AI and nuclear doomsday

0:00To watch the new Catherine Bigelow movie, A House of Dynamite, on Netflix, is to immediately start freaking out about a nuclear war.. We've already lost one American city today. How many more do you want to risk? What kind of fucking question is that? Some enemy, it could be Russia, it could be North Korea, launches a nuclear warhead at Chicago from the Pacific Ocean. It's okay. The U.S. has interceptors. They blast up into the air and they knock the nukes out. That is actually true in real life. That's our nuclear defense system. Except in the movie, the interceptors fail. Except it's just a movie.

0:40Except a few days before it came out on Netflix, The Pentagon started freaking out and talking internally about how to calm people down. Coming up on today explained from AI to Golden Dome to Netflix's new hit, why everybody is suddenly talking about nuclear war. Weihnachtsschmaus und Silvesterer-Kletz sind gerade vorbei. Da kommt auch schon die nächste Tradition um die Ecke. Das große Opel-Angrillen, bestimmt auch bei deinem Partner vor Ort. Mit Musik, Leckereien vom Grill und exklusiven Angeboten, like the Opel Frontier already at 225€ monthly in Leasing plus Sonderzahlung. And in the showroom is also the rest of the Modell palette ready.

1:20Up to the Opel Grill on Saturday, 24 January at your own Opel Partner. More on opel.de

1:43Oh, my God. I'm Noelle King with Tony Capaccio. Tony covers the Pentagon for Bloomberg News. He's been covering defense for almost 40 years. So recently, Tony found out that in the days before A House of Dynamite dropped on Netflix, the U.S. Pentagon, of all people, started circulating a highly unusual response to the movie. So on October 16th, the Missile Defense Agency issued an internal memo, not publicly releasable, it said, underlined, to give a heads up to leaders that this movie was coming out broadly. And if there are questions about it, this is the current state of the U.S. ground-based missile defense program.

2:25The goal is to ensure leadership has situational awareness and is not surprised by the topic, which may come up in conversations or meetings. It was not a slash-and-burn memo, trash in the movie in total. It made points pretty clinically and, like, dryly that the movie's not accurate in terms of its portrayal of the success of the missile defense system. That's kind of the bottom line. And they gave four or five pages of technical insight into why they thought it was inaccurate. The fictional interceptors in the movie miss their target, and we understand this is intended to be a compelling part of the drama intended for the entertainment of the audience.

3:05But results from real-world testing tell a vastly different story. If you look at the last decade with improved warheads, our success rate is 100%. Literally, that is accurate. The track record on this thing over the last four tests indicates 100 % they did intercept the missile. It does say this, the 50-50 chance of intercepting narrative is based on our earliest prototype versions of the GBI system. When we operated with only a few early warning radars that were already decades old and a rudimentary command and control systems. We have since upgraded one of those early warning radars in California and added other upgraded early warning radars to the missile defense system.

3:51Plus, we also have added a sea-based radar in the Pacific capable of discriminating threat objects with great precision. End quote. That's what they're laying out. What they have not done a good job of lately is putting this stuff out in public.

4:09The movie describes the process of interception chillingly as trying to hit a bullet with a bullet. Once the kill vehicle separates, our mid-course intercept system has a success rate of 61%. So it's a fucking coin toss? That's what$50 billion buys us? We are talking about hitting a bullet with a bullet. Which sounds like this will work if you're damn lucky. Is that about the size of it? You're going to love this. Inside the memo, there's actually a clinical description of hitting a bullet with a bullet. But here's what I'm going to read this. Hitting a bullet with a bullet in terms of closing speeds is significantly less than hitting a missile with a missile.

4:56The highest velocity rifle rounds travel at 4 ,500 feet per second, giving closing velocity of two bullets around 2.75 kilometers per second. Missiles in space will have three to four times higher closing velocities. velocities. So I guess in layman's language, hitting a bullet with a bullet is not as difficult as hitting a missile with a missile. Okay. Hitting a bullet with a bullet is not as difficult as hitting a missile with a missile. And the whole point is we're supposed to be terrified by the idea that like hitting a bullet with a bullet is really hard. So question then to you, Tony, is we've talked on the show before about how President Trump would like to build a Golden Dome missile defense system.

5:40Once fully constructed, the Golden Dome will be capable of intercepting missiles, even if they are launched from other sides of the world, and even if they're launched from space. Where does something like that fit into this conversation about whether or not our systems that handle the nuclear threat are operating properly? So if I knew nothing about missile defense, and I'm a lay viewer of this movie, and I know a little bit about what Golden Dome is supposed to be because the president says so, I would scratch my head and say, well, if this limited defense system didn't work against one missile, maybe they got a point.

6:22Maybe it should be a broader system to protect parts of the United States. Maybe I should wait and see what they disclose. That's a logical way to look at this film. That is what I thought, truthfully, last night as I was watching this movie. Okay, maybe we need Golden Dome. The movie gives one pause in terms of, well, maybe he's got a point. The problem with his point is that he's said nothing about it beyond that it's going to be$175 billion and it should be deployed later by the end of his administration. The Pentagon said virtually squat about the configuration of it, where this$175 billion estimate came from.

7:02And the technical difficulties involved. So we're in a wait and see mode on Golden Dome. But the movie does raise a question. If this limited system doesn't work, do we really need a broader system? How rare is it for the Pentagon to sort of have a response to something that's in theaters now? We're talking incredibly rare.

7:30In 1982, the Pentagon was asked by ABC to vet their script for the famous Day After. This is a movie of a nuclear holocaust in Kansas hit by a, we're not sure where the missile came from, but a nuclear holocaust the day after a nuclear bomb. Roger, we've got 32 targets in track and 10 impacting points. I want to confirm, is this an exercise? Roger, copy. This is not an exercise. The Pentagon reviewed the film. They didn't want to give aid to the filmmaker, ABC, in the form of National Guard or military equipment because they felt the script implied that NATO started World War III. But, and I saw some of the documents, I wrote about it at the time, they actually complemented some of the film's depiction of the aftermath.

8:22We're talking, though, rarely does the Pentagon come out and blast a movie. They support a lot of films. They supported Mission Impossible, the last one, on the last two Mission Impossible's with V-22 Ospreys, with the use of the George Bush carrier. Remember in the trailer when the Secretary of Defense said, You gave him an aircraft carrier? Well, that was actually true. They gave him the George Bush carrier for Tom Cruise and the rest of the film cast. I need you to trust me one last time. So they do routinely support films. It becomes news when they don't. But Platoon, a famous one from like 25 years ago, was one of them.

9:06They didn't support that. And they didn't support Zero Dark Thirty, the famous movie that Kathryn Bigelow made about the killing of Osama bin Laden by SEAL Team 6. Bin Laden is there. And you're going to kill him for me. Zero Dark Thirty was kind of the code name for when the operation would take place in Pakistan, about about Pakistan. A number of organizations were and members of Congress were a little peeved that the Pentagon helped them or even helped vet their script. This is at a time when Secretary Robert Gates was telling the rest of the military, keep your mouth shut about the bin Laden raid.

9:51So, Tony, how might you interpret what happened here? Why did the Pentagon respond? Why did the Pentagon talk about this at all? I got to think they are concerned that the Golden Dome will be trashed because of this. There's a line in here that says, was this film made in response to the planned Golden Dome project? And it says, please direct all questions related to Golden Dome to the office of the Secretary of War. Which, so there was some sensitivity. This wasn't emphasized throughout. But I think part of it was it showed a miss. And they were worried about an erosion, probably, of U.S. confidence in the public about the system that nobody's really thinking about at the moment, but to be proactive in terms of if they're asked about the subject.

10:50Bloomberg's Tony Capaccio. Coming up, the prospect of nuclear war was scary enough. It always has been. But now we are actively adding AI to our nuclear infrastructure. How freaked out should we be?

11:13Support for the show comes from Grow Therapy. Everyone thinks they're therapists, especially in the new year, when you might be getting advice from every corner of your life, your feed, your group chat, even your barista. But here's the thing. Advice isn't therapy. Therapy is therapy. And if you're looking to start therapy with licensed professionals, Grow Therapy is there to help. Grow can connect you with thousands of independent licensed therapists across the country, offering both virtual and in-person sessions. You can search therapists by what matters most to you, like insurance, specialty, identity, or availability, and get started in as little as two days.

11:52And if something comes up, you can cancel up to 24 hours in advance at no cost. Grow says there are no subscriptions and no long-term commitments. You just pay per session. Whatever challenges you're facing, Grow Therapy is here to help. Sessions average about$21 with insurance, and some pay as little as$0, depending on their plan. Visit growththerapy.com slash explained to get started. That's growththerapy.com slash explained. growththerapy.com slash explained. Availability and coverage vary by state and insurance plan.

12:33support for today explained comes from bombas putting on a new pair of socks can instantly feel refreshing according to bombas especially according to bombas if they're bomba socks bombas make socks for just about any activity your warm merino wools your comfortable compressions your cushioned running socks and so much more high quality slippers t-shirts mudans Nisha Chital tried Bombas. Here's what she thinks. Bombas has great kid socks. I have a three-year-old. She runs around a lot indoors in her socks, and she's often slipping on the hardwood floors. But Bombas kid socks have grips on the bottom, and they're really great to prevent that slipping around on hardwood or tile floors.

13:15So we love those for her. They're definitely our preferred socks for our daughter. Bombas says that for every pair of Bombas you buy, they donate one to someone facing homelessness. Anytime you get something cozy, someone else can too. According to Bombas, you can go to bombas.com slash explain and use code explain for 20 % off. Your first purchase, that's B-O-M-B-A-S dot com slash explain. Code explain to check out.

13:40Support for today's show comes from Rubric. Many companies are deploying AI agents at this time. They automate tasks. They handle workflows. They make decisions. And here is the thing. Sometimes, according to Rubric, they mess up. They might delete the wrong file. They might make changes you didn't authorize. They might go off script. When that happens, you can get stuck trying to figure out what went wrong and how to fix it. Enter Rubric Agent Cloud. Rubrik Agent Cloud is a platform that they say lets you monitor, govern, and rewind AI agent actions. It's one platform. They say it helps you use more agents faster and without the risk.

14:22They say it runs in the background. They say it watches what's happening. They say it makes sure things stay on track. They say you can get full visibility. They say you can set guardrails so the agents don't go rogue. If your business relies on AI agents, you need the ability to monitor, govern, and rewind their actions, says Rubrik. Right now, our listeners get exclusive early access to Rubrik Agent Cloud. You can go to rubrik.com. That's R-U-B-R-I-K.com, rubrik.com.

14:59I listened to this podcast. And the guy said, this is Today Explained. I'm Josh Keating. I'm a senior correspondent at Vox. And for the last few months, I've been working on a fellowship where I'm writing a series of articles on the intersection of AI and nuclear weapons. Human beings have had fears around AI and nuclear weapons for as long as we've had AI, I would think. and you see them quite vividly sometimes in movies. The system goes online on August 4th, 1997. Human decisions are removed from strategic defense. Skynet begins to learn at a geometric rate. It becomes self-aware at 2.14 a.m.

15:44Eastern Time, August 29th. General, what you see on these screens up here is a fantasy, a computer-enhanced hallucination. Those blips are not real missiles, they're phantoms. Most recently, there's a part in A House of Dynamite where they're trying to figure out what the hell happened and whether AI is involved. The Chinese Navy has been experimenting with AI-assisted launch systems, so this could be a technical mishap. Are these movies with these fears onto something? Well, I mean, I think the interesting thing about movies when it comes to nuclear war is this is a kind of war that's never been fought.

16:22We don't have real world examples of it. there are no sort of veterans of nuclear wars other than, you know, the two bombs we dropped on Japan, which is a very different scenario. And so I think like that movies have always played a kind of outsized role in debates over nuclear weapons. You can go back to the 60s when, you know, the Strategic Air Command actually like produced its own rebuttal to Dr. Strangelove and Failsafe and movies like that. You yourself have flown profile missions over and over again but there is one area in which you've had little or no actual experience nuclear effects in the 80s the tv movie um the day after was kind of a galvanizing force for the nuclear freeze movement and president reagan apparently was like very disturbed when he watched it and it influenced his thinking on arms control with the soviet union and in a specific topic i'm looking looking at, which is AI and nuclear weapons.

17:25I think there's been a surprising number of movies that have that as the plot. And it comes up a lot in the policy debates over this. I mean, I've had people who are advocates for integrating AI into the nuclear command system saying, look, this isn't going to be Skynet. And they're referring to the computer system that takes over in the Terminator movies. and it was actually General Anthony Cotton, who's the current commander of Strategic Command, the branch of the military responsible for the nuclear weapons. He also advocates, you know, greater use of AI tools, but he referred to the 1983 movie War Games, where there's a computer system called Whopper that, you know, accidentally almost starts a nuclear war.

18:10And he said, look, you know, we're going to have more AI, but there's not going to be a Whopper in Strategic Command. So where I think it falls a little short is the fear tends to be that a super intelligent AI is going to take over our nuclear weapons and use it to wipe us out. You know, for now, that's a theoretical concern. What I think is the more real concern is that as AI gets in the more and more parts of the command and control system, I'm like, do the human beings in charge of the decisions to make nuclear weapons really understand how the AIs are working and how is it going to affect the way they make these decisions, which could be, you know, not exaggerating to say some of the most important decisions ever made in human history.

18:53Josh, answer that question. Do the human beings working on nukes understand the AI? We don't know exactly where AI is in the nuclear enterprise. Computers have been part of this from the beginning. I mean, some of the first digital computers ever developed were used during the building of the atomic bomb in the Manhattan Project. But I think people will be surprised to know how low tech the nuclear command and control system really was. Up until 2019, they were using floppy disks for their communication systems. I'm not even talking about the little plastic ones that look like your save icon. on Windows.

19:30I mean, the old 80s bendy ones, like they were using those in their command and control system. So, you know, there's reasons for that because they want these systems to be secure from outside cyber interference. So they don't want everything like hooked up to the cloud for very obvious reasons. But, you know, as there's this ongoing multi-billion dollar nuclear modernization process underway. A big part of that is updating these systems. And multiple commanders of Stratcom, including a couple I talked to, have said they think AI should be part of this. What they all say is that AI should not be in charge of making the decision as to whether we launch nuclear weapons.

20:15They think that AI can just analyze massive amounts of information and do it much faster than people can. And if you've seen House of Dynamite, one thing that movie shows really well is how quickly the president and senior advisors are going to have to make some absolutely extraordinary difficult decisions. And so AI can help provide options and help take the decisions the humans don't have to make out of their hands and leave humans in charge of making the really important ones. Okay, so that brings us to a thing that makes everybody nervous, which is AI being integrated, more integrated. You told us about some of the pros, and truthfully, an indecisive president or a president that has to make a decision real fast being helped by something, that's a very good point.

21:09What are the big arguments against getting AI and nukes in bed together? Well, I mean, part of it is like even the best AI models that we have available today are still prone to error. You know, another worry is that there could be outside interference with these systems. It could be hacking or a cyber attack or, you know, foreign governments could come up with ways to sort of seed inaccurate information into the model. There has been reporting that, you know, Russian propaganda networks are actively trying to seed disinformation into the training data used by Western consumer AI chatbots. And, you know, another is just like how people interact with these systems.

21:55I mean, there is a phenomenon that, you know, a lot of researchers pointed out called automation bias, which is just that people tend to trust to a remarkable extent the information that automated systems, the computer systems are giving them. There are abundant examples from history of times when technology has actually led to near nuclear disasters, and it's been humans who've stepped in to prevent escalation. And so there was a case in 1979 when Spigniew Brzezinski, the U.S. national security advisor, was actually woken up by a phone call in the middle of the night informing him that, you know, hundreds of missiles had just been launched from Soviet submarines off the coast of Oregon.

22:44And, you know, just before he was about to call Jimmy Carter, the president at the time, to wake him up and tell him America was under attack, there was another call that had been a false alarm. For six minutes this morning, the signals from the North American Air Defense Command headquarters indicated a nuclear attack against the United States. It was a false alarm. The cause reported this evening a computer error. A few years later, there was a very famous case in the Soviet Union. There was a colonel named Stanislav Petrov who was working in their sort of missile detection infrastructure, who was informed by the computer system that there had been a U.S.

23:24nuclear launch. And, you know, under the protocols, he was supposed to then inform his superiors who might have ordered immediate retaliation, as their doctrine would have required them to. But it turned out the system had misinterpreted sunlight reflecting off clouds as a mistill launch. And so that was a case where it's very good that Stanislav Fetrov took the decision to wait a few minutes and confirm that before he called his superiors. Yeah, it sounds, I mean, I'm listening through to those examples. And the thing I might take away, if I'm thinking about it really simplistically, is that human beings pull us back from the brink when technology screws up.

24:00It's true. And I think there's some really interesting recent tests on AI models given sort of military crisis scenarios. And they actually tend to be more hawkish than human decision makers are. You know, and we don't know exactly why that is. I mean, one thing I think about is that if we look at why we haven't fought a nuclear war, why 80 years after Hiroshima, nobody's dropped another atomic bomb, why there's never been a nuclear exchange on the battlefield, I think part of it is just how terrifying it is, how humans understand the destructive potential of these weapons and what this escalation can lead to, that there are certain steps that may have unintended consequences.

24:43And fear is a big part of it. And so, you know, from my perspective, I think we want to make sure that there's fear built into the system, that intelligences, entities that are capable of being absolutely freaked out by the destructive potential of nuclear weapons are the ones who are making the key decisions on whether to use them. It does sound like, you know, watching a house of dynamite, you can, as I did last night, vividly think perhaps we should get all of the AI out of this entirely. Just like just like let's not have any mistakes. Let's have any accidents. It sounds like what you're saying is we may have reached a point at which we are not going back in time.

25:29AI is a part of nuclear infrastructure for us, for other nations, and it is likely to be that way. Well, it's interesting. One thing one sort of advocate for more automation told me, he was like, if you don't think humans can build a trustworthy AI, then humans have no business with nuclear weapons. And, you know, I think he's kind of right. But the thing is, I think that that's a statement that people who think we should eliminate all nuclear weapons entirely would also agree with. Like, it's, you know, I think I maybe got into this worried that like AI was going to take over and take over nuclear weapons.

26:08But I realized like, maybe, but like right now I'm worried enough about like what people are going to do with nuclear weapons and the risk that, you know, it's not that AI is going to kill people with nuclear weapons. It's that AIs might make it more likely that people kill each other with nuclear weapons. And so, you know, to a degree, like, the AI is the least of our worries. Like, the real thing we should be concerned about is that, like, we have these weapons at all. And that, as I think the movie shows well, just, like, how absurd the scenario in which we'd have to decide whether or not to use them really is.

26:53Josh Keating. You can find his reporting at Vox.com. Josh's reporting on this one was supported by the Outrider Foundation and Journalism Funding Partners. Kelly Wessinger produced today's episode. Jolie Myers edited. Laura Bullard and Avishai Artsy checked the facts. And Patrick Boyd and Adrienne Lilly engineered. The Vox membership sale, you should know, is still ongoing. ad-free podcast folks, other perks too. Good on you if you've taken advantage of the sale. And thank you to the person who said they signed up for the Today Explained hosts. Our boss reads those messages, so we appreciate it.