development that had some backlash, as you referred to a little bit earlier. And that drama unfolded and it ended after a year and a half or so of a partnership with them. But the idea of how fast can you really move? How could you change what do we call this sclerotic ways of doing business in the Department of Defense? There were lots of small innovation organizations beginning to sprout up. One of the most important was the Defense Innovation Unit also known as DIU in Silicon valley.
Rob Atkinson: Yeah, we’ve had Mike Brown on the podcast before.
Jack Shanahan: Yeah. So Mike, is the current, that was started by Ash Carter when he was Secretary of Defense. And so, the encouraging thing was there were more and more of these innovative organizations. But what none of us would solve, unless you start bringing an organization like The JAIC together is how do you get to scale across the entire Department of Defense? So we did what I... We delivered our first model or algorithm to combat operations within six months to standing up Project Maven, which was unheard of. Now, was it a relatively crude and brittle algorithm? It was. It got better and better and better. The users started getting more comfortable with it, they began to build some trust in the system. And so that process has just continued to unfold. But the big thing was, just showing it could be done. And based those lessons then standing up The JAIC and taking a very similar approach because of my experiences, I took all the bad things I did, we learned the hard way, tried to fix those when we stood up The JAIC.
Rob Atkinson: That’s interesting. I want to ask sort of a couple questions and then, and turn over to Jackie. But one is, at least with AI, as it is today, ML or machine learning algorithm, you need a lot of data. And it’s not always clear to me, how much of that you can get in a military? There’s a lot of one-offs, there’s a lot of uniques in a military situation. So how did you deal with that problem or that challenge of being able to get enough data to have a machine learning system learn?
Jack Shanahan: Yeah, unfortunately you have to relearn the lessons that every other organization who has done AI for the first time had to learn. It’s data, compute, and algorithms. Think of those three big categories. Data turns out to be as complex and difficult as the other two categories, in some ways even more so. It’s not an insolvable problem, it’s just hard to go at this. As you were just suggesting, the data is all over the place in the Department of Defense. So we had to build a data management pipeline. How do you go get to data? I mean, this is no exaggeration whatsoever. We had people driving pickup trucks or round two sites to pick up tapes with video on it, to bring it back to then begin to build the algorithms against it, which is a crazy, crazy way of trying to do business.
Rob Atkinson: Wow.
Jack Shanahan: Now, the data was there. The Department of Defense is awash in data. Well, is it the right data? Is it dirty data? Will it help you, or will it actually hinder you? So we had to build an entire data management process, which actually has come a long way and more and more companies are focusing just on that aspect, because it is so instrumental. I hope two years from now, it’s less of a problem, because companies are coming up with different and better and more efficient and effective ways of going at the data problem.
But that data problem took a lot of time to work through. As you can imagine, I’ll just say a very quick example. If a drone is flying a 12 hour mission, sometimes 10 hours of that is nothing but flying back and forth, or looking at clouds and, and nothing happening on the ground. All of had to be reviewed manually at first, eventually, you sort of get to an automated process. So there was so much more than the algorithm that Maven helped establish procedures for, that then has helped the rest of the Department of Defense. But that was a big one and it continues to be a real challenge.
Rob Atkinson: Yeah, that’s really interesting. We could go a lot deeper on that, but I want to ask you, to me, the other challenge, and I know when I’ve talked to defense contractors, they’ve raised the issue. One of the advantages that Google has, or a Facebook, they can hire anybody from really around the world, as long as they can get a visa. And obviously the DOD, and national security spaces were limited. How did you cope with just skill development and skill acquisition within what you were doing?
Jack Shanahan: Yeah. Well, so there’s a DOD part of this, and there’s an industry part. The DOD part has a long way to go. So one of the things we took on in The JAIC was how do we build an education and training strategy for the entire department that Congress has been again, very helpful in leading us to a solution, which is we have to get an entire workforce to understand this in a much better and different way than they’ve understood it in the past. So accepting that DOD was going to take a long time to get there, we had no choice but to turn to commercial industry. First of all, we weren’t interested in reinventing the wheel. I hope as you would appreciate both of you, our philosophy from the beginning of Project Maven, from one was commercial first.
Don’t reinvent what already exists in commercial industry. I took that philosophy in with me to The JAIC. Now, some people say, “Well, there are a lot of really talented people in the department.” There are, but you cannot compete with a Google, a Microsoft, or an Amazon, and the number of PhDs machine learning they have. You just can’t get there. Now, I will also to say, we did play a very deliberate strategy of going after startups because they were thirsty, they were talented and they needed work. And to a person, they were not reluctant to work with the Department of Defense. So they just had this energy level that they wanted to help us and do it very quickly.
And then it’s a little bit of a signal to the big defense beltway, contractors, the traditional defense industrial base, “Hey, wait a minute the department’s going out and spending hundreds of millions of dollars, and it’s not coming to us. What’s going on here?” It’s almost an entire episode in itself of sort of how the defense innovation base looked around and saw this national security innovation base beginning to blossom and a lot of money going into it. So, you got into a little bit of, “Hmm, maybe we need to invest some resources into that same sort of machine learning or computer vision, natural language processing.” So we went out to those companies deliberately because their talent level was extraordinary. And whatever is written about the Google Project Maven episode, I’ll tell you what, those software engineers working on the project were world class, absolutely world class. They were not the ones protesting, that was a different part of the company. And I was very a proud of the work that we we did together.
Rob Atkinson: That’s great. Just a quick a side, I’ve mentioned before, maybe in an episode. My son, when he graduated from college, he got a computer science degree in natural language processing. And we live in D.C. area and I said to him, I said, “Hey, it’d be really great if you come here and work here,” I’m happened to mention this to a defense contractor I was talking to. “Oh, have him apply. Really, we need people like that.” He’s out in Silicon Valley working for a VC based company because it’s so exciting.
Jack Shanahan: Yeah. And that’s going to happen a lot. The one thing that we could always try to entice people with, I mean, patriotism, okay that may or may not work, money may or may not work, but people want to work on really hard challenges. And the military has some wicked hard problems, as we would say in New England. And we have data that doesn’t look like any part of commercial data. I mean, it’s one thing to label objects for sale on Amazon marketplace, or to make movie recommendations on Netflix. It’s an entire another to look at how do we actually fight, or help our back offices get more effective and efficient business processes in place or medical. We spend a lot of time on the medical field in healthcare, which I think is one of the biggest areas of boom in the next decade, is the medical part of this.
Jackie Whisman: It seems like one of the many difficulties in managing a department like this, especially a new wing of it is how to define and measure US leadership in AI for defense. So for example, the US military may be able to better leverage AI and cyber operations, but the Chinese military could have better advances in hypersonic weapon systems that can operate autonomously. So, when we’re talking about overall leadership in AI for defense, how is that being measured? Or how did you measure it when you were running the show?
Jack Shanahan: Yeah, it’s a good question. And I will, again, sort of make this binary. You’ve got commercial world AI and then military operations. The commercial AI race between various countries or just their state of progress is well covered. And there’s a lot of think tanks writing about this. And I think good writers that Kai-Fu Lee have written about well, China has certain advantages in implementation. They’re very practical about how quickly they will put something in the field, where we might spend a little bit more time doing research and development and testing and evaluation of it. They have all sorts of data, but is it the right data? So, I’m not sure data is inherently advantage, even though everybody says, “Well, they’ve gone to an entire digital process.” That may or may not be an advantage.
We have the best education system still, I believe. Research and development in the world on AI, with a lot of people catching up very quickly. So, there’s no one area think anybody would say the United States or China has such a strong advantage that it will change the game for the next decade. It’s pretty close in all areas. I will say, and I’m hardly the first person to say this, China is catching up very, very quickly. China innovation is a real, real thing, and they’re getting very good at it. Now, on the military side, it’s a little bit of a black hole, and this is what I had to deal with because our intelligence community didn’t know what to collect against, why they needed to collect against it. It’s one thing to tell me that China has terrific computer vision to catch jaywalkers in the straight streets of Beijing.
What I really want to know is, how is the PLA using it for military purposes? That is a much more challenging problem. I stood up a part of The JAIC just to focus on that part of the problem, working with the entire intelligence community across the entire United States national security enterprise, because it’s not very well understood. And I would think China probably says the same thing about us. We make assumptions about how fast the other are moving. I don’t want to get back into sort of the missile gap myth that we experienced in the 1960s. I’m more interested in the reality of practically where is each side. And when we know sort of strengths and weaknesses, we know much better on what to focus on. So very good understanding on the commercial industry side, still a long way to go on the military side.
Rob Atkinson: Yeah. That’s interesting. Daniel Castro on our team led a project, and now we’ve done it actually two years, where we looked at about 15 or 17 objective, very quantitative variables that were available for the EU, the US and China on AI, like university quality. Again, you pointed out, we lead on that, publications, venture capital, new startups, all this, and the US, at least looking at these variables leads China, where China’s ahead of Europe, but China appears to be catching up. Their rate of acceleration is great, so suggests that policy makers really need to pay attention and make sure that they’re doing the right thing.
I did want to ask you this question. One thing that we’ve talked about earlier, one time. I get so frustrated when I hear people talk about AI. And it’s almost like it’s some sort of magical thing that a wizard can just do to make anything happen. And then the problem, when you think of AI, that way, in my opinion, it leads you off in dystopian nightmares pretty quick. “AI is going to eliminate everybody, there’re going to be no more think tanks because you’ll have an AI running ITIF” And I know Jackie might like that, but I’m not sure I would.
Jackie Whisman: You’re not a hologram Rob? I haven’t seen you in a while.
Rob Atkinson: According to Elon Musk, the odds of us all living in a matrix like reality are quite high. But my point is, AI is really just computer software code, it’s cool. It’s a lot better what we had before, but it has limits. Love to just hear your thoughts on that, Jack?
Jack Shanahan: Yeah. So, the dirty little secret is I had a can of AI’s spray in my desk drawer, and I’d pull it out and sprayed that around, instant results everywhere. Now, the people that... There is a problem with AI hype today. There’s no doubt about it. I’m not worried that we’re being an AI winter, but to steal a phrase, I think from a Chinese resource, who said, “But there may be an AI fall coming up if we’re not careful.” Because people talk about it in very unrealistic ways. The way I look at it right now is, for the foreseeable future. First of all, I take the discussion of artificial general intelligence off the table. And I give that to the researchers who look 50 to a hundred years down the road.
For the foreseeable future, for the world that I did, I inhabited and the people behind me are dealing with every day, it’s what can humans and machines do together to make it all work better? And I’ve used the two words twice in this session already is effective and efficient. What can we do to just make better and faster decisions and get through data faster? The idea of AI, even the idea of autonomous vehicles in the edge case conditions is a long, long way away.
I think Elon Musk woke up to that reality recently, and even admitted as much, this is a very hard problem to solve. So I think we should spend a lot of less time focusing on those dystopian parts and what can I use AI for to get that human more time, to think about context, to think about a particular problem he or she is dealing with in a way that the machine does the things that a machine does best, which is get through lots of data very quickly, finds signal in noise, and help the human do what the human does best, which is contextual understanding or rich, contextual understanding.
I think if we focus on that, the idea of a human machine in a partnership of some time, that we will begin to see more and more benefits over the next couple of years. I mean, there are some things that are happening today that we call them AI, but they’re more sort of what I would call, smart systems or expert systems, a little bit more rules based, that are just making businesses do back-office operations much faster, much better than they were doing them in the past. And the medical piece, some big, big breakthroughs. I think pharmaceuticals will see very big breakthroughs. Some parts of sort of the Maven intelligence work will show rapid progress over the next few years, but it will not be replacing humans, it will be complimenting humans. And I think that’s what I would always want to emphasize in this conversation.
Rob Atkinson: No, that’s great. That’s great.
Jackie Whisman: Well, maybe we should close by... One thing that I wanted to ask was how important is it to have a strong AI ecosystem in the US for national security? But also, how much does US military investment in AI help overall AI innovation?
Jack Shanahan: Yeah, I think this is such an important foundational question. We spent a lot of time in Maven, and then again at The JAIC thinking about what we could do to help the AI ecosystem because it truly is an ecosystem. When you look at the components of data, compute and algorithms, those are different components with different companies working on all of that. There were times when we had negotiations ongoing with a startup company who was incentivized to take money from China, if we weren’t going to come through with sort of cash in the next quarter, because they needed to put money on the books for the quarter. Now, fortunately because of the aggressive people we had running Project Maven, they were able to get through the acquisition and contracting process, fast enough to put these companies on contract with us, and they turned away from the China money.
That’s a micro example of a macro problem that the country is facing. And I know Rob and Jackie, you’ve both talked and written about this idea of industrial policy. I don’t want to go too far down that path. But I think there is unquestionably an element of that, which is maybe nothing more than saying the country is going to make some strategic choices about the areas of technology called emerging and disruptive technologies. Not trying to say one company’s going to survive over another or anything like that. As you say, Rob, “This is not the end, right? It’s a means.” Now, I think there’s a strong case to be made for a little bit more interventionist policies to help the entire ecosystem get there faster. Because without it, I think the competition is brutal in commercial industry, and China’s going to catch up very quickly and other countries will be there ready to take the place if China doesn’t fulfill somebody’s needs.
So I think we really have to focus on the entire ecosystem here. The good news is, it’s striving in so many ways, and I would like to see it blossom even more, but that means there has to be more investment into these companies from places like the Department of Defense or other parts of the government. There is a lot of work that has to be done on that. So, the good news is I think this administration’s put putting some very deliberative processes in place to start really going after that idea of, what does an industrial policy look like? What should we be doing to really encourage and foster innovation in a way that it’s no longer a hardware industrial age world, it is truly a information age, software driven world? And we have to look at things differently than we did a decade ago.
Rob Atkinson: Yeah. I mean, if there’s one lesson from studying evolution and Jack, you know more than I do on this, obviously, but one lesson from studying the evolution of war and changes in leadership it’s that, technology change is a core component of that. And countries that don’t catch up with that. I mean, you being from the military, was it Billy Bishop in the air force who realized you could use an airplane to sink a ship?
Jack Shanahan: Mitchell, Billy Mitchell.
Rob Atkinson: Billy Mitchell. Yeah. Billy Mitchell. And-
Jack Shanahan: He got court martialed for it by the way.
Rob Atkinson: He got court martialed for it, but he was right. And it’s the ability, and certainly Ash Carter, and others, and Bob Work, others saw in you, saw the importance of we are not going to be able to be a viable or strong, capable military and defense power if we don’t move into this new area robustly.
Jack Shanahan: Well. And to drive that point home even more so is, the technology we’re talking about with AI is a commercial technology first and foremost, that we are now trying to adapt to the Department of Defense. Unlike almost every other technology department has sort of produced, invented, designed, fielded in the last a hundred years, this is different. It’s mostly coming from commercial industry. And we have to think carefully about how do we protect those industries in a way that we haven’t had that conversation when it comes to IT, or emerging tech, the way that we used to about submarine industries or aircraft production in Detroit and World War II or Willow Run. I mean, these are new questions that we really have to go after, or we’re going to lose.
Rob Atkinson: Yeah. No, that’s an excellent point. We weren’t worried about offshore, the sea wolf, but we certainly could lose commercial capabilities in here that is critical for us. So Jack, we could just go on, I could go on for a long, long time, and I know Jackie probably could too, but in the sake of keeping on time, we have to stop here. So I really want to thank you so much for being with us today.
Jack Shanahan: Well, thank you. Both of you for what you you’re doing with ITIF. I know you’re driving a lot of big discussions in the administration, in commercial tech industry and even in academia. So I’m grateful for a chance to come in and spend a little bit of time. But more importantly, I’m proud of what you’re doing with ITIF.
Rob Atkinson: Well, thank you so much.
Jackie Whisman: And that’s it for this week. If you liked it, please be sure to rate us and subscribe. Feel free to email show ideas, or questions to [email protected]. You can find the show notes and sign up for our weekly email newsletter on our website, itif.org and follow us on Twitter, Facebook and LinkedIn @ITIFdc.
Rob Atkinson: And we have more episodes and great guests lined up. New episodes will drop every other Monday. So we hope you’ll continue to tune in.
Jackie Whisman: Talk to you soon.