An AI-Driven Second Nuclear Age

The world is in an AI-driven second nuclear age. Over the next 15 years international security will become intrinsically more dangerous, perhaps unacceptably so. This is not the result of any specific US policy, or of any other country’s strategy. Nor does it follow from how conflicts currently underway turn out. The danger arises for deeper structural reasons.[1]

One is that a collection of new technologies is changing the nature of warfare. Militarized AI, cyber, space weapons, and new kinds of political warfare. Many items on this list are interconnected by AI.

Another structural change is the return of nuclear weapons to strategic competition. Nuclear weapons are even more deeply embedded in the international system now than they were in the Cold War. This is the case for superpowers and now for second and third-tier countries. While the “classic” nuclear dangers of the Cold War are recognized, those of successor worlds—our world—are not. They will have to be discovered along the way, and the world is only at the beginning of this learning curve. What few people recognize is that new AI technologies will unavoidably spill over into nuclear forces and strategy. The consequences of this are too jarring, too unsettling to think about at the present time. But this will change by necessity as that process unfolds.

The third factor that makes successor worlds to the first nuclear age dangerous are major shifts in international power. This stems from changes in economic clout, business, and the spread of advanced technologies to many parts of the world that historically have had little access to them. Others increasingly have access to advanced technology. This has major consequences in both peace and war.

New Technologies

Technology and innovation will play an outsized role in strategic competition in the future, even more than it has in past. AI is the standout exemplar. But other technologies are undergoing extraordinary innovation as well. Cyber, weapons in space, hypersonic missiles, precision strike, missile defense, autonomous weapons, drones, lasers, stealth, superintelligent machines, hyperscale data centers: these technologies are revolutionizing warfare. They are changing competition just as they have in virtually every business, from automobiles to retail and from communications to health care. Competition based on technology and innovation is much more unpredictable than standard forms of rivalry. This is because performance is highly uncertain in the beginning. Moreover, no advantage lasts very long because others copy it or come up with something better. Each performance improvement leads to a counter that tries to offset it. Agility—the ability to change direction quickly—becomes a significant competitive advantage. As a result, much more effort has to go into research and development, both to stay ahead of rivals and to keep a close eye on what competitors are doing.

AI, in particular, is transformative in many different parts of the military. Here, I define AI in the common, widely used sense of the term: computer systems performing tasks that would typically require human intelligence. It will have such wide-ranging consequences that it is difficult to make broad generalizations. It’s better to get into specific applications of AI, much the way business schools frequently use case studies to teach broader lessons.

One major application of AI applies its computational power to the problem of tracking and destroying mobile targets—those whose survival depends on movement, hiding, or relocating. Examples are aircraft carriers, satellites, leadership groups, airplanes, submarines, and many missiles. Such targets have been around for a long time, clearly, but until quite recently they could not be reliably tracked. AI in combination with other technologies like cyber, phones hacks, computer vision, sensors in space, and drones provide a new capability to track—and destroy—these moving targets.

A business example shows how this works. It is highly suggestive of what is going on in the military in the United States and China now, as will be made clear.

US automobile insurance companies offer downloadable apps for your phone. A customer receives a discount on their insurance premium if they download the insurer’s app, so there is an incentive to do so. The downloaded app accesses other parts of the phone, especially GPS and the cell phone network. Given this access, the insurance company can track how fast you accelerate from a red light. Should you run a stop sign without slowing down, they will know. They can record your speed, where you go, and even how many left turns you make—left turns lead to more accidents than right turns.

The insurance company integrates this data into a behavioral life pattern of the customer. They can look to see how your driving changes after you’ve been out to dinner and perhaps had a drink or two. They can see how your driving changes after one hour at a pub, or after two hours. In the near future, they will have ways to turn off your car if you drive erratically. China has this ability today.

If you add a new driver for the car, such as a teenager, they can determine from the driving pattern whether it’s you driving the car or the teen. Should you claim the teens aren’t allowed to drive the car, they can determine if they are doing so, or whether you are driving.

AI can process all the diverse data on this “mobile target,” your automobile. They can compare your driving to other drivers. Moreover, and this is important, the company can scale the tracking because the data is in digital form. The data is used to train their AI models. State Farm insures 45 million automobiles in the United States. Their algorithms set your premium and decide whether to terminate your policy. A cancelation notice is immediately sent to the insured customer, and perhaps to the Department of Motor Vehicles.

Almost every business school in the United States has courses that go into the processes used by AI to track customers. There are case studies of DoorDash, Uber, and Amazon on their algorithms and processes. AI for mobile tracking is even more advanced in China than it is in the United States.

A big difference between commercial and military applications of AI, obviously, is that the former are cooperative targets. In the military they are not. Nonetheless, there is a tremendous amount of data that exists for moving military targets. Data collectors include satellites, drone video, heat (infrared) signatures, radar tracks, sonar, telephone hacks, computer breaches, gravitational and electric field disturbances, telemetry, automated license plate readers, security camera hacks, radio intercepts, and data from tracker beacons. Each of these is an “information touchpoint.” There is an extraordinary amount of information in aggregate about moving targets because the data is updated frequently. This is one reason why hyperscale data centers are sprouting up everywhere—storing and processing this data is a major challenge.

Downloaded apps to mobile devices like phones and tablets are as useful to military intelligence as they are in business delivery services. Tracking apps may be concealed into social media accounts, sports betting apps, contests, or a special discount. Such “cover” apps conceal intelligence tracker software of various kinds. These apps can be disguised as automatic system updates, or people can be enticed to download them with a promise of winning a free meal.

Users, of course, would have no idea they were being used by an enemy intelligence service. Enemy soldiers, say, assigned to a missile brigade could be tracked without their knowledge or cooperation. Their proximity to nuclear warhead storage sites, launch sites, or support centers could then be estimated. This is where machine learning comes in. Data collected for years of military exercises could be used to map out the timing and spatial layouts of, say, an enemy missile brigade as it sets up to fire. Patterns in timing and physical distances would emerge. These patterns would tip an observer off about the purpose of the exercise. In this manner one could learn about the brigade’s vulnerabilities that even its commanders were unaware of. For example, they might find that certain trucks break down frequently and cannot be relied on to carry out their role with reliability.

We can take this example a step further. Suppose that in addition to tracking mobile phones with covert apps installed on them there are additional data streams: computer email/text hacks, radio intercepts, or drone video of the brigade. Data from these sources could be combined into an overall picture of the missile brigade. This leads to something called multimodal AI, meaning there are multiple modes of description of the target’s behavior pattern. It is not hard to see how this could be useful for targeting purposes.

Militarized AI has many other applications. One is that it integrates the weapons used to destroy a target into the tracking process. AI can synchronize surveillance, sensors, and shooters. Synchronization of killer drones, aircraft, and missile readiness is complicated. AI drives the kill chain that ensures its parts are coordinated to destroy the target.

AI powered cyber-attacks are another application. The key driver here is AI’s ability to analyze cyber defenses very, very fast. Moreover, they can do this analysis for large scale cyber defenses. What used to take months and years to locate weak spots can now be done in much shorter time periods because computer search can go through billions of lines of code in far shorter time. The implications of this for cyber war are enormous.

There are many other applications of militarized AI. Autonomous weapons, cryptography, and new kinds of political warfare are coming and are described in my cited book. We are only at very beginning of the learning curve for this.

It is hard to believe that these applications will not spill over into nuclear forces and strategy. After all, these weapons are key targets because they can do so much damage. In addition, the broad shift in nuclear force structure in recent decades is to make nuclear weapons mobile. They gain survivability by moving, hiding, or relocating. Since nuclear stability depends on survivability after a first strike, a technology that undermines their survival, like AI, has enormous consequences for national leadership decapitation, destroying mobile nuclear missiles and ships, and blinding enemy satellites. The technologies to do this are in gestation right now. They are not yet ready for employment at scale, but they are growing in performance and budget.

Further advances are based on digital technology. This means that they are likely to be scalable in ways that older innovations based on industrial capability were not. To produce tanks, ships, and missiles at large scale is difficult in the United States. History shows this. The new AI technologies give a much bigger, faster bang for the buck.

Possibilities for creative thinking are wide open in this field. Audacity, guile, and deceit play a key role. They could be more important than technical innovations. Press reports of military use of AI in the wars in Gaza, Lebanon, and Iran show this. For all of these reasons AI is certain to spill over into the world of nuclear forces and strategy.

The Nuclear Context

No country today is giving up its nuclear weapons. Every nuclear-armed state is enlarging or modernizing its force. As a result, the nuclear context of international affairs is “thickening.” The background, setting, and circumstances that shape how international politics, crises, and wars are seen will increasingly have nuclear weapons in it as a factor. The question of “where things might go next” in a conflict or a crisis will be shaped to a greater degree by this nuclear context more than it has been since the days of the Cold War.

China’s nuclear break out is the most jarring example of a thickened nuclear context. It is big and bold, with intercontinental ballistic missiles (ICBMs), submarine-launched ballistic missiles (SLBMs), bombers, medium range missiles, new warheads, expanded testing facilities, and satellite-based warning and navigation systems. The breakout is accelerated. Beijing could have rolled out a larger nuclear force slowly, over many years. Instead, Beijing chose to compress development into a small number of years.

China’s nuclear breakout is an intellectual shock because few experts in the 2000s and 2010s thought China would develop into a nuclear superpower. Now, nuclear parity with the United States is coming in the early 2030s. We do not know if the buildup will stop there, at parity. Beijing could go for nuclear superiority. I can think of a wide range of future possibilities for a nuclear China except one: China’s returning to minimum deterrence.

Russia, for its part, is building new kinds of nuclear weapons: a megaton “tsunami-maker” autonomous torpedo, giant ICBMs carrying 8 to 15 hydrogen bombs, orbital nuclear weapons. At the same time, Russia’s conventional army and navy have been shown to be dysfunctional in the war in Ukraine. This is an example of a new danger that is yet to be identified and named. I call it a “half made” military. It is a “modern” military with advanced technology weapons like AI, cyber, drones, and nuclear weapons. It has “half” of a modern force in technology. What it doesn’t have is the “other half,” of a modern force: training, organization, and system integration.

In the nuclear age, a half-made military is a danger to itself and the world. This is because it may be forced into a position where it has no other options to defend itself except the use of nuclear weapons. I very much doubt that Russia will be the only example of a half-made military in the years ahead.

India’s nuclear weapons also are modernizing. Delhi has a more complex nuclear structure now. It has a nuclear triad, a three-tiered nuclear force, like the Cold War superpowers had. ICBMs form one leg of its triad, SLBMs are a second leg, and bombers make up the third leg. This gives Delhi a reliable second-strike capability against China or Pakistan because the triad concept is that at least one part will survive a first strike to retaliate.

Pakistan and North Korea are building nuclear weapons and missiles as fast as they can. Both have weapons with short, medium, and long-range capabilities. North Korea, in particular, has operational ICBMs that can reach the United States. In the 2030s they will have many more of these.

North Korea will not have a high-tech nuclear force like the United States, Russia, or China. They will not have space-based warning satellites and modern command and control. It is likely to be more accident prone for this reason if it is pushed too far in a crisis. Indeed, that is the main point. No one in their right mind would want to press North Korea too far with its “jittery” nuclear command and control. It just might go off—to launch something at someone.

North Korea will possess an adequate deterrent against the United States based on a large missile force—and on Pyongyang’s inability to control it if it were raised to high alert level. Call it a threat that leaves something to chance.

What is also interesting is the emerging competitive dynamic between North Korea and the United States. Pyongyang may plan a nuclear force against a US conventional counterforce attack. In other words, North Korea would size its nuclear force against a conventional attack that exploits the new AI and related technologies that allow the United States to track mobile missiles. This is quite a change from the “classic” Cold War dynamic, which assessed nuclear weapons versus nuclear weapons. Missile exchange models, accuracy, and the number of warheads on a missile (MIRVs) described that competition. Now, we are looking at something quite different: An assessment of the military balance has to compare nuclear versus conventional weapons. AI drives a conventional counterforce attack against an opponent with mobile nuclear missiles.[2] This leads to a very different kind of competition than in the Cold War. Again, we are only in the early stages of this learning experience.

Even France and the UK are getting into nuclear modernization. Paris worries over its ability to penetrate the new technologies of missile defense. Consequently, the French have a multiple-warhead hypersonic missile in development. The UK has a massive nuclear modernization underway too. It includes new warheads and submarines, and nuclear capable F-35s.

The outcome of the war in Iran will likely have a large impact on what Israel does on nuclear issues. Though this eventuality is uncertain, it is hard to imagine that Israel’s response will be to reduce its nuclear forces.

The United States has its own nuclear modernization underway. The legacy triad of bombers, ICBMs, and SLBMs is receiving replacements in the B-21 bomber, the Sentinel ICBM, and the Columbia class of submarines. Command-and-control improvements and new warheads also are in the works. So too is “thicker” missile defense for living in a nuclear world where more countries have these weapons.

Still another part of the US modernization is nuclear-capable F-35 aircraft. A new warhead, the B61-12 precision guided thermonuclear gravity bomb, has been developed for this mission. This brings together the stealth technology of the F-35 with AI-driven target tracking of mobile missiles described above. It further shows how advanced technologies—stealth and militarized AI—spill over into US nuclear forces and strategy.

 Taking together these national trends means that nuclear weapons are more deeply embedded in the international system than they were even in the Cold War, the first nuclear age. That world had two nuclear superpowers that mattered. There were other smaller nuclear powers, like China, France, the United Kingdom, and Israel. But for the most part, the smaller powers didn’t matter in the overall strategic rivalry. They were contained in taut disciplined blocs like NATO or they were greatly dependent on a superpower, like Israel was on the United States. The two superpowers drove the strategic competition.

Today, in the second nuclear age, nuclear superpowers like the United States, Russia, and China still matter very much. But second- and third-tier nuclear states matter very much as well. They matter in ways that France, the United Kingdom, and China did not in the Cold War. For one thing, they have independent command and control. North Korea, Pakistan, India, and China really are independent decision-making centers. Others, like Iran were they to field a nuclear force, would be independent as well. This has far-reaching implications for war and peace in the years ahead. We are only beginning to understand what they are. The classic dangers of the Cold War are understood. But the new dangers of a second nuclear age are only beginning to be identified.

Can the World Absorb the Coming Arms Race?

It is useful to take a step back to understand the structure of the military technological world we are in. People worry about the nuclear arms race. They worry about militarized AI.

What I worry about is a concurrent arms race in all of the technologies—cyberwar, space, militarized AI, stealth, and nuclear weapons. In a few years we may have to add quantum computing to the list. It makes little sense to separate out the competition technology by technology. Each technology reinforces the other. What matters are the consequences that result from the synergy between them.

We need to change how we think about technology and war. In the Cold War, strategists and arms controllers had the luxury of a binary concentration on nuclear versus conventional weapons. Complex models were built on this dichotomy. But the days of this intellectual luxury are over: the familiar categories need to be rethought, repositioned, in a world with a wide array of new and interacting technologies. Our old strategic concepts were based on this separation of conventional and nuclear weapons.

But there is nothing sacred or immortal about old strategic concepts. We should be willing to look beyond old concepts and assumptions in regard to the new technologies. To label military AI, cyber, and space as “conventional weapons,” and anything with a nuclear warhead on it as a “nuclear weapon” overlooks what the weapons in combination mean.

Admittedly, the old binary labeling—conventional and nuclear—is easier, intellectually speaking. Our strategic concepts and organizations use them for this reason. They are familiar. But continuing to call AI and cyber “conventional weapons” overlooks the far-reaching implications they have on nuclear stability and more broadly on the gestalt shift that AI, especially, is having on the world.

This leads me to question whether we understand the arms race that is now underway. It is likely to be unfathomable if it is analyzed with old intellectual capital and outdated categories. Countries are unlikely to see “only conventional threats” if AI driven counterforce targets their central nervous system and the brain of their armed forces.

The technological arms race is going to be very messy for other reasons beyond outmoded intellectual capital. The “newness” of the technologies themselves creates problems. We are early in the learning curve of each technology described in this paper. How they work together and how they are entangled with politics are only beginning to be identified. There is reason to doubt that a president, paramount leader, or a general staff may not understand their own forces. This has happened before in history.

The difference now is that the technologies are more complex, and more escalation prone because reaction times are so short. There is no time for human beings to make all the decisions in the time needed. Algorithms—and AI—have to do it.

For these and other reasons some people believe that the coming arms race itself could be as dangerous as the threats it was designed to deal with. This belief extends to many countries. It only takes one country to make a lot of trouble for the United States. But there are many countries who will have the capability to do so. It is asking a lot to expect all countries with modern military technology to be prudent, risk averse, and conservative. There are more decision centers that matter. Some may not be cautious and careful. New military technology has spread and so have nuclear weapons. Absent a broader, sober analysis of the world that is coming we are in serious danger for reasons we do not yet even recognize.

Featured image: Brecht Corbeel/Unsplash