无人机战争：作战新规则已经到来

The industrial capacity of China is fearsome.

Being able to deploy highly autonomous, AI-driven drones at scale is still a domain that we can win in.

I think the technologies that matter most to the future of war are right there in front of us. I think a modern conflict becomes basically like a software-writing fight. It will be the pace of deployment that is the make-or-break for militaries around the world.

The game theory here is just as simple and obvious as it can be.

Unless we find a way for industry and government to work together, we will find ourselves in a very tough situation.

We're here to chat about drones, autonomy, and great-power conflict. Both of you have started incredibly successful and innovative drone-focused companies. At the same time, both of you started when the industry was truly nascent. It was seen as a field for hobbyists rather than something that could actually shape the future of warfare. What led you to be interested in this field and to spend so much of your time, blood, sweat, and tears on this now-incredibly-important industry?

I had started and sold the company to Qualcomm, and I've always been somebody who has had an intense passion to compete, fight, and win at whatever I was doing. In my last year at Qualcomm—I was there for about 4 years—I didn't have the fire in the belly that I had had throughout my life.

So I started thinking about what it was that would really motivate me, not for the next 5 years but for the next 50. I decided that if I could find the intersection of 3 things—a noble mission, the chance to work with extraordinary people, and a chance to define the possible—I would have that fire in the belly.

My brother went on to become a Navy SEAL, so it was a totally different track in life. He was getting ready to go to business school, and I encouraged him to think about what he wanted to do in life. He came to me with the idea of bringing the best of what was going on in the autonomous-driving sector to the mission of protecting service members and civilians.

He felt like if somebody would do that, it would have brought home a lot of his friends safely to their families, and he felt like it would be a pillar for the future of American military dominance. I thought it was an extraordinary mission. I told you earlier, I thought it was a stupid business, so I wished him luck and suggested that he come up with a better business to make a mission impact.

But SEALs are very persistent people, and my brother is no exception. Long story short, I started to spend more time with him, learning about the challenges, and I was just shocked by the scope of the problem and how little was being done about it. I've been proud and humbled every day for the last 10 years to have an opportunity to contribute to a mission that I think is so important.

Incredible. I'm glad SEALs are persistent in general, and I'm especially glad that your brother, the SEAL, dragged you along in this important mission. What about you?

We come at this from very different places. First, I think that Shield AI deserves enormous credit for, 10-plus years ago, recognizing the need and the opportunity for what we now think of as defense tech at a time when that really did not exist and people thought they were crazy for even trying. Candidly, I don't think we get that same kind of credit.

I grew up flying radio-controlled airplanes. I've basically been doing drone stuff my whole life. That led me to be a grad student at MIT in the late 2000s and early 2010s, when you could basically take radio-controlled airplanes, put computers and sensors on them, and write software to get them to do smart stuff.

I really became obsessed with trying to build AI systems that could fly better than people could. I wasn't thinking so much about applications at the time. But in 2013 or 2014, my lab mate and I started to look out and see that there were interesting things starting to happen with small, lightweight quadcopters, and we felt like the applications and implications of the technology could be enormous.

Needing to have an expert pilot there flying it was just a fundamental restriction. The big bet that we made when we started Skydio was that AI and autonomy built into a small, lightweight quadcopter was going to be very powerful for a wide range of industries. Government and enterprise applications were always part of the vision, but we explicitly decided to start with a consumer product because we thought something lightweight, integrated, and easy to use would be a really good platform for this other stuff.

For me personally, when I was in grad school, I was wrestling with, “I love this technology, and I wonder if I want to keep working on it. Am I going to have to go work for a defense contractor?” I deeply believe in the mission of the U.S. military, but the idea of working at a traditional defense contractor was very unappealing to me. So we started with a consumer product.

It's reflective of the trajectory of the space. Very quickly, in the 2018–2019 time frame, I think, to the credit of the U.S. military, they realized that these small, lightweight civilian quadcopters had enormous value on the battlefield. The technology from consumer to enterprise to military is so tight that, in the span of a year, we basically won our program of record, the Army's Short-Range Reconnaissance program. It was officially announced in 2021, but it really started in 2018 or 2019. That was the beginning for us of expanding to serve a much broader set of markets beyond the consumer space.

Adam was also part of my early story. I don't know if you know this, Adam, but I read all of your papers from MIT when, in 2015, we were working on autonomy technologies. I was like, “Who knows what they're doing?” Then there was Ryan, and he had all these great papers. We've learned a lot since then, but it was the foundation.

Yeah. A lot of the ideas that we use at Skydio came from the research community. Some of it was research that we had done, and some of it was research that other folks had done.

Incredible. One of the things that I think has most shocked many and fascinated military strategists is just how drones have reshaped the nature of warfare in the past decade, particularly in the past 3 years. There's been huge proliferation, right? Far more mass is being brought to the battlefield via drones.

It's enabling much more distributed, decentralized, and lethal force structures. Anybody running around in a truck can pop out with a drone that might go 1,000 nautical miles, or a collection of drones that might go 1,000 nautical miles, and hit who knows what far off into the distance. There used to be much more concentration of forces and dependence on large, exquisite assets to deliver capabilities.

It's been a complete transformation, and the world has seen a lot of that unfold in the conflict in Ukraine. I think one of the major questions is how the United States and its allies can adopt those lessons. The things that we're doing have a lot of merit. It has been a force structure that has dominated the last several decades, but our adversaries have spent a long time thinking about how to counter what we're doing.

I think it's going to be important for us to think about how we can embrace 100 times more systems to empower our service members to be lethal and effective and to come home safely to their families. I think drones, in many ways, are the future of war.

Absolutely.

Adam, one of the things that particularly shocks those observing the Ukraine conflict—referring to, say, a commercial quadcopter like a Skydio—is that, in some instances, they'll go and take out an exquisite system like a tank. Two or three of those packed with sufficient munitions could take out a large armored vehicle. How have you seen that shift impact how the military looks at dual-use commercial drones?

To be honest, I think that's a question that the U.S. military has not fully digested yet. One of the realities of drones is that they create this massive asymmetry: A system that costs a few thousand dollars can take out a system that costs a few million dollars. I don't think we've fully grappled with that yet, to be honest.

We're seeing evidence of this in Ukraine. The Ukrainians, largely out of necessity and through incredibly impressive ingenuity, have a broad array of all different kinds of ground systems—whatever you want to call them, ground robots or drones—along with air drones and sea drones. They're building those at an incredible rate, iterating very quickly, and being very scrappy.

The U.S. military may not be subject to the same kind of constraints that the Ukrainians are, but I think that we need to understand that the possibility for that asymmetry is real. The possibility of building very low-cost systems that are very capable and capable of delivering strikes or maintaining surveillance is very real, and our adversaries are likely to take advantage of it.

I think that's a journey that we still need to go on to some extent. There's still quite a bit of inertia and momentum in our military toward larger, more traditional, exquisite systems. I think that's starting to pivot, but I think that's a real question for us for the future.

There's a phrase often used in the military context: quantity has a quality all its own. There is an incredible disparity between what our chief adversary, the People's Republic of China, can produce in a month for drones, whether they're formal military-style drones or dual-use commercial drones. At that scale, I think a lot of people are wondering what that means if we're trying to deter a future conflict—or, God forbid, if there were a conflict in the Taiwan Strait or elsewhere. Adam, would you mind talking about the differences in scale, roughly speaking, between the production capabilities of both countries, and why that is particularly important from a regulatory perspective?

I mentioned that I grew up flying radio-controlled airplanes. Basically, all radio-controlled airplanes were made in China in the 1990s and 2000s, and nobody thought too hard about that. They still are, right? They probably still are.

That didn't seem like a national security issue at the time, but if you think about what a drone is, it's basically a combination of radio-controlled-airplane-type stuff—motors and consumer electronics, a lot of the same stuff that goes into a phone. As a country, we basically outsourced manufacturing to China, and I think that was a series of policy decisions and expediency on the part of business. That was a mistake from a military standpoint, because one of the major themes is that the gap between civilian and consumer technology and military technology is closing in many of these domains.

The general manufacturing capacity in China for low-cost, capable compute systems, which are really now becoming robotic systems—not just drones, but other kinds of robots—is substantial. It's not just one single thing; it's a whole ecosystem. It's not about cost either at this point. It's really about technical expertise and built capacity in terms of all the different things it takes to mold and machine, build PCBs—the circuit boards—and place components on them.

I don't think this is something we can solve overnight. The ultimate thing—and TBD on whether this is attainable—I always say, wherever they're building iPhones, they're going to have a really good ecosystem for building drones and other kinds of electronics. The real prize is: can we bring that level of scaled manufacturing back to the U.S.? I don't know if we can, to be honest, but I think it's worth a shot.

From the outset as a company, we've been manufacturing our drones in the U.S. I would say that when we started doing that in 2016, it felt like we were swimming upstream into a fast-flowing river. I would say that we're maybe starting to get some signs of tailwinds, especially with the new administration, which is cause for optimism. But I think this is one that we just can't give up on, because robots are going to become more and more important.

They're going to be using the same kinds of ingredients from consumer electronics and other kinds of relatively low-cost systems. Cars are going in this direction as well. Cars are starting to look more like laptops and phones in terms of the components that are in them. This is a combination of industry and policy, and all of us working together. I will say that I think there's cause for optimism over the last couple of years. There's still a lot of work to do, but I don't think it's an insurmountable hill for us.

I've got a good-news, bad-news take. I think the bad news—you sort of led with it—is that the industrial capacity of China is fearsome. Try as we might, I think that's going to be a very difficult thing to close out in 1 year, and even a decade. I'm an optimist, and I always like to believe that there's a way. But it is a substantial gap, and I think we're approaching, to use a rocket term, a Max Q moment from a national security perspective, where we're undergoing a force transformation.

Huge technology changes are afoot, things like AI are coming into play. We've got an adversary that's become extremely wealthy, with huge industrial capacity, also investing massively in its military capabilities. The question then becomes, through this Max Q moment over the next decade, given our production capabilities and the entire capabilities of this nation, what's the right play?

History has shown—World War I, World War II, Vietnam—that tremendous amounts of mass were brought to the battlefield. In some cases, despite tremendous amounts of mass being brought to sections of the battlefield, it just turned into a grind-it-out war of attrition without a lot of movement on either side. We start to see some of that in the Russia-Ukraine conflict as well: tremendous amounts of mass being brought to the battlefield, but front lines that are extremely difficult to move.

I think a reason for that is there's a difference between mass and effect. The world is big, and targets are relatively small compared to the scale of the world. It turns out that just throwing a bunch of mass downrange can still be pretty hard to hit the things that matter.

Where the United States has dominated over the last couple of decades is software prowess, AI, and autonomy. If we can combine the fantastic work that's going to reindustrialize the United States to build more mass and more capability, we have to do that. But if we can combine it with software and autonomy capabilities, if we can close the OODA loop so that we can push software updates at a moment's notice and make every ounce of charge and every minute of flight time maximally effective, I think that's how we can compete.

There's a second wave that hasn't really broken yet, which is AI and autonomy. The vast majority of what's happening in Ukraine is still basically one-to-one. You've got these FPV pilots who are expert operators flying the thing, or they're flying drones with very limited, pretty simple missions, like, “Go to these coordinates and deliver a strike.” I think we're going to see, over the next decade, another fundamental change: rather than having these drones animated by an operator on the ground or by relatively simple algorithms on board, they become animated by really advanced autonomy, and they can communicate with each other.

The implications of that, I think, are probably going to be even more significant than the first step to just having these things be unmanned. I do think that's an area where, as a country, it plays more to our strengths. You can't forget about the hardware and the manufacturing capacity—you need to be able to do that stuff—but winning on the AI front, I think, is even more important.

So, to ensure we dominate, to ensure we win, Vice President Vance just spoke a few moments ago at the American Dynamism Summit, and he said our goal is to essentially remake the economy, to fix that mistake you guys were referencing earlier: we just offshored all of our manufacturing; we can't make things anymore. There are leading members in Congress committed to that and also committed to defense procurement reform.

If you had an opportunity to sit down with anyone—the leaders in the executive branch, leaders in the legislative branch, or anyone who'd be listening right now, whether those leaders or staffers for them—and you could give 1 or 2 recommendations, if you just fixed 1 or 2 really pivotal points, and those were resolved, we could do so much more, either for drone manufacturing in particular, manufacturing generally, or unleashing AI. What sort of recommendations would you give?

Number 1, I would ask that they continue to do what they've been doing, which is reinforce their belief in investments in the incredible people who sign up to serve. The one thing that I think we've got absolutely right is that we have brilliant people who are brave. They believe in the values of this country, and they just go and do extraordinary things. It's been a privilege of mine to be able to meet them.

We strive to contribute to make sure they can be as effective as possible, to fight, win, deter, come home safely to their families, and provide them the best possible tools. And look, what do I know? The administration has hard jobs, so let me put that disclaimer up front.

I think the technologies that matter most to the future of war, that matter most to the future of this country and all of our allies around the world, are right there in front of us. I think the security challenge of our time is whether or not we can mobilize the bureaucracy to go reach out and pick up what's on the table. Everybody already knows that this is one of the most important capabilities of the future.

That is taking and making real the large-scale deployment and operationalization of autonomy technologies. If you look at the space that we play in, which is small, light quadcopters that weigh a few pounds and are kind of soldier-carried, the Ukrainians today are using these things at the rate of millions per year.

Literally millions per year. It’s the primary method through which they’re delivering strikes and surveilling the battlefield. The U.S. military, I think to their credit, has programs generally pointed in this direction, but those programs are operating at the scale of thousands—single-digit thousands. So we’re literally off by 3 orders of magnitude, I would argue, relative to what evidence suggests the modern battlefield demands. There are similar trends in Israel as well. Israel has rapidly adopted these kinds of systems at substantial scale.

This is an area, to the point about reindustrialization, where military purchasing power makes a massive difference. The consumer civilian quadcopter markets are measured at a scale of single-digit billions, which is, I would argue, comparable to what the military should be spending in this space. They’re not spending anywhere close to that today for this class of system.

I think that’s a pretty obvious lever that has a bunch of benefits from a national security perspective. One, you’re equipping our soldiers with modern, relevant technology. But two, the purchasing power that the military can bring to bear there is significant enough to actually move the needle from an industrial-base standpoint for this class of technology that serves other markets. We all—not just us, but the companies in our space—serve public safety and critical-infrastructure inspection, and there’s huge technology overlap between the products that you use to do that and the products that folks use on the battlefield.

I think if you just went down all the quantities of everything and added a 0 behind all of them, and then, of course, there’s a cost challenge. We have to figure out how to do that.

Yeah. I think there are some things that you can delete, and deleting those things is also painful. It’s maybe not a 0. Sometimes it’s a 0 in front of something—a decimal, the big 0, right? I think that’s the trade-off that needs to be made, right? There’s a trade-off between buying legacy exquisite systems that cost hundreds of millions of dollars, in some cases billions of dollars, and buying 1 or 2 of something or 10,000 or 100,000 or 1 million drones.

If you just look into the future, which of those things is going to be more powerful? Large quantities of AI-driven drones isn’t the only thing that you need, but I think in many situations it’s the right answer.

Well, let’s dive into a little bit more. As you both noted a few moments ago, the U.S. military, starting around the 1980s and 1990s, really went all in on highly expensive, exquisite systems: small numbers. That worked fine for us, but our now-near-peer adversary, China, developed an asymmetric military designed to counter that. So they have a carrier strike missile. For a cost of, say, $100 million, they take out tens of billions, if not more, on our side, plus the horrible lives we lost.

There’s been a move within the DoD to counter the counter with Project Replicator. Let’s presume that, even if we move those decimal points so that all of a sudden the older systems have less procurement and the newer systems have more dollars allotted to procure them, China still has certain advantages—perhaps because they’ve had more time working on it, perhaps because they’re more subsidized.

One of them, people would argue—and I welcome you to say that this is wrong—would be in the area of swarm technologies. You can see the light shows over Shenzhen and other cities for the Chinese New Year, where extraordinary shows with hundreds of thousands of drones broke the Guinness World Records again this year. Do we have that level of sophistication, and why or why not? What can we do to make sure that we not only achieve parity, if we have not, but achieve superiority in that space?

This is an area that we’re quite focused on. Swarm is one of these terms that can mean a lot of different things, and it’s really about what task you’re trying to accomplish. When you see the drone light shows, they’re 100% reliant on GPS. They’re using GPS to figure out where they are and to position themselves precisely. They’re also 100% reliant on a comms link between all the drones all the time.

Both of those technologies are basically irrelevant on the modern battlefield. It’s very easy to jam GPS, and comms are always contested. I think those shows are representative of their ability to build a bunch of drones and get them into the air, which is certainly part of the equation. But from a core technology standpoint, they’re less relevant for the things that I think would be impactful on the battlefield.

We have the great joy of competing against the leading Chinese drone company, DJI, in civilian markets that are unregulated, where customers can buy everything. Our biggest advantage competing against them is AI and autonomy capability. The stuff that we’ve been able to build into our drones is far more advanced than what DJI has been able to do in terms of sensing the environment in real time, responding to it, and automating complex tasks and missions.

The advantage that we have is reflective of our strengths as a country. We were talking about building on academic research. The technology of Skydio is reflective of a lot of smart investments that the government has made over the years. I think being able to deploy highly autonomous, AI-driven drones at scale is still a domain that we can win in. In my view, it’s still up for grabs, and it’s something that we’re quite focused on as a company.

And Ryan, your company has also invested heavily in autonomy.

Yeah. Our investments in autonomy, going back to one of my earlier statements, are predicated on this belief that to make mass effective, it has to be intelligent, right? That’s the difference between grinding out trench situations and being able to assert dominance in a space: if you can find, fix, and finish targets at scale.

We just think that’s fundamentally important. The first 10 years of our journey were defined by striving to make the world’s best AI pilot and climbing the aviation food chain, which culminated in us doing some work on F-16s that has circulated the internet.

When we thought about the next 10 years, the question was: Is the future really about Shield AI building the world’s best AI pilot, or is it about making a contribution to the industrial base? That way, everybody who’s building these systems across America—building incredibly sophisticated machines and striving to do it at larger scale—can now deploy the best possible AI pilot for their vehicles and for the customer’s missions.

We think our best and highest contribution is to enable the industrial base to fast-forward the large-scale deployment of the world’s best AI pilots.

So in Ukraine, you see drones come up, drones come down. Oftentimes they’re DJI drones, in a matter of minutes or seconds, many times failing in the mission. How have you thought through how your drones can operate in a battlefield where there is a high degree of electronic warfare? You don’t have access to GPS, and you don’t have access to reliable signals to the controller and operators.

When we started the company, we made a huge bet on computer vision as the right technology for AI and autonomy. This was back in 2014, when it was much less clear than it is today. Natively, our drones have a bunch of cameras. They look out and see the world, and they use that to figure out where they are, how they’re moving, and what’s interesting and important around them.

We come from a place of not being reliant on GPS and having more of an ability to do things onboard. Having said that, candidly, our first round of drones in Ukraine basically failed. We had built the system largely informed by the U.S. Army’s requirements for what they thought was important for a quadcopter in this space, and electronic warfare was just nowhere on that list.

From a radio standpoint and from a navigation standpoint, our first-generation system was just not set up for success. It was a painful process for us. I’ve been to Ukraine twice myself, and we had a bunch of folks on our team spend a bunch of time there. We learned a bunch of hard lessons about what it takes to really operate in this environment, and that started to drive our development.

More so, honestly, than the requirements coming from the U.S. military, we made an explicit decision as a company that we think this is the real-world situation that matters, both from an immediate-impact standpoint and because, whether or not they realize it, this is what everybody else is going to need as well.

There was a process for us that really took a couple of years of adapting the native vision-AI primitives to work in this environment, which we’ve now gotten to with pretty phenomenal results. The drone is incredibly resilient and capable, both from a comms standpoint and from a GPS-denied navigation standpoint, in extremely harsh environments.

It’s a real technology hurdle to get across, and we’re now seeing the bet that we made pay off in many ways.

We're seeing this with other militaries around the world. They're starting to come around to the idea that this is the kind of thing that matters, and our systems are performing. Electronic warfare testing is becoming part of the evaluation protocol for a lot of purchases.

I think Adam hit the nail on the head. The effectiveness of your systems in an EW environment is the difference between whether they're relevant or not. A lot of people will say that their next-generation product is going to work in an EW environment, but I think that's hard to say unless you're actually doing it. That's a pretty tall claim.

Another element of effectiveness in these complex battlefields, I think, is just the adaptability of your capabilities and your software. An anecdote from Ukraine: to solve for this environment, which is constantly changing, we were going to go out—we have a product called the V-BAT. It's a plane; it's about 12 feet tall with a 12-foot wingspan, and it can fly for about 12 hours.

We got the brief from the Ukrainians and the situation we expected: “Here's the situation. You guys are going to take off, go this way, maybe 70, 80, 90 nautical miles. You'll have GPS on the ground. As soon as you get to 200 feet, the jammer is going to hit you, and you have to have no GPS from that point forward.”

The engineering team was like, “Great, got it. Good. We'll get the fix on the ground. We'll take off. We'll go take care of business. It's going to be sick.”

We go out there, and at about 3 feet, the airplane loses GPS. The Russians and Ukrainians are both constantly jamming, too, because if the Ukrainians stop jamming, they find themselves at risk. Even if you're launching on the friendly side, there's intense jamming 3 feet off the ground.

It turns out the airplane takes off and just starts flying in the other direction. I don't know if you've seen the movie Interstellar, where he's going through the cornfield. The guy has a laptop, and he brings down the drone.

Our team and the Ukrainians took off in a truck and chased it for about 2 hours. They found it orbiting over a cornfield about 80 kilometers away, I think. In the span of 24 hours, the engineering team rearchitected the stack not to use GPS at any point in the mission.

We validated it at our Texas facilities, and then we pushed it forward. Twenty-four hours later, the team took off and conducted a demonstration that was written about in The Wall Street Journal. The outcome was ultimately a Russian SA-11 being found, fixed, and finished by a HIMARS.

I tell that anecdote because it will be the pace of deployment that is the make-or-break for militaries around the world. If we want to make a software change in some of our programs, it can take up to a year. It is considered a big deal to change the software.

When we took that aircraft forward to Ukraine, we wrote new software in 24 hours and pushed it. Then we needed 24 hours to properly plan the operation. Unless we find a way for industry and government to work together to deploy software capabilities at that pace, we will find ourselves in a very tough situation.

In a world where you're using autonomous systems, everything is basically software-defined, right? The entire behavior and capability of the system are software-defined. Electronic warfare is also software-defined; the behavior of the jammers and everything is coming largely through software.

In many ways, I think a modern conflict—and you see this in Ukraine—becomes basically a software-writing fight. The speed at which you can write it and deploy it really matters.

I just have no choice but to be hopeful and optimistic here, because we've had the same experience where sometimes it's taken us 2 years to push new software into deployed systems. I would like to think that if we were actually in a conflict, that would just evaporate and we could do it in a day. Maybe that's overly optimistic and hopeful, but the status quo is unacceptable.

You have 2 exceptional stories. We're discussing autonomy, and one of the concerns you either of you may hear from time to time is that, if you have autonomous drones, that means humans are not in the loop, and that means we're ceding our authority to essentially software.

Is that an accurate assessment? What does autonomy mean in terms of humans actually having control over the end state of what the drones are doing? How do we best configure the military, and also civil society, for a future where autonomous drones are not only available but cheap and widely adopted?

The first thing to say—and this is really, I think, the important backdrop for all of this—is that this is terrible stuff, right? We're talking about weapon systems that kill people and create immense suffering. That is the reality of war.

The real goal for everything is to act as a deterrent, to make conflict less likely, and to make it such that, if conflict does happen, you can be maximally targeted and precise and minimize human suffering. I think that's an important backdrop for all of this when we're thinking about what we do and the kinds of systems that we're building.

There are a bunch of legitimate concerns about what it means to have these AI-driven robots, how automated they're going to be, and how much authority we're going to delegate to them. I think we also have to keep in mind that, unfortunately, the game theory here is as simple and obvious as it can be.

Nobody thinks nuclear weapons are good for humanity on an individual level. Deploying a nuclear weapon is a miserable, terrible thing. But the only world worse than one where you and your adversary have nuclear weapons is one where only your adversary does.

One of our strengths as a country is our values and the way that we try to conduct conflict with a high ethical standard. I think it's important to maintain that.

One of the things that I've been impressed by is really the level of sophistication within the military on these issues. There are people whose job it is to think deeply about the implications of different kinds of weapon systems and how authority is delegated. There are very robust controls in place for how the military thinks about these systems.

Those things are evolving as the technology evolves. But this is something I've thought quite a bit about personally. The more I've thought about it, and the more time I've spent with the U.S. military in particular, the more comfortable I've gotten that this is a robust organization that cares about doing the right thing and is thinking deeply about the implications of technology.

My general view, which I think is shared by U.S. military policy and doctrine, is that ultimately human judgment is really important. A human exercising judgment in how force should be used is super important.

But the other thing that people have to understand is that the status quo is not great. Oftentimes, your choice if you want to take out a target is dropping a 500- or 2,000-pound bomb, which is going to cause widespread destruction and a lot of collateral damage.

An AI system that might be using autonomy to a pretty intense degree to figure out where something is and what to do about it is probably better than dropping a 2,000-pound bomb and blowing up a whole city block. You've always got to be thinking about what the status quo is and whether we can use AI and autonomy to better and more precisely accomplish the thing that we care about while inducing as little human suffering as possible.

Maybe the good and true news right now is that human-machine teams are far more effective than machine-only teams. For the next several years, I think that'll continue to be true, and maybe it's true for longer than that.

The frameworks that we have in place, where people are in the loop on the decisions, make a tremendous amount of sense.

I also think Adam brings up an excellent point. The game theory is this: if somebody finds that a machine-only team is the most effective team in certain missions and circumstances, the question is, why and when would that be used? What do you do about it, and how do you make sure that you're ready for it?

I don't think you can live in a world where you have blinders on. I was speaking to a senior military leader who had a nice framing: his expectation is that the more defensive you end up being, the more likely you are to turn things over to machine control to get the very fast reactions and the dominance that you need to come out of that situation.

A great, very practical example of that today is the Phalanx gun system that protects ships, right? If you come into that thing's weapon engagement zone and it's turned on, it will kill you.

You can think about that at a larger scale. If a force is pressing on another force, they find themselves in a defensive situation, and they flip the switch and go full auto. What does that mean? How does that get put back in the box tactically? What does that mean for your forces?

Were they trained to face something that had that level of capability, that level of discretion, and that level of speed? How does it play forward from there? I think there are a lot of hard questions.

I think the convenient answer, the easy answer, is that humans are always going to be in the loop. It's going to be fine. Don't worry about it. But I think the world is a little bit more complicated than that. I don't have answers for it other than a strong conviction that America needs to lead.

So, no matter what you believe about any of this, whether you have conviction on one side or the other or you're just uncertain what the future is going to be—and I think there's a lot of uncertainty about how the future will play out—American leadership is the answer.

At the conceptual level, a lot of these things are less new than they seem. Dropping a bomb in World War II: once that thing leaves the bomber, it's autonomous. It's pretty dumb autonomy, but there's no human judgment left, right? That thing is falling, and it's guided by gravity, wind, physics, and other things. At some point in that trajectory, if it starts heading toward the wrong place or you realize it was the wrong target, there's nothing you can do about it.

So we're used to relinquishing control over the end outcome. Usually, that results in much less precision and much less ability to actually accomplish the thing that you care about. AI fundamentally changes that equation. But I don't think the concept of a human relinquishing control over the ultimate thing that happens is actually new.

Both of you questioned the premise that this is a new thing that's never happened before. It is not the case. Second, both of you zoomed out and said, “Listen, we are not at the end of history. Fukuyama was wrong. There are rival systems of government: totalitarian states and free states. And we have to make sure that, while addressing your concerns in a conscientious, thorough, and democratic manner, we are the ones who dominate this technology to ensure deterrence.”

Now, let's imagine 2 different futures 10 years from now: 1 where totalitarian states—China, Russia, et cetera—have dominance over this technology, not only in terms of procurement and production but also in terms of technology; and another where we maintained and then advanced our technological lead over them and also achieved manufacturing parity, if not superiority. What do those 2 futures look like?

It's hard to predict exactly, but I think AI is the most important technology, really, in the history of humanity. And a world where our adversaries have it and we don't is not a good one. I mean, a world where the Soviet Union had nuclear weapons and we didn't, or a world where Nazi Germany had nuclear weapons and we didn't—these are not pleasant things to think about.

From my standpoint, I think we should just view that as an unacceptable outcome, one that we cannot allow. Some of this plays out beyond the military domain. But I think there is, in AI—and I think this is changing rapidly for the better with the new administration—a lot of talk about safety and regulation: We can't do this, we can't have this many parameters, and if you use more than this much compute, it's not allowed.

If you're sitting in Beijing or Moscow, that's got to be music to your ears, right? “Please, yes, slow down.” It's not to say that we shouldn't be thoughtful, but we've got to be real about the game theory dynamics here and the implications for this technology.

So, to close out, there are, I'm sure, a number of founders watching online, and I'm sure many of them would be interested in getting into public safety or defense tech and wanting to make a difference. They see the mission. They see the need. Perhaps they're scared; they're afraid of failure. They don't know how to start. If you were to give just 1 piece of advice to these founders, what would it be?

Number 1: The mission is worth it. I don't think there is a thing in the world that you can care about that doesn't build from a foundation of security and stability. I wake up every day incredibly excited and honored to have the opportunity to contribute to something that I think is so important.

The other thing I would say is that this stuff is just extremely challenging. Especially if you're building hardware and serving these critical industries where the stakes are really high, you should expect it to be really hard. It's hard in different ways at different points in time. But at some level, I think to be successful at it over the long term, you've got to love that.

I think probably both of us, through different things in our backgrounds, do. I love the challenge of trying to solve these difficult problems. And I think our company—probably both our companies—tend to attract people who want to be really pushed and challenged, work on hard problems, and struggle with things over days, months, and years. That is definitely part of the journey here.

If you want to get to something that is going to have real impact, you've got to embrace the struggle.

Wise words.