[10th EAI Academy] ⑦ The Military Use of AI and the AI-Nuclear Nexus

YouTube Link: https://www.youtube.com/watch?v=UdZB8QT-Ztg

■ Kim Yang-gyu_Professor, Korea National Defense University.

Video Transcript

I teach at the Korea National Defense University. Before coming to the university, I was a senior research fellow here. Just three years ago, I was involved in planning this academy and inviting speakers, so it feels quite special to be back. Every year, distinguished speakers have come to create meaningful sessions, and I feel it is even more significant that I now have the opportunity to lecture at this academy. Since my doctoral studies, I have been researching deterrence and have also worked on nuclear weapons strategy. After coming to Korea, I learned that South Koreans have a strong desire for nuclear armament. I have also conducted research on public opinion showing that 70% of the population supports nuclear armament. More recently, since 2022, I have been studying artificial intelligence (AI) and am researching how it will transform the military landscape.

Today, I want to discuss the changes that could arise from the combination of AI and nuclear weapons, and why this makes the Korean Peninsula particularly vulnerable. While it is a somewhat somber topic, I hope that by thinking through it together, we can find new breakthroughs. The question I want to pose to you today is this. During the ongoing war involving Iran, a very innovative change has occurred: digital targeting. During this war, 1,000 targets were struck in 24 hours, 1,700 in 72 hours, and over 13,000 in 38 days. This means it took an average of four minutes per target. I would like you to consider how much deliberation was possible in the process of a human thinking, deciding, and striking a target within such a short span of four minutes.

The Combination of AI and Nuclear Weapons: A Shift in Military Strategy

According to Iranian claims, a missile that struck an elementary school in the Minab region resulted in the deaths of over 100 children. Someone must have made this decision. I want to talk about the gravity of the human decisions made during targeting, how this is changing future warfare, and what transformations it will bring to the Korean Peninsula. Today's lecture will be divided into three main parts. In the relationship between AI and nuclear weapons, the nuclear aspect is more important. AI itself is not an independent weapon system or technology; rather, it typically acts as an enabler for force enhancement. In other words, AI combines with existing technologies to enhance their effectiveness.

In understanding the connection between AI and nuclear weapons, the most crucial first step is to understand nuclear weapons themselves. It is logical to first understand how nuclear weapons have operated, and then to grasp the changes that occur when AI is introduced. Therefore, in the first part, I will explain what nuclear weapons are and how they have functioned until now, how they are changing with the introduction of AI, and what this portends for the Korean Peninsula.

The Impact of AI on Nuclear Deterrence

To summarize the main points I will cover today: Nuclear deterrence is fundamentally based on a second-strike capability. I will explain this in more detail shortly. As many of you may know, the concept of Mutual Assured Destruction (MAD) is the foundation of nuclear deterrence. This concept is very important, so I will explain it in detail today. However, the introduction of AI has enabled faster detection, more accurate targeting, and more precise operational execution, which in turn makes nuclear second-strike capabilities—that is, the survivability and stability of nuclear weapons—vulnerable. The core argument of my lecture is that this could lead to the collapse of strategic stability. There is one thing I will not be addressing in my lecture today: the concern about what would happen if machines, or AI, were to use nuclear weapons at will. Currently, there is an agreement between the United States and China on the important principle of not connecting AI to nuclear command and control. Therefore, we do not yet need to worry about a scenario where a machine suddenly decides humanity is not worth preserving and eliminates us. We don't need to delve into the concerns of sci-fi movies like *The Terminator* or *The Matrix*. I want to focus on how AI is changing the existing nuclear order and its mechanisms, and how this is altering the incentive structures and cognitive frameworks for humans. AI is not abolishing nuclear strategy; rather, it is changing the informational and temporal conditions under which nuclear deterrence has operated, and I intend to argue that this is extremely dangerous.

In the context of the nuclear nexus, what I will not be covering are sci-fi threats like killer robots, fully autonomous launch systems, or *The Terminator*. While some argue these threats could materialize in five or ten years, I want to focus on the technologies actually being used in current conflicts: the fusion of AI-based intelligence, surveillance, and reconnaissance (ISR) capabilities; target identification and prioritization; enhanced decision-making support; and the execution of decapitation or precision-strike operations. These are the real changes I want to highlight. I want to emphasize that these changes alone have already made the situation extremely vulnerable. In short, the introduction of AI has accelerated the speed of operations to the point where human judgment is becoming almost meaningless.

I am going to discuss how the reality of drastically shortened timelines—from target identification to strike—will change future warfare and the stability of the Korean Peninsula. Now, we need to talk about how nuclear weapons operated in the past. What would happen if a nuclear weapon were to be detonated? I've said many critical things about North Korea, which has led to me being hacked and disliked by many. I apologize for that. Let's imagine that you all die because of me. You've come to this lecture, and suddenly you receive an emergency alert on your phone. It says an object, presumed to be a missile launched from North Korea, is expected to strike Yongsan in 10 minutes. It appears to be a nuclear missile. Evacuate. You only have 10 minutes.

Now, I'll ask a question. In the remaining 10 minutes, is there anyone here who would scramble to get as far away from Yongsan as possible? No one. Then is there anyone who would rather head towards Yongsan? Why would you want to go towards Yongsan? You said, 'To die cleanly.' Let me explain what happens when a 100-kiloton nuclear warhead detonates.

When a nuclear warhead is dropped and explodes, what is the very first thing that happens? Immense heat. A fireball, a sphere of heat about 200 to 500 meters in diameter, is created. How hot is a 100-kiloton nuclear warhead? It's comparable to the surface temperature of the sun. You probably thought of nuclear weapons as just being scary, but you likely never considered how hot they are. If you are hit by a nuclear warhead, you would experience heat on the scale of the sun. That's the first effect. So, what happens next when that much heat is generated so suddenly?

Everything will melt. When this immense heat is suddenly released, what happens to the surrounding air? It will heat up and rise in an instant, leaving behind a near-vacuum. Then, the surrounding air will rush in to fill that empty space. This creates a massive hurricane. For a 100-kiloton warhead, the wind speed at the center would exceed 800 kilometers per hour, which is eight to ten times that of a Category 5 hurricane. I experienced a Category 5 hurricane in the United States. I was safe in a shelter, but when I came out, everything was destroyed. If a nuclear warhead were to hit Yongsan, all life within 370 meters would be vaporized. Buildings would be destroyed as far as Seoul National University. There would be complete destruction up to 4.5 kilometers, and even at Seoul National University, buildings would be at least half-destroyed.

After that comes the fallout effect, radiation exposure, which we are familiar with from Fukushima. Earlier, you said you'd want a clean death. This is the answer to that. Even if we try to flee for 10 minutes, how far could we get? At best, to the area of partial destruction. Ultimately, you would be crushed by a partially destroyed building, suffer second- or third-degree burns, and die from radiation exposure. I apologize for delivering this gruesome story so casually. If you are near ground zero, you will die before you can even think 'I'm being obliterated.' The heat is so intense that your body itself will melt. The more you try to escape, the more horrific your death will be.

So, there's something I always tell those who have heard my lecture. If such an event occurs and you are not confident you can get more than 30 kilometers away, I tell them that simply heading towards ground zero is the way to a much cleaner death. This is no laughing matter, but a nuclear warhead renders all our previous calculations about security meaningless. I don't know if you've seen the movie *2001: A Space Odyssey*. In that movie, when two political communities had a conflict of opinion, one community would wage war to impose its political will on the other. Why did they wage war?

What was gained by conquering Troy through war? Soldiers acted as a defensive wall between the enemy's population and their own forces. Therefore, to impose one's political interests or will on an adversary, one had to make the adversary's population suffer. One had to inflict suffering by burning homes and taking women as slaves to make the other side surrender. The logic of old wars was that you had to eliminate the enemy's military forces. Only then could your own military power be projected directly onto the enemy's population. This was the rationale for war. But now, that is no longer necessary. You can just fire a missile. Into the rear areas, completely bypassing the enemy's military forces. If North Korea fires a missile now, it arrives in five minutes. In five minutes, you can kill the enemy's soldiers and civilians indiscriminately. This is a revolutionary weapon system.

Even a single attack can have enormous consequences. When a strategic nuclear missile of about 100 kilotons detonates, calculations using NUKEMAP show that the number of immediate fatalities would be 200,000, with the death toll reaching 1.4 million within a day. If it were to hit Yongsan in Seoul, the suffering would be immense. Because it can inflict tremendous suffering, wiping out property and lives in an instant, even a single attack can lead to devastating political consequences. Right?

If a nuclear missile hits Yongsan, all life within a 370-meter radius will be vaporized. The range of building destruction extends to Seoul National University, which would also be at least half-destroyed. Afterward, there would be a fallout effect similar to the Fukushima accident. In response to the comment about wanting a clean death, it would be difficult to escape the fallout even if you fled for 10 minutes.

At best, you could reach an area of partial destruction. You would be crushed by a partially destroyed building, suffer second- or third-degree burns, and die covered in fallout. I apologize for relaying this gruesome story so casually, but near the blast's epicenter, the intense heat would cause your body to melt. The more you try to escape, the more horrific your death will be.

Nuclear weapons have completely overturned traditional security calculations. As in the movie *Odyssey*, in the past, wars were waged to impose political will. War was the act of eliminating the enemy's military forces. This was because the enemy's military served as a defensive shield between their population and your soldiers.

To impose one's political interests or will on an adversary, one had to inflict suffering on the adversary's population. Inflicting suffering, such as by burning homes and enslaving women, was necessary to make the other side surrender. Therefore, the objective of past wars was to eliminate the enemy's military forces so that one's own military power could be projected directly onto the enemy's population.

Now, war is no longer necessary. You can just launch a missile. Regardless of the enemy's military forces in the rear, a North Korean missile can arrive in five minutes and kill civilians and soldiers indiscriminately and in great pain. This marks the emergence of a revolutionary weapon system.

If a 100-kiloton nuclear missile were to hit Yongsan, Seoul, NUKEMAP calculations estimate 200,000 immediate fatalities and 1.4 million deaths within a day. This is horrific suffering that wipes out property and lives in an instant, and a single attack can lead to catastrophic political consequences.

Deterrence Strategy in the Nuclear Age

In 1946, upon witnessing the development of the atomic bomb, the scholar Bernard Brodie said, "Thus far the chief purpose of our military establishment has been to win wars. From now on its chief purpose must be to avert them." In other words, the objective shifted from the traditional goal of 'defense and victory' to 'deterrence'—preventing war from happening in the first place. With the development of nuclear weapons, deterrence became the core of all security policy.

The word 'deterrence' comes from the Latin 'deterrere.' 'Deterrere' is a combination of 'de,' meaning 'from' or 'away,' and 'terrere,' meaning 'to frighten.' Thus, the etymological meaning of deterrence is to frighten an adversary away from taking an unwanted action. Since the development of nuclear weapons, military strategy has been centered on deterrence.

In the evolution of nuclear weapons, second-strike capability was crucial. Second-strike capability means the ability to retaliate with nuclear weapons even after suffering a nuclear attack from an adversary. Possessing a credible second-strike capability became the core of deterrence. There are broadly two types of deterrence. For example, let's say I have a bad relationship with Mr. Lee Jong-bin and I want to hit him.

What would Mr. Lee have to do to stop me from hitting him? He could stay far away, bring strong friends to prevent me from hitting him, or arm himself. The key is to instill fear by showing that if he is hit, he can retaliate with greater force.

There are two ways to achieve deterrence through fear. First, you can threaten to inflict unacceptable damage through retaliation, preventing the action, like saying, 'If you hit me, my strong friend will destroy you.' This is called 'deterrence by punishment.' It's like threatening, 'If you hit me, I'll cut off your arm.' It's the same logic as when our parents told us, 'No homework, no allowance.'

Second, you can threaten that the adversary will not achieve their desired policy goal or will end up getting hurt more, like saying, 'Go ahead and hit me. If you do, you'll break your fist.' This is called 'deterrence by denial.' In other words, it makes the adversary's action impossible to achieve. These two concepts operate in the same way with nuclear weapons.

The former corresponds to a 'counter-value' strategy, and the latter to a 'counter-force' strategy. It is a grave mistake to harbor the vague hope that possessing nuclear weapons will enhance South Korea's security without understanding the fundamental dynamics of how they work. We must properly understand the principles of how nuclear weapons operate.

To explain how revolutionary the advent of nuclear weapons was, Professor Daniel Deudney uses the concept of the 'reciprocity of violence.' Unlike economic interdependence, this refers to how easily each side can kill the other. Humans are very vulnerable and can be killed easily, which is why we crave security. This security is achieved depending on whether mechanisms to control violence exist.

There are two types of mechanisms for controlling violence. First are natural constraints. When geographical conditions like mountain ranges or rivers make it difficult to project violence, the use of violence can be controlled. For example, it is difficult to project violence across oceans to places far from the sea. Second are man-made institutions. This involves creating control systems like governments, treaties, and constitutions to prevent the use of violence.

Depending on how well these controls function, the reciprocity of violence can be non-existent, weak, or strong. In human history, the reciprocity of violence has never been completely absent. With the development of weapon systems, the reciprocity of violence has progressively intensified.

In the age of Odysseus, weapons like swords, bows, and horses could only inflict damage at the city-state level. The wars of the Three Kingdoms period in Korea also had a small scale of damage compared to modern times. However, with the invention of gunpowder and the development of sailing ships, bombs, trains, and fighter jets, the scale of damage grew larger and larger.

With the development of nuclear weapons, we enter a state of immense reciprocity of violence that can surpass the city-state level. Therefore, the entire planet becomes extremely vulnerable, and without global-level institutional controls, the whole world could be at risk.

According to Professor Deudney's theory, all of humanity should have become much more vulnerable after the development of nuclear weapons. Nuclear weapons were developed in 1945, the same year we gained independence. So, can we say that the post-nuclear era has been more insecure than the pre-nuclear era?

Before the development of nuclear weapons, there were World War I and World War II. But after their development, there have been no great wars. This seems to contradict Professor Deudney's theory. Despite the increased vulnerability of the entire planet after the development of nuclear weapons, why haven't more wars occurred? Why do we call the period from 1945 to 1990 the 'Cold War'?

Professor Robert Jervis posed this question and called it the 'nuclear revolution.' He said, 'An explosive increase in quantity entails a qualitative change.' The core of the nuclear revolution is as follows.

When both sides possess a mutual second-strike capability, they share a mutual vulnerability. This creates an effect similar to holding each other hostage, rendering military victory meaningless. As long as the certainty of retaliation holds, political victory and peace become the core logic. When both sides possess a counter-value strike capability with nuclear weapons, 'Mutual Assured Destruction (MAD)' occurs, which enhances strategic stability.

This operates on the same logic as the exchange of hostages between great empires in the past. If I send my son and daughter as hostages to an enemy nation, I must be prepared to sacrifice my own children to wage war against that country, which in itself creates deterrence. Nuclear stability works in the same way. 'Mutual Assured Destruction (MAD)' means that each side can assuredly annihilate the other.

Mutual Assured Destruction works when both sides have a second-strike capability. This means that even if one side launches a nuclear attack, if the attacked side can retaliate with its own nuclear weapons, the attacker will suffer the same level of damage it inflicted. The key to second-strike capability is credibility. That is, the belief that 'If I am hit with a nuclear attack, I will definitely retaliate with nuclear weapons.'

However, from a rational standpoint, retaliating with nuclear weapons after being attacked is difficult to explain. President Eisenhower himself said that waging a nuclear war would be more terrible than winning one. Since surviving a nuclear attack means facing a horrific future, what good would it do to launch a nuclear attack on the adversary in return? Therefore, rationally, the credibility of nuclear retaliation is diminished.

To solve this problem, states have resorted to appeals to emotion or to automated systems, like Russia's (formerly the Soviet Union's) 'Dead Hand' program, which ensures automatic nuclear retaliation in the event of an attack. This is designed to operate as an automated process, devoid of human thought, calculation, or emotion.

It's not just about being able to retaliate; one must know what constitutes 'unacceptable damage'—violence equivalent to cutting off the adversary's arm. The United States calculated that the death of 25% of the Soviet population and the destruction of 50% of its industry would constitute an unacceptable blow. If attacking an adversary with nuclear weapons means sacrificing over 25% of my own population and over 50% of my industry in a retaliatory strike, then a threat like 'If you destroy New York, I will destroy Moscow' becomes possible. What happens then? Each side holds the other's

population hostage. The very act of using nuclear weapons leads to one's own destruction, making their use impossible. This idea of unconditionally suffering unacceptable damage operates on the same logic as a mutual hostage exchange. Stability comes from a mutual forfeiture of the security of each other's populations. This makes guaranteeing a second-strike capability the central issue. What is the most important thing for this deterrence to work?

The nuclear weapons must survive. You must protect your nuclear weapons from an adversary's nuclear attack to maintain a second-strike capability. If your nuclear weapons are destroyed, you lose your second-strike capability. Therefore, you must protect the weapons and give up on protecting the people. The 'bad' weapons become not those that kill more people, but those that destroy other weapons. The first arms control agreement between the US and the Soviet Union was the Anti-Ballistic Missile (ABM) Treaty, which banned widespread missile defense systems. Missile defense may seem defensive, but deploying it negates the adversary's second-strike capability, causing this entire mechanism to break down. The US and the Soviet Union shared this understanding.

If you want to have nuclear weapons, how many should you have? Ten? A hundred? What is the basis for the claim that 100 warheads would ultimately guarantee South Korea's security? Does having 100 warheads guarantee a second-strike capability? No, it does not.

Having just one is even more dangerous. Recently, there have been arguments that since nuclear armament is difficult, it is necessary to possess nuclear latency. But this is like an invitation to 'hit me first.' You could be attacked before you even acquire the weapons, making your situation far more vulnerable. It might be different if a thousand nuclear weapons suddenly appeared, but the development process is extremely vulnerable.

The Logic of Nuclear Stability and Mutual Hostage Exchange

When North Korea first developed its nuclear bomb, it also faced the threat of a US surgical strike. It was President Carter who prevented this, and during the Clinton administration, a surgical strike on the Yongbyon facility was also considered. Thus, a very strange mechanism exists that actually invites an attack. This form of deterrence, based on a mutual hostage exchange, is called a 'counter-value' strategy.

A counter-value strategy maintains peace by having both sides put what they value on the line and threatening to destroy it. The opposite concept is a 'counter-force' strategy, which involves threatening to destroy the adversary's weapons. This enables a preemptive attack, which is called a 'first-strike capability.' A first-strike capability means the ability to negate an adversary's second strike. Possessing a first-strike capability creates a much more unstable situation, as it tempts a country to attack first without fear of retaliation. This is why the first agreement between the US and the Soviet Union was to forgo mutual missile defense.

Nuclear stability comes not from protecting the lives of the people, but from keeping the weapons safe—that is, from the survivability of the nuclear arsenal. This is the core of nuclear logic. There are three main ways to protect the survivability of nuclear weapons. First is to place them in underground facilities and protect them robustly. When Iran protected its missiles inside a granite mountain base 1,500 meters deep, the US only destroyed the entrance, leaving the missiles usable again. Thus, hardening silos is important. Second is to make their location unknown. If you can destroy all of an adversary's nuclear weapons, you become a first-striker. How do you prevent that?

You hide them. The most common method is to constantly move the warheads. This is a method North Korea often uses; moving nuclear weapons on trains makes them difficult to detect, identify, and strike. But the ultimate form of concealment is the nuclear submarine. Why? Do you know the difference between a diesel submarine and a nuclear submarine?

Diesel submarines run on batteries, so they are much quieter. However, due to battery limitations, they must surface to recharge after operating for a certain period. At that point, the submarine becomes extremely vulnerable. In contrast, a nuclear submarine can theoretically stay submerged indefinitely.

There was a movie where South Korea was devastated in a war with Japan, but a surviving South Korean nuclear submarine prepared to turn Tokyo into a sea of fire. Even if a nuclear-powered submarine devastates enemy territory, it is difficult to destroy because it operates in the vast and concealing environment of the ocean.

A nuclear submarine guarantees the ultimate second-strike capability. This was the first impetus for arms control negotiations between the US and the Soviet Union. Equipping a nuclear-powered submarine with nuclear warheads makes it perfect, but unfortunately, we currently do not have nuclear weapons. A more brutish method is to simply have a lot of them. Would 100 be enough? The US and the Soviet Union competed to see who could build more nuclear warheads. The numbers grew to 5,000, 6,000, even 7,000. It costs an enormous amount of money.

If we were to pursue nuclear armament, we would have to keep building warheads until we reached an agreement with North Korea on the number of nuclear weapons needed to feel secure. This means entering a nuclear arms race until both sides recognize that they have reached a point where they can no longer negate the other's nuclear capability. China is currently in this situation.

China was originally content with 200 warheads, but it is now increasing its arsenal by 100 per year and has reached 600. All of this stems from the need to preserve a nuclear retaliatory capability. So, why does North Korea say it will not give up its nuclear weapons? North Korea's 2022 nuclear force policy made South Korea feel extremely vulnerable. The reason former President Trump was willing to meet Chairman Kim Jong Un, even reducing US-South Korea military exercises, was for new negotiations. What were Chairman Kim's conditions for meeting with the US?

There were three conditions. First, recognition as a nuclear-weapon state. Second, the withdrawal of the US's hostile policy toward North Korea. Third, South Korea should not be involved. They are demanding that the US first accept the reality of their nuclear status, even though that status is what should be negotiated. The conditions for meeting are themselves very demanding.

Kim Jong Un is full of confidence, judging that the current balance is in his favor. The catalyst for this confidence was North Korea's 2022 law on nuclear force policy. Whereas the first nuclear force policy law in 2013 stated that nuclear weapons were for deterring the threat from US forces, the 2022 law specifies that North Korea can use nuclear weapons preemptively under five conditions.

Which of the five conditions is the most threatening? Conditions 1, 2, 3, and 5 concern the use of nuclear weapons as a deterrent—that is, in situations where the country's survival is threatened. Condition 4, however, outlines the use of nuclear weapons as a means of warfare, intended to prevent the escalation and protraction of a conflict and to seize the initiative. A year later, Kim Yo Jong specified that nuclear weapons would be used to "seize the initiative in the early stages of a war, destroy the enemy's will to fight, prevent a protracted war, and preserve North Korea's military strength."

Changes in North Korea's Nuclear Strategy and Its Vulnerabilities

This means North Korea intends to be the first to use nuclear weapons if a war breaks out. It is a highly threatening nuclear strategy unlike that of any other country in the world. North Korea is striving to acquire nuclear capabilities commensurate with this strategy. At the 9th Party Congress earlier this year, it mentioned improving its nuclear weapons operating system, modernizing conventional forces, equipping the navy with nuclear weapons, and developing special strategic assets as secret weapons, such as ICBMs, AI-based unmanned integrated attack systems, and electronic warfare systems. Why has North Korea adopted such an aggressive nuclear doctrine?

Ever since this was developed, I have been contemplating how to evacuate my family if war breaks out. The disaster alert messages you receive may not feel realistic, but this could happen at any moment, as North Korea would start a war by dropping a nuclear warhead first. Yet, we have never conducted military drills for such a scenario.

We need to consider this point. In fact, ever since I began developing AI, I have constantly worried about how to evacuate my family if a war breaks out. You probably don't give it much thought. The disaster alert messages you're receiving may seem very surreal, but a real disaster could happen at any time. To be precise, the moment North Korea decides to start a war, it will begin by dropping a nuclear warhead first. Yet, we have never conducted military drills for such a scenario.

We have never contemplated a war that starts with us being hit by a nuclear warhead. This is a very serious change, so why is North Korea taking such an extremely offensive posture? In truth, this stems from its vulnerability. When I refer to deployed ballistic missiles, I mean warheads that are mated to their missiles. The United States is estimated to have about 1,770 such deployed systems. North Korea does not; its warheads are currently separated from its missiles. The U.S. has a total of about 3,700 ballistic missiles, or over 5,000 including those slated for retirement. North Korea has between 40 and 50 warheads—though recently, President Trump clearly stated the number is 57. So let's say 57. In terms of delivery systems, the U.S. has a nuclear triad of strategic bombers, missiles (ICBMs, SRBMs), and submarine-launched ballistic missiles (SLBMs). North Korea only really has missiles. And the missile defense system (GMD), well...

Have you ever seen 'House of Dynamite'? Yes, it's the missile from that movie. It's the primary missile system for intercepting threats in space. We also have the Patriot system. North Korea is extremely vulnerable. During the Kim Jong Il era, whenever South Korea and the U.S. conducted military exercises or flew advanced fighter jets, Kim would stay out of sight out of fear. In fact, when we deploy our F-35s, North Korea lacks the capability to detect and strike them. That is why South Korea considers its air force a crucial component of its three-axis system. North Korea is purchasing weapons from Russia to supplement its capabilities, but it remains in a very vulnerable state.

The combined ROK-U.S. forces have the capacity to overwhelm North Korea's nuclear capabilities. So, what about its second-strike capability? It is highly questionable whether North Korea could conduct a second strike with only about 57 warheads. Are 57 warheads difficult to detect, or easy? They are relatively easy. China feels insecure with 200 warheads despite its vast territory; with only 57, we can certainly identify and strike them in advance. As the alliance's surveillance, precision-strike, and missile defense capabilities become more sophisticated, North Korea's nuclear capability becomes increasingly meaningless.

This is why, starting in 2022, North Korea began making very strong threats to "annihilate the Republic of Korea." It used to be a "sea of fire in Seoul," but now it's the "annihilation of the Republic of Korea." When did this start? From 2022. What happened in 2022? There was the 8th Congress of the Workers' Party of Korea in 2022, and the Washington Declaration in 2023. What did it say? That if North Korea were to use nuclear weapons, its regime would not survive. It spelled out the end of the Kim Jong Un regime. In response, Kim Jong Un talks about the "annihilation of the Republic of Korea." So now, both sides are talking about annihilating each other...

They are each targeting what the other values. We have put the end of the Kim Jong Un regime—Kim Jong Un's life—on the line, while North Korea has put the annihilation of the Republic of Korea—the lives of all South Korean citizens—on the scales. They have become mutual hostages. But North Korea is changing. It is raising the stakes, telling South Korea to stay out, that it cannot be a partner or a counterpart for dialogue. This is North Korea's way of saying, "The balance of terror is broken, and our hostage is no longer secure."

So, a very serious change is underway. North Korea started saying these things from December 2023. Let me give you one representative example. In its five-year defense plan following the 9th Party Congress, North Korea has always emphasized three primary targets for a preemptive strike. First, major command centers. Where are ours? At Gyeryongdae, Yongsan, and other key headquarters. Second, ports, to strike where U.S. reinforcements could arrive. Third, airfields. This last one is very frightening. What we emphasize as part of our three-axis system is our ability to fly our F-35s and launch air-to-ground missiles to bring about the end of the Kim Jong Un regime. But what is North Korea saying? 'We'll just strike your airfields first with our long-range artillery. Then your planes can't even take off, can they?'

The Advent of AI and the Revolution in the Speed of Warfare

It's a situation where both sides are constantly making moves against each other, like a chess game. From North Korea's perspective, the calculation is, 'We still hold you hostage, but it seems you can no longer hold ours. You can no longer threaten Kim Jong Un's life.' This puts us in a difficult position. And that is where AI comes in. AI took a long time to emerge... oh, I'm running over time. I need to wrap this up, but I'm talking too much. Let me explain what changes when AI enters the picture. As I said before, it's not the scenario you imagine from movies like *The Terminator* or *The Matrix*. What did I say it was? It's about making the entire process, from detection to strike assessment, extremely fast. In the past, we would acquire information from satellites and drones, along with signals intelligence, cyber intelligence, and open-source intelligence. We often couldn't even process the open-source intelligence.

How can anyone process all of social media? There's a tremendous amount of data. Human commanders used to integrate all this information from battlefield sensors and radars, then deliberate for days, or even months, to create target lists and formulate strike plans. That was the standard process. But how long does it take now? During the recent conflict involving Iran, strikes were being carried out every few minutes. That's how incredibly fast the process from detection to assessment has become. There has long been debate about the revolution in speed that AI would bring.

In the past, when I gave lectures or talks on this, I could only explain it in abstract terms. For instance, when ChatGPT 3.5 came out, theoretical estimates based on its parameters suggested a processing speed of 20 quadrillion operations per second. That's not trillion, but quadrillion. To put that in perspective, think about solving a probability problem in a math textbook. What do you do when you can't figure out the formula? You start counting every single possible case. Now, imagine doing that 20 quadrillion times per second. It's an incredible speed, isn't it?

That's what I imagined. 'Wow, it could be that fast. When AI does it, that's the AI speed revolution.' That's what I thought. But recently, experimental results have begun to emerge. A very interesting paper I saw this year from a Greek research team took ten real-life international conflicts and obscured all identifying information that could be easily Googled. However, they used the actual battlefield data—the maps, the situations, etc. Then, they pitted human captains, human colonels, and AI models against each other in a competition to devise independent operational plans.

So, the captains. How long do you think it took them to create an operational plan? When I ask captains I know, they usually say it would take about a week. But in the experiment—perhaps the participants were exceptionally competent—how long did it take? The results show an average of about three hours. Now, what about the colonels? They would be faster than the captains, right? How long did they take? Just take a wild guess. Anyone... yes.

>> Two hours. They were even better than that. Right, they're colonels, they have experience. How long did it take? The results show about 58 minutes for the skilled ones. So, the difference between three hours and 58 minutes is already huge. Then they gave the task to DeepSeek, ChatGPT, and Claude. These were older models, like version 4.5. ChatGPT and DeepSeek showed similar results. How many minutes do you think it took? Oops, I already said 'minutes,' spoiling the surprise. How many minutes? One minute? Wow, you rate them highly. It took about four to five minutes.

So, an operational plan that took a captain three hours and a colonel nearly an hour was completed by ChatGPT in five minutes. The most shocking result was from Claude. How long did Claude take? 56 seconds. This shows just how fast these systems are. And as you may know, in the recent assassination of a senior military official, the target was reportedly identified by hacking traffic light CCTVs and cell phones, and by aggregating and integrating all available information from social media. We have already entered this era.

AI-Based Kill Chain and Counter-Force Strategy

So, what's the problem here? Earlier, what did I say was the core principle of nuclear weapons? Survivability. If a high-ranking military leader's location can be found that quickly, how long would it take to locate a nuclear warhead? It's the same logic. Concealed nuclear assets can be identified more easily and struck more easily. What happens then? The survivability of nuclear weapons plummets. So, when targets can be found easily and the 'sensor-to-shooter' cycle—the time from detection to strike—becomes extremely short, what suddenly appears to be achievable?

You start to think, 'Why obsess over counter-value strategies? I can just execute a counter-force strike.' What do we call that? The Kill Chain. Why not just destroy the adversary's missiles before they're even launched? Why even bother with missile defense (MD)? If you can identify and strike first, that becomes the logical path. And what's more important here is that it doesn't matter whether AI actually possesses such a capability or not. What matters? You only need to instill that fear in the adversary. If I were an AI military expert, I could threaten them, saying 'I'm watching your weapons, I know whether you will press the button or not...''

and I'll either kill you or destroy your missile launchers before you can.' That's a threat you can make. What is this? This is deterrence by denial. You are making it impossible for the adversary's nuclear operation to succeed. This isn't about cutting off their arm; it's about making sure they can't even press the button. 'I will prevent your missile from launching. I will deny your capability.' This kind of deterrence starts to seem possible. Sounds good, doesn't it? Our happiness is gone, but isn't this great? One might think that.

But why do I say this is frightening? Let me explain. I've already mentioned it. When the focus is on counter-value, there's mutual restraint. But what happens when the focus shifts to counter-force? You start to believe you have overwhelming superiority over the adversary. And what about the adversary? How will they feel? They will think they are at an overwhelming disadvantage. What kind of dynamic does that create? That's what I've been explaining. Previously, to conceptually distinguish these, it was said there are four ways to use AI. An academic named Rusenko explained it as Mosaic, Centaur, Minotaur, and AI Commander. It's about whether you use AI tactically or strategically. Using it strategically means relying on AI for all decisions leading to victory in a war. Tactical use is just about how to strike effectively on the battlefield.

That's how AI is used. And how much control will be exerted over the AI? Will humans maintain full control, or will humans be controlled by AI? If AI is allowed to make its own judgments and humans simply execute its commands... So, it's interesting: what does a Centaur look like? The top half is human, the bottom half is a horse. Haven't you seen one? They appear in fantasy movies. The head is human, but the execution is done by a machine. It's used that way, tactically. And what is a Minotaur? The Minotaur of King Minos... the head, the decision-making, is AI, and the body, the execution, is human. Mosaic warfare means humans control everything strategically, and AI is used only at the tactical level. An AI Commander is just letting the AI do whatever it wants. U.S. military officers were asked about this. Which method of AI utilization do you prefer?

What do you think came in first place? Mosaic was number one. Second place... >> You'd think Centaur, right? It was Minotaur. This means commanders are very afraid of uncertainty at the tactical level. They would rather rely on a machine. 'I'll make the strategic decisions about the war, but we can still use AI tactically.' This was most extremely demonstrated in the Gaza war, in the way Israel uses AI. Autonomous weapon systems are always a hot topic. A program called Lavender... As I mentioned earlier, how long did it take the U.S. to strike Iran? Four minutes.

What do you think was number one? Mosaic came in first. Was number two Centaur? No, it was Minotaur. This shows that commanders are very fearful of uncertainty at the tactical level. They'd rather rely on a machine. 'I will make the decisions about the war, but we can still use AI tactically.' This was most starkly revealed in the Gaza war, in Israel's use of AI. The debate over autonomous weapons systems was always heated. I mentioned the program called Lavender earlier; how long did I say it took the US to strike Iran? Four minutes.

How long does Lavender take? It's a drone. An autonomous drone takes 20 seconds. 20 seconds. It took 20 seconds to strike and kill a single target. 20 seconds. So it flies around, calculating the probability of someone being a terrorist. If it's over 88%, what does it do? It kills them. Automatically. Above a certain percentage, like 90-something percent, it just kills. When it operates like that, isn't a Minotaur system fine at the tactical level? The efficiency of war increases. That's the kind of change people thought would happen on the battlefield, and it was most radically demonstrated in the war with Iran.

They were hitting one target per [few minutes], and the speed of war became incredibly fast. That's why the Mina mis-strike became such an issue. Because decisions were being made so quickly, with one target being struck every four minutes, the CENTCOM commander explained it as 'human in the loop.' 'Humans are involved in every decision. Don't worry. We are in control,' he said. But the question is, when you're taking out a target every four minutes, what does 'human in the loop' even mean? Of course, it's better than 20 seconds. But if you're making a decision every four

minutes, how carefully can you deliberate? It's a speed that inevitably raises serious questions. So because this is all happening so fast, human control is progressively being lost. Experts say humans are being relegated to 'rubber stamps.' Maybe think of it this way: when you go to a website and it asks about cookies, how long does it take you to click 'Agree'? For me, it takes about one second. I mean, I need to get to the site quickly, and the pop-up is so annoying.

That's what's happening. The AI processes that much data, performing 20 quadrillion calculations a second to reach a conclusion, and it's easy for a human to think, 'Do I really need to supervise this? Do I need to check it?' That's how absurd incidents like the Mina mis-strike happen. The target had already changed years ago, but the information wasn't updated, and the human didn't cross-check it. This was revealed in the recent Iran conflict, and most seriously, they say this is how Palantir was used. It would take too long to explain in detail, but this five-part system is how the Maven Smart System reportedly operated in this case. It all proceeds so quickly that it's difficult for a human commander to intervene and say something is wrong.

Four Consequences of AI Warfare

It has become an era where it's very difficult to say something is wrong. I call this the 'paradox of speed.' Because it's too fast. To explain it briefly, a system called Apollo transmits the data it collects. They say it can operate even if the internet is down because it's equipped with a very small AI. So, it gathers information in the foundry. This is then structured into a workflow, and then the Ontology—this is the core of Palantir's technology—creates a digital twin. It takes all the gathered data, gives it meaning, and builds a virtual world identical to the real situation. Based on that, there's the AIP, which was initially in the cloud and is now in ChatGPT. It allows human commanders to ask questions and understand the information in the Ontology through conversation. It translates it into natural language. There's AIP to help humans understand the Ontology data that only machines can comprehend, and on top of that

is the Gotham system. Gotham displays common situational pictures and allows commanders to issue orders like 'Strike this,' or 'Move that unit here.' It operated in this manner. The speed was incredibly fast, the war was conducted that way, and as a result, four phenomena occurred. The first, as I mentioned, is the reduction of deliberation time. Is this target worth hitting? What will be the effects after we hit it? What is that? That's strategic judgment. There's no time for that strategic judgment anymore.

Humans are just approving things like civil servants. One every four minutes, boom, boom. So what happens? Operations are carried out haphazardly, without a concrete picture of how to win the war. And the result? The war still isn't over. What was the most defining characteristic of President Trump's approach? This war has no objective. The objective keeps changing. The goal you're told yesterday is different from the goal today, and it will be different again tomorrow.

So, if you ask specifically how victory will be achieved, what the definition of victory is, there's no concept of it. So what are they doing? They're just smashing things indiscriminately. Very precisely, very quickly, one target every four minutes. What happens then? There's no time for deliberation. And humans fall into an automation bias. Because things happen so quickly, they think, 'It takes me three hours to plan maybe one operation, but you bring me a new one every five or six seconds.' So they come to trust the AI's judgment, the data it saw, the processing speed it used. They stop criticizing it. When these two things operate simultaneously, what problem arises?

AI has what's called a black box problem, though I don't have time to explain it in depth. As I said earlier, it cannot explain the conclusions it reaches after performing 20 quadrillion calculations per second. If you ask it, 'Why did you reach this conclusion?' it doesn't know. The most it can do is say what it did: 'I tried it. I counted.' This is called induction. What it basically does is provide the most plausible explanation for a situation based on the theories it holds. For example, if the ground is wet, it concludes it must have rained. Why?

Because it has a theory that when it rains, the ground gets wet. But does it have to rain for the ground to be wet? No. Someone could have sprayed water, or an underground water pipe could have burst. But it doesn't consider those possibilities. AI is a token-based prediction machine, which means it's a machine that only comes up with the most plausible explanation. It's not a machine that knows the correct answer. So what's the problem? If there's a lot of data, the judgment is more likely to be accurate. But if there isn't enough data, or if bad data is fed in, it makes wrong judgments, and humans, with no time for deliberation and stuck in automation bias, cannot catch the error.

And then, accountability becomes unclear. 'Why did you mis-strike Mina?' We still don't know whose fault it was. So people start saying, 'The AI told me to do it.' So, who made the AI? 'Someone at Google, I hear.' So if it's your fault, you say, 'It was the data's fault.' Responsibility keeps getting passed around. Whose fault is it? Who is at fault? Who should be held responsible for the civilian casualties? These problems keep emerging. And something I haven't written here is another critical problem of AI warfare revealed this time: you run out of ammunition too quickly. The pace is too fast. When you use it like that, what happens? You never imagined you'd use up ammunition this fast.

A CSIS report with estimates came out, and you've probably seen it in the media recently, but today an economic war was declared. Why can't the U.S. strike again? Didn't they say they would destroy civilization? Why can't they say that now? They have no missiles. To what extent? They've used up all their precision missiles, the most advanced precision-strike missiles. They're gone. THAAD, we say THAAD is important, right? How much did they use? They say it's depleted. Two-thirds of the inventory. Not fighting Iran, not fighting China, not North Korea, but fighting with Iran, and they used one-third. With the one-third that's left, they have to deter China until new ones are deployed.

They have to deter North Korea. They have to deter Russia. Fifty percent of the Patriot missiles are gone. In my paper, I call this the 'window of vulnerability.' A window of vulnerability has opened. What's most needed to make those missiles again? Rare earth elements. And who produces a lot of those? China has them. China's leverage in this dilemma has grown enormously. The United States is now in a very vulnerable state. It cannot start another war. That's the situation. So these fundamental problems have arisen. In conclusion, no matter how excellent your tactical efficiency is and even if it's proven, if the humans making strategic judgments don't make the right calls, the war gets completely tangled. So what was the content of M? We'll give you economic sanctions. We'll pay for all the reconstruction costs. Right? Who won? They just have to open the entire Strait of Hormuz. So it's becoming very strange.

Many military experts are now saying that for the United States, this will be the worst defeat since the Vietnam War. This is the new face of AI warfare. On top of this, at least Iran didn't have nuclear weapons. What would happen if such a war were waged against a nuclear-armed state? That's a problem. Of course, let's assume we do everything better than this time. Let's say from the start we set our strategic goals well and hit exactly what needs to be hit, responding efficiently so that North Korea cannot launch a counter-attack. Even then, there are problems. First, the distance between North and South Korea is too short. It's five minutes. Five minutes. It doesn't even take 10 minutes for a missile to arrive. Actually, 10 minutes was generous. It's usually five minutes. What can you even think about in five minutes? We'd be even more reliant on AI, wouldn't we? If we had a system like Palantir, a smart system, we would rely on it even more.

The Advent of Precision Strikes and the Transparency Revolution

We would have no choice but to rely on it. And with the DMZ in between, too many troops are concentrated there. It's a powder keg situation where escalation could happen very easily. The center is too close. But North Korea's nuclear arsenal is still small, 57 warheads. What is that telling us? President Trump is saying, 'I know exactly how many nuclear warheads you have, down to the single digits.' What does that mean? It means he knows where they all are. This is what an academic named Lieber and Press called... In a paper they wrote in 2010, as I mentioned earlier...

I said there's counter-value and counter-force, right? They published a paper called 'The New Era of Counterforce.' What do they argue in it? They say that weapons are becoming increasingly precise. In the past, when an SLBM was fired, its accuracy was only about 10%. But it didn't really matter. Why? Because it was a nuclear warhead. Even a near miss could kill a lot of people. Now, the accuracy of those missiles exceeds 90%. ICBMs have an accuracy of 96-97%.

The accuracy is so high, we can strike with extreme precision. Plus, there's the transparency revolution. The transparency revolution means we can now see where the adversary is and where their weapon systems are as clearly as looking at the palm of your hand. And they use the Korean Peninsula as an example. What are they talking about? Military reconnaissance satellites. They calculate that with six U.S. military reconnaissance satellites, there is a 91-minute surveillance gap. That means there's a 91-minute gap between one satellite passing over and the next one arriving. Does that gap seem small or large to you?

Right? In 90 minutes, a system like a TEL could move away, and how would you know where it is? It would be somewhere around here. How far can it go in 90 minutes? Well, if you move it by train, I'm not sure, but it took me an hour and 20 minutes to get from Nonsan to Seoul today. Of course, that was by KTX, but it's a considerable distance. So 91 minutes is a very large surveillance gap. But what they say is that if you combine the satellites of NATO, Japan, and South Korea with those of the U.S., you get 60 satellites. With 60 satellites, how many minutes would the surveillance gap be?

it comes out to 27 minutes. What happens when there are 1,000 of them? It shrinks to a matter of minutes. That's what Elon Musk is talking about with low-Earth orbit. If we know where everything is like that, we know where all 57 of those small nuclear weapons are. So what can we do? If we consider an extreme case, no matter if they're in a secret base or an underground facility, we could just use our own nuclear forces to strike them and achieve a first strike. So, as things stand, even though North Korea's arsenal is small and mobile, it is still very vulnerable. So, as the Kill Chain of the South Korean or ROK-U.S. forces becomes more sophisticated, what happens from North Korea's perspective?

If they don't use their nukes quickly, there's no point in having them. This is what they call 'use it or lose it.' They become plagued by the fear that if they don't press the nuclear button early on, having these weapons might be completely meaningless. This happens when the precision revolution, the transparency revolution, and the AI speed revolution are combined with the existing strike capabilities of South Korea and the U.S. North Korea would have no choice but to feel immense fear.

From our perspective, it's defensive. We don't have nuclear weapons, but North Korea is a threat, so we have no choice but to enhance our Kill Chain and decapitation strike capabilities. Paradoxically, however, once those capabilities are perfected, North Korea finds itself in an extremely vulnerable state. This could actually provoke North Korea into a preemptive nuclear strike. Our defensive modernization could be perceived as preparation for a nuclear attack, creating a spiral model problem where mutual military actions are interpreted with offensive intent. The first risk is that South Korea or the U.S. becomes overly optimistic about its counter-force strike capabilities.

As in the Iran case, we might attack first, believing we can eliminate Iran's nuclear capabilities with a preemptive strike. Second, North Korea, under pressure that 'if I don't shoot first, there was no point in developing nuclear weapons,' or if it judges that a ROK-U.S. joint attack is imminent, will have an incentive to use its nuclear weapons sooner. Third is the fact that North Korea employs a combined nuclear-conventional missile strategy. A KN-23 missile can carry either a conventional or a nuclear warhead. If launched at a high angle, it's impossible to tell which it is.

North Korea integrates its conventional and nuclear strike capabilities, having a single artillery unit manage both conventional and nuclear weapons simultaneously. From our perspective, we can't tell if it's nuclear or not. Since North Korea doesn't have many nuclear weapons, it seeks to maximize the coercive effect by mixing them. From our side, the logic becomes 'since we don't know if it's conventional or nuclear, let's just attack first and see.' From North Korea's perspective, it will consider an attack to have already begun and will be under 'use-it-or-lose-it' pressure. Fourth is the case where AI advises a first strike, judging that a nuclear attack is imminent during a U.S.-China war.

Have you ever questioned an AI's answer? According to a recent study, only six out of 1,000 high school students questioned an AI's answer when solving a math problem. Humans tend to be overconfident that AI is better than them because it processes more data faster. This overconfidence can reinforce all four of these pathways.

The Dilemma of AI Warfare and South Korea's Challenges

The fourth pathway reinforces the first, second, and third. We are moving toward an environment where war can break out much more easily. It's a problem whether we adopt AI or not. This presents a dilemma for South Korea. It raises questions about how we integrate AI into the Kill Chain and KMPR. I support enhancing KMPR capabilities, but strengthening the Kill Chain could subject North Korea to a 'use it or lose it' fear.

Restoring Human Control and Counter-Value Strategy

The capabilities for the Kill Chain and decapitation strikes are similar, which is a major physical problem. First, we must restore meaningful human control. Even if we develop our AI military intelligence, we need a system that allows for deliberation and suspension of action. Second, more important than the Kill Chain is the certainty of retaliation.

U.S.-Soviet nuclear arms reduction began when they confirmed the despair and frustration of being unable to stop each other's second strike. Since a nuclear submarine cannot eliminate an opposing nuclear submarine, it is meaningless to build more and more nuclear warheads. This sense of despair is what actually pushed the U.S. and the Soviet Union to talk.

The important thing is for us to also focus on the counter-value domain, not the counter-force domain. We must threaten the Kim Jong Un regime that 'even if you annihilate the Republic of Korea, you will certainly die.' If we can credibly threaten that South Korea's nuclear submarines or F-35 fleet will kill Kim Jong Un, we can restore this perception. AI can be utilized when enhancing our counter-value strategy.

We must consider a system that can operate even in an extreme situation where all internet is down, such as if North Korea detonates an EMP bomb to disable South Korea's electronics before commencing an operation. We should use AI to enhance the resilience of our AI-based command and control system. If we can credibly convey the threat that the Kim Jong Un regime will end even if it annihilates the Republic of Korea, the mutual vulnerability and the status of being hostages will be restored.

AI Utilization and Restoring Mutual Vulnerability

The U.S. and South Korea must discuss and consider what kind of force enhancement on our part would cause North Korea to feel extreme fear. This is a point that should be discussed at the NCG. The U.S. and South Korea must work together to prioritize which effects need to be intentionally avoided and which need to be strengthened. We must avoid falling into a spiral model and use AI to secure more time for judgment. We need to think about how we use AI. A recent paper also states that the possibility of mutually assured destruction (MAD) lies in the certainty of retaliation.

It is necessary to develop military capabilities in a way that enhances the certainty of retaliation, and the military applications of AI must be considered carefully. A counter-value, rather than counter-force, strategy would help prevent the increased vulnerabilities stemming from AI adoption and avert a rapid collapse of strategic stability.