The End of American Military Dominance? A Conversation With Paul Scharre

24:04Would we have seen a fundamentally different conflict? Probably not. Maybe a somewhat slower tempo of operations on the U.S. side, slower ability to strike emerging targets. There were certainly instances where the U.S. was like rerouting planes in flight to new targets that were emerging. We might have still seen that without a bit maybe less of that or less effective. So I would sort of think about it as kind of like we're at the biplane days of AI playing a role in warfare. Very early, but a lot of promise for where this might go down the road. I was struck by a line in your piece about periods of disruptive technological change when you say that what determines a military's relative success is how well it employs new technology.

24:46not necessarily being at the forefront of that technology, but as you pointed out, in the AI case, catch-up happens pretty quickly. That does seem to kind of map onto these debates about Chinese priorities in AI development versus American priorities, the priorities of the big American labs, at least, which are really focused on being at the frontier and being first to AGI or superintelligence or whatever. Do you think that the Chinese focus here is actually the right one when it comes at least to the military applications? Yeah, I think probably that's right, that for the military, AI adoption will matter more than AI development.

25:21And I think this is an important asymmetry between how the frontier AI labs, how Anthropic and OpenAI and Google DeepMind are thinking about AI versus how the military ought to be thinking about it. So for the frontier labs, they see themselves, and this may turn out to not be correct, but I think there's valid reasons why they believe this, in a winner-take-all or largely winner-take-all dynamic to gobble up market share for AI. And the reality is that we've seen in the last 20 years or so that there are a lot of digital technologies that have natural monopolies and there's room for one or maybe just a handful of players.

25:58That's true in social media, in operating systems, in handsets, in search, for example. There's like one company, Google, that dominates search. Lots of other companies have tried. It's hard to break into. There are these network effects that often happen for a lot of digital technologies. So for a lot of these companies, they're anticipating that AI will turn out to be similar and there will be like one company that everyone goes to for their AI. Or maybe there's like two, maybe max three players in the mix. That's true today, actually, for Frontier Labs. It's not actually true. Like if you just go like six months behind, maybe, then the model space explodes and you've got a whole bunch of Chinese open source models that are in the mix, and it widens considerably.

26:41I think it's very much an open question what that looks like for AI models. But I would say for the military, that's just not going to be the case, in part because our competitors are going to not rely on American technology. And so we're not going to be able to sort of like rely on, even if there is one US company that controls a big portion of global market share, that's not going to give the US an advantage the same way that like That being true for search means that the Chinese military can't access search technology. There are Chinese competitors in the search space. And so what's going to matter much more is how do you adopt this technology, how to use it?

27:16Or to maybe put this a little bit simpler, if the leading U.S. companies are three months ahead of the leading Chinese companies in AI development, but the U.S. military is five years behind the leading U.S. companies in AI adoption, it doesn't matter. That three-month lead negates to basically nothing. And then it's an adoption race between the U.S. and China of what do I do with AI? How do I use it effectively? We'll return to my conversation with Paul Shari after a short break.

28:06Paul Scharre:Thank you.

28:32on Museum Mile. See why Art News Magazine calls Independent 20th Century a staple of the season and experience historic modern art from around the world and one of New York's most iconic spaces, September 24th to 27th. To learn more, visit independenthq.com. This July, save$10 on an all-access subscription to Foreign Affairs. Get exclusive access to thought-provoking analysis by the world's leading policymakers, scholars, and thinkers. Join the global conversation on the issues that matter most. Visit foreignaffairs.com slash July to subscribe. That's foreignaffairs.com slash J-U-L-Y.

29:18You look at some of the history of this in the essay. You talked about being in the biplane era of this at this particular moment. When you look to, you know, especially that period between World War I and World War II when tanks and airplanes and everything were being integrated in new ways. What determined which militaries were able to figure out how to best use those technologies? Was it idiosyncratic? Was it individual leaders? Was it something about systems? And what are the lessons of that history as you see it? Yeah, I think it's a great historical example here. You know, at the tail end of World War I, we'd seen tanks and airplanes introduced in fairly marginal ways, but everyone agreed in the aftermath, in the interwar period, that tanks and airplanes would play a big role in future wars.

30:00The question was, what do you do with them? And it wasn't obvious to people at the time. There were lots of debates about the roles of airplanes and tanks, the role of carrier aviation, for example. One striking example to me is that Britain was actually out first in terms of adopting carrier aviation, but fell behind Japan and the United States, not because their aircraft technology wasn't good, but because of internal bureaucratic and cultural squabbles within the British military about who controlled the technology and who gets to own these planes. And so those kind of mundane things about bureaucratic politics turn out to matter a lot because it creates incentives for who's willing to adopt this technology.

30:45I mean, I think we can see that today in the way that the different services think about drones, how that the Army and Navy and Air Force, for example, think about drones very differently, or even within parts of the services within the Navy, aviation versus surface versus undersea community, for example, how they think about robotic systems and uncrewed systems, often is driven by bureaucratic incentives. Is this taking away some power from some existing power center? You know, tanks are taking away from the cavalry, for example, or is it giving power to some community and enabling them? And you'd like to think that it should be strategic factors that dominate, but that's not always true.

31:25There are these bureaucratic factors that matter a lot. Is the Chinese military doing better than the U.S. military when it comes to integrating AI especially, but also in kind of adapting to new technologies more broadly? I wouldn't say that they're ahead today. I think the risk is that maybe they're hungrier, and the Chinese leadership is willing to, particularly at the political level, at Xi Jinping is willing to force change on, you know, the People's Liberation Army is not necessarily the most nimble and efficient organization. I mean, they're sclerotic and slow in their own ways, but that there is a sense of being the underdog of wanting to catch up and be willing to force hard cultural change from the Chinese Communist Party onto the Chinese military and get rid of people that are underperforming.

32:16And I worry that being on top for so long has led a sort of assumption within the U.S. defense enterprise of we'll always be the top dog. And, you know, look at how great we did in Iran and, you know, look at how great we did in Iraq and Afghanistan. And all these failures are political failures and they're not failures on the part of the military. And we'll always be ahead and we'll always have better technology or better ability to fight. And that may not be true, you know? And I worry about something like a Suez crisis moment for the United States ahead where we just run into a point where we actually can't perform and we fall short in a big way.

32:54And so I think that to get there requires an acknowledgement that U.S. technology leadership is not a birthright. U.S. military dominance is not a birthright. And we've got to take these challenges like what we've seen a country like Iran, which really shouldn't be able to do the things to U.S. military that it did, pose a lot of difficult challenges, and we've got to adapt to that. Yeah. I did think back to a piece that we ran late last year called America's Drone Delusion by Justin Bronk that made, I think, an argument that's certainly in tension with yours. These are not, obviously, these pieces are not point-counterpoint.

33:28And with apologies to Justin for doing this without him here to defend himself. Well, yeah, I am interested in how you think of his argument. The core of it was that he thinks that the kind of, you know, focus on large scale acquisition of drones or A-enabled drones does not meaningfully enhance deterrence against China. That's just kind of overreading the lessons of Ukraine and that, you know, in fact, China, as he argues, is spending more time prioritizing, you know, crude military hardware rather than unmanned systems. I'm curious how you react to his argument and how you see that difference.

34:01So I think there's a strong version of the drone argument that I'm sure some people make that I want to make clear to listeners I am actually not making, which is that all we need is cheap, low-cost drones. That's not what I think. I think there is a rule for expensive assets. There are things that large stealth bombers and submarines and aircraft carriers do that you can't replicate with drones. I think we need a balance across the force. And right now, the US is woefully underinvested in large numbers of low-cost systems. and we need to rectify that. You know, I think if the goal is, for example, deterrence against China, what is going to get you there?

34:36I think the answer is not like a revolutionary idea. It's getting to the place where Xi Jinping believes that an invasion will fail. And how do you get there? You have to demonstrate the ability to find and sink enough Chinese amphibious ships early enough in an invasion. And if you do that early and you can demonstrate to the Chinese that they're not going to be able to make an invasion succeed, then I think that that accomplishes deterrence. I think drones play a role in that. They're not the only thing that plays a role in that. Anti-ship cruise missiles and long-range bombers and submarines and mines all play a role in that.

35:16I think there's probably a role for drones there. And I think if you want to create this sort of, you know, hellscape is the term I've used over the Taiwan Straits, where anything that's at sea gets sunk, Drones actually have a lot that they can bring to bear in that environment. And we've seen across a number of different conflicts, a drone can create this transparency and lethality in a way that's cost effective that, like, you're not going to do with F-22s and F-35s. There's no way. Where are they based from? All right? They're not going to be able to be operating in that way. And so that's a potential solution or a piece of the solution to this operational problem the U.S.

35:50military is facing. And I suppose the Ukrainian success in using undersea drones or maybe surface drones, as well to sink Russia's Black Sea fleet, but just despite not having like a, you know, Ukrainian Navy that could compete with Russia is a lesson how to do that. Is that relevant here? Oh, absolutely. Absolutely. I mean, so if the argument is like, look, quadcopters are not going to do it. Yeah, absolutely. As we talked about earlier, it's not the right kind of drone. The better analogy here, you're right, I think is what Ukraine was able to do, for example, in crippling Russia's Black Sea fleet using relatively low cost.

36:23I mean, a few hundred thousand dollars for like a drone bow that was able to damage and in some cases sink some really expensive Russian warships. You know, you've got to balance that against a lot of these Russian warships, you know, not well maintained, not well crewed, don't have adequate defenses. Fair enough. But I think you could see the potential of these technologies. And that's a place where there's clearly a lot of opportunity. one way to think about this is technology sort of creating a design space that you can create these low-cost aerial drones or sea drones or undersea drones that are creating new operational opportunities to build things that are pretty cheap that are laden with explosives or sensors that can finally destroy things that are much more valuable and so you know not all the drones that you would want in the taiwan should necessarily aircraft some of them might be surface.

37:17Some of them might be at subsurface. Some of them might look more like missiles or mines. But I think there is an opportunity there with uncrewed systems to create a pretty deadly environment for the Chinese military. How do you assess the state of American deterrence in the Taiwan Strait? How close are we to being able to convince Xi Jinping or instill that doubt in Xi Jinping's mind? I mean, what I'm still hearing, I think it's probably others are from U.S. military leaders, is that doubt exists now, today, that U.S. deterrence is still effective. The question is, what does that look like over the coming years, over the next, say, two to five years, as the Chinese military continues to modernize?

37:52And how do we make sort of investments and, most importantly, make them on a quick timetable so that we can sustain that deterrence and ideally even strengthen it, right? So that rather than getting to a world where in the next couple of years, Xi Jinping thinks, well, maybe this would work, maybe the time is right, in fact, is confidence in the ability of the People's Liberation Army to execute a successful invasion goes down over time. But part of it, there's like a capability dimension to this, being able to demonstrate what the U.S. military could do. But the time dimension is critical, because if the U.S.

38:27military doesn't get there until 2040, that might be too late to need. How has the war in Iran affected that deterrence? Is it, you know, obviously the effect on stockpiles, kind of huge portions of American Patriot missiles and other interceptors are gone, and then some of the harder lessons that you were talking about earlier are surely the Chinese are studying. Does that undermine deterrence? Does that make the Chinese more confident in some ways? I mean, I think that the conflict with Iran has made the U.S. military look quite weak. And I think that, in fact, it's undermined a lot of the deterrent effect that had come out of the U.S.

39:01raid to grab Maduro and Venezuela, which was remarkable as an operational success. I mean, remarkable in every possible respect. And I think a lot of countries sit back and go like, that's crazy that the U.S. military could do that could waltz into a country, grab its leader and spear them away in the dead of the night in like 90 minutes. I mean, that's just unbelievable. And then for the U.S. to be floundering about still really in the Middle East against Iran to see U.S. bases under attack, the U.S. really unable to, I mean, sure, there's bombed targets, but unable to certainly coerce the Iranian leadership to compel them to some kind of even a negotiated settlement over the Strait of Hormuz.

39:41And we're still in this debate where on a daily basis, is the truth on, is it off, are we striking, are we not, do we have a deal? I think all of that makes the U.S. military look very weak. Never mind the discrete material problem of we've depleted a lot of our missile interceptors here. By one estimate, 50 % of the Patriot inventory, 50 to 80 % of the THAAD missile interceptor inventory, these are investments that are not only expensive, but it's going to take a long time to replenish that arsenal. And this is like the global arsenal that would be used in other conflicts as well. We've already known that in a conflict with China, we don't have enough munitions, both offensive and then defensive, like missile defense munitions, for a sustained conflict there.

40:28And we've already reduced that stockpile even further. So I think that all of that definitely undermines U.S. deterrence against China. We've talked about Ukraine a lot as a kind of model and source of innovation. As you look at the war in Ukraine, how do you see the battlefield dynamic there evolving? And, you know, I think there's an impression now that things have turned in Ukraine's favor on purpose of its ability to strike targets so deep inside of Russia. What is your read of that back and forth as each side has looked for new innovations in drone technology or ways of integrating drones?

40:59Yeah, I mean, I think like in a marginal sense, the momentum or perceived momentum perhaps has shifted in Ukraine's favor over the last several months. But I think that some of the dynamics you've seen over the last couple of years are unlikely to change. Of one, in an operational sense, the war being relatively static. I don't think we're going to be in a place, even if Ukraine starts maybe making small gains on the ground, we're not going to see like massive breakthroughs. And all of a sudden, a transformation of the static kind of trench lines into some kind of very fast-moving Blitzkrieg-style maneuver warfare.

41:35Like all of those reasons why it's been static would likely be the case. And at a political level, certainly Ukraine's ability to reach deep inside Russia and blow up things is, you know, better for Ukraine and demonstrate the cost to Russia. Putin's been willing to absorb a lot of costs. I don't think that's going to fundamentally change Putin's calculus. I think on both sides, both sides have settled in for several years now for a long war of attrition, both operationally on the battlefield and politically, both sides are willing to endure a massive amount of costs. And so I think absent some outside dynamic that's going to change things in a major way, I think we're likely to see the war continue to grind on for some time.

42:13Let's look a little bit farther into the future. I noted that you wrote a piece in Foreign Affairs two and a half years ago now called The Perilous Coming Age of AI Warfare. And you, at that time, said that the window to regulate autonomous weapons was closing fast and that quote, without limits, humanity risks barreling toward a future of dangerous machine-driven warfare, even then I think Ukraine had at least claimed to have fielded a fully autonomous weapon that could make, you know, really its own decisions about who to kill on the battlefield without a human in the loop. How much has that developed in the two and a half years since then?

42:46And how do you see both the U.S. and more global debates about putting the right kinds of constraints on autonomy when it comes to those lethal decisions. Yeah. On one level, I think the, um, we've seen over the last couple of years increases in the use of drones and the numbers of drones by other countries and other conflicts and sort of incremental growth and autonomy, I think was predictable to those who were looking at this. And I think it's kind of a long-term trend. I've been a little bit surprised by the way the war is unfolding in Ukraine in two dimensions. One, the scale of drone production is much, much higher than I would have imagined a few years ago, getting to the place where Ukraine is producing 4 million drones a year now.

43:25That's a lot more than I would have imagined Ukraine was capable of doing. On the autonomy side, it is candidly developed slower than I would have expected. Now, we have seen what Ukrainians refer to as sort of last mile autonomy, where basically a human is piloting a first-person drone, and then they lock onto a target, could be a human, could be a vehicle, and then the drone is able to carry that out for the last miles sort of metaphorically, anywhere from maybe like 500 meters to a kilometer, maybe more in the face of jamming. I would have expected that we see more autonomy given how much jamming there is on the front lines.

44:03I don't have a great reason actually why that hasn't occurred. I think the simplistic answer is maybe it's just not needed that both sides are able to get to the point where they can still find targets. And it's really just at the really terminal phase, is getting really up close that the jam is significant enough to not do that. The other reason, of course, is there is concern about fratricide. The front lines are static, but are to some extent a little bit porous, partly because they are so thinned out. And so both sides want to avoid situations, certainly Ukraine, at least, I should say, want to avoid situations where they end up targeting their own forces.

44:41And that's much harder for an AI. To get to the point where an AI could say, oh, that's a Ukrainian versus a Russian soldier, that's harder. It's a lot harder to do than simply saying like, that's a human. So we're not quite there. You know, on the one hand, the political discussions about AI more broadly in society are just heating up. What's AI going to do for our jobs and how do we deal with that? I do think regulating military aspects of AI is probably getting more challenging. You know, the idea that you're going to bring the US and Russia and China to the table to some verifiable treaty to ban autonomous weapons.

45:13It didn't seem realistic five years ago. It seems even less realistic these days. And so I think that getting to a place where there's political momentum actually achieves some kind of regulation over military ad is meaningful, seems hard. I saw recently the U.S. Secretary General was speaking out against autonomous weapons, but that doesn't mean that's going to lead to actually changing what militaries do. Where is U.S. military policy on this question now? And where, in your mind, if you were kind of put in charge of crafting this, what would you be doing? So on autonomous weapons specifically, There's been a policy in place since 2012.

45:48It was updated a few years ago. The Pentagon has said that they are in the process of launching a new review of that policy. That really is an internal review process that doesn't say the U.S. military can't build autonomous weapons. Basically says if the U.S. military were to build weapons that are different than what we've done in the past, they look different than, say, mines or some of the automated features on, like, air and missile defense systems that we already have, it had for decades, then they'd have to go through a more extensive review process to make sure not just that they're lawful but that they work properly like we're going to send these things out into the real world that they're not going to kill civilians or target our own forces for example personally i think those are sensible practical steps disclosure i was involved in writing the original policy back in 2012 so i do have some attachment to it i don't know where we'll end up on this needless to say in the wake of the sort of big, messy public breakup between the Pentagon and Anthropic earlier this year, there's been a lot more public scrutiny over how the military is using a lot more congressional scrutiny about what the rules are going to be and who sets those guardrails.

46:54You know, if the Pentagon were to sort of scrap the policy entirely and say our policy is, you know, unleash the killer robots, they might get a little bit of flack for that. I'm not sure how much this team would care. It doesn't mean they wouldn't do it. So I think, you know, in terms of outcomes, it could be anywhere from they have maybe renew the current policy with some slight updates. It doesn't restrict development of autonomous weapons, doesn't say you can't do it. Just kind of says, if we're going to do it, here's some ways to be thoughtful about it. Or they could say, we're going to, you know, press delete on this policy and it's anything goes.

47:26I think there's a lot of possibilities here. How scary does that get to you if it's anything goes? I mean, I'm not like, I worry more about in the broader sense, what happens if we see all sort of race to the bottom in terms of human control over warfare between not just the US, but also competitors. You know, I think the US military is inclined to act pretty responsibly. And even if you took the policy away, people are likely to be pretty thoughtful. The US military doesn't want weapons that they can't control. in some ways one of the motivations at the time behind this policy was in the absence of any policy what we were finding was people were just uncertain about what to do they were like can i build this am i allowed to build this having a process in place gives people a place to get a yes which is advantageous if you want to think about how do you make it possible for the u.s military to actually move forward with this technology so i'm actually not sure that like no policy is a good move, not just from a safety standpoint, but also from the standpoint of enabling the U.S.

48:30to seize the technology. You might get a place where people are kind of paralyzed because they don't know what to do. They're like, well, I mean, am I allowed to build this thing or not? I worry more about a sort of broader dynamic of surrendering human judgment to machines. I think in my mind, the risk is more insidious and gradual. We end up in a place 20, 30 years from now, when we kind of look back and we're like, wow, this is scary, a world where humans are no longer in control of warfare. The U.S. a few years ago said that it would keep a human in loop for decisions regarding the use of nuclear weapons and made that kind of public commitment.

49:04I don't think Russia or China or North Korea has. Do you worry about this undermining deterrence, causing even, you know, kind of mistaken use of nuclear weapons to go to one of the most, you know, alarming of those scenarios? Yeah, I mean, I was very encouraged to see that. I think that's a, that's kind of a low bar to clear, right? Hey, you know, while we're building the cover, let's make sure he's been in charge of nukes. It seems like a pretty reasonable argument to make. We haven't seen anything official out of the Chinese military. There had been a statement that had come out of a summit between President Biden and Xi Jinping during the Biden administration, where China had sort of also agreed to maintain human control over nuclear weapons.

49:47Well, fuzzy, what if anything that means? but it felt like maybe a half step in a positive direction. We've not seen anything like that from Russia. I mean, in a perfect world, I'd love to see something like a P-5 statement about the importance of human control over nuclear weapons. Again, that seems like a low bar to clear. On one level, you know, one might look at this and say, like, this is, who would even suggest otherwise? On the other hand, there are examples like the pro-nitor system that the Soviets had reportedly built and allegedly deployed in some fashion, and that according to some stories and reports by Russian generals, maybe Russia can just continue to have in some fashion an automated sort of like nuclear retaliatory system that would allow Russia to strike back if there was a first strike against where humans would still have to authorize this.

50:33And there'd still be technically a human in the loop, but it would automate a whole bunch of the processes. And I do think this is a place where like machines break. I mean, you know, like anybody who interacts with a computer knows they break all the time. They're doing weird things. Your Bluetooth is connecting to the wrong thing. I mean, like, if we're struggling to get basic technology in a place where it's foolproof for these really high-stakes applications, there's a lot of value in having humans still at not just one place, but like multiple steps to that process. I think that there's a lot of value that human judgment brings.

51:08You mentioned the blowup between Anthropik and the Pentagon. I mean, the role of private companies and their workforces on some level is another fascinating dimension of this. Do you imagine continuing friction between AI labs and the Pentagon as we try to hash out a responsible policy? Or has the government kind of asserted its power and that's where it's likely to stay? No, no. I think this is going to be a continued source of tension. I think it's unfortunate because in many ways it's a revival of the tension that we saw, you know, almost a decade ago when Google decided to discontinue their work on Project Maven.

51:39And we saw in response to that, like kind of a big existential panic from a lot of people in the national security space that they could get cut out of this transformative technology that was coming out of the commercial space. And that is what's fundamentally different about AI. In some ways, it's the opposite of stealth technology that was invented in defense labs. It doesn't really have a commercial application. This is coming out of commercial labs. And for these companies, the government is important insofar as the government has vast regulatory powers at its disposal to make their lives difficult, to put in place export controls or gum up the release of their models.

52:17We've seen the US government do this most recently with OpenAI and Anthropic in terms of some of the cyber capabilities of their models. But the government's not a big customer, the same way it is for defense companies. In fact, the government's a pretty marginal player in this space. We think of a total dollar amounts. And so, you know, a lot of the AI engineers, they're not coming from the national security space. They may not be supportive of military applications. They may not want to work on that. In the wake of the Maven dispute with Google several years ago, what we saw is the military leadership and defense leadership went on a listening tour among universities and tech companies, talking to people in the AI space, understanding their concerns.

52:58the outgrowth of that was the DoD AI ethics principles stating how the military would use AI in a responsible way. It wasn't putting in place limits on what the military was going to do, but it was communicating both externally and internally to the defense enterprise, sort of how are we going to approach this technology in a way that's thoughtful and responsible. And they want to kind of this bridge building, you know, kind of tour in a way that's opposite of the current approach, which has been, you know, kind of go nuclear on this, on both sides. I mean, that's certainly true. Pentagon leadership, I think also to some extent of people in the tech space.

53:35But, you know, we've seen letters from, it's not just anthropic, letters from employees, open letters at Google and OpenAI calling on not working with the military. That's a real problem. And I think we've got to find ways to bridge this divide between Washington in Silicon Valley, I mean, the government could forcibly compel U.S. companies to give up models using the Defense Production Act or some other kinds of authorities. That's not going to be effective in integrating the technology in a way that's effective. To do that, to actually use AI effectively, you've got to have a tight relationship with AI engineers who can sit side by side with people in the military to talk about, here's what it can do, here's what it can't do.

54:18Like, imagine if we were trying to build like advanced fighter jets and the engineers and the military were never able to talk to each other. Jets wouldn't fly very effectively. And so I think we've got to find ways to repair that relationship. Well, Paul, I'm glad we could end on a note that was not an accidental nuclear attack caused by a hallucinating AI. But thank you for doing this. There's a lot we will, I hope, pick up in the coming months and years in our pages from you and others. But thanks for the great piece. It's called Losing the War of the Future. And thanks again for joining me today.

54:48Thank you. Thanks for having me on the discussion and thanks for your support with this article. Very excited to take this out there.

55:19make sure you subscribe to the show wherever you listen to podcasts and if you like what you heard please take a minute to rate and review it we release a new show every thursday thanks again for tuning in

55:48We hope you enjoyed this episode of the Foreign Affairs interview. Don't forget to visit foreignaffairs.com slash spotlight to sign up for Dan Kurtz-Valen's free weekly newsletter featuring more of the clear-headed analysis that you enjoy here on the podcast. Sign up today at foreignaffairs.com slash spotlight.