Weeping for the Mathematicians

Okay, so we already talked for like an hour, so now I've got no voice and yelling at you. This is your idea. You wanted a prepod meeting. You wanted a meeting. Oh, I didn't want to meet you. We should never do prepod meetings. Hi, everyone. Welcome back to CheapTalk. My name is Jeff Kaplow. I'm an associate professor of government here at William & Mary. And joining me, as always, is my esteemed colleague, Marcus Holmes. Hello, Marcus. Jeff, we're recording this on October 14th, and it is still quite hot in Williamsburg. We had like a little bit of a temperature dip the last couple of weeks. And now it's like almost 90 degrees today. We're back to summer. This is kind of crazy. I don't like the hot weather. I like fall. I like Thanksgiving. I like the holidays. I just want us to get into cooler, cooler temps. That's what I'm feeling right now. Marcus, the opinions expressed on this podcast are solely our own and do not reflect the policies or positions of William & Mary, our employer. Good. I'm just glad we got that straight. We were all for a little break. We were both in the UK having a good time out there. How was your trip to the UK, Marcus? It was good. I kept on looking for you. It's a little bigger than Williamsburg. So I don't see you here either, but I was looking for you in the UK and I didn't, I didn't find you. But yeah, we were both there at the same time. I was in London and Birmingham with my good friend and co-author Nick Wheeler, friend of the pod who's been on before. And we were launching our new book, Interpersonal Diplomacy, the making of atomic bonds and international crises at his university, which is the University of Birmingham. And it was quite lovely. So the way that this works because we published with Oxford University Press, which is based, at least the version of Oxford University Press that we use is based in the UK. The book is out in the UK and it's not out here yet. So they got it. They're enjoying it. They're reading it. It's making big, big news. And there's a big splash, of course, and everybody's having a good time with it. And now here in the United States, you can't get it yet. So I was like, in this awkward situation. So I flew to the UK. We did what's called a book launch, which is basically just you talk about the book and, you know, people ask questions and you have a little reception. We sold some copies there, which is great. But now I'm like itching to do the same thing here, but the book's not out yet. So we can't really talk about the book or do a book launch because nobody can buy it yet. So I got to wait. It's going to be out in November. They say, I will say the hard copies. So Jeff, you have this experience as well. There's nothing like, like, finally touching, like, physically, the book after all is said and done. Because you work on these stupid things for so long, at least, you know, I do. It takes forever to write these things. You know, multiple revisions and edits and additions and subtractions and the proofing and the index and the whole thing. And then you finally get it. And then what happens? And I maybe you share this as well. I am terrified to, like, actually open it up and look at it because I'm going to, I'm going to find a typo. I'm going to find an error. I'm going to find something horrifying. It's going to, like, really not your name wrong. That part, I made sure to get right. If nothing else, get the name right. As long as it appears in Google Scholar, I'm happy. But it just, I'm so afraid of, like, finding an error that I can't, I can't look at. And that goes for all my books. Like, I, all my books, I have three books. But like, you know, I can't, I can't bring myself to, like, open up and look at them just because I'm afraid I'm going to find something that I don't want to find. So it's, it's like trepidacious, trepidacious trepidate with trepidation. I took the book and I kind of flipped through a couple pages just to make sure that there was words on the pages, which there appeared to be. So that's good. But yeah, it was good. It was a fun time, had a good time, and, you know, look forward to our American launch. In November, when the book is finally out here, and the book will be out just in time for holiday giving. That's right. That's like, as long as you get it out before the holidays, I think, because I think it's going to be a big gifting purchase. Well, you know, it may be Oxford, else what they're doing, which is they, they do the, the release. You know, it's kind of like movies, right? Like in the old days, they would do like New York and L.A. Got the mood to generate some buzz, right? Like a couple weeks. And then us like, please got it, you know, like a couple weeks later, right? So it's like the important people got to see it first. It's a little buzz going and then, you know, people that live in like Chicago or Williamsburg, they can watch it a couple weeks later. So like the UK is like devouring the book as we speak. They're loving it. Everything's great. And we're going to have the same experience now in a couple months. So I think it's all good. I got to say my favorite part of the new book coming out is your traveling. You're like, you're like, you're at an airport, you're at JFK on a layover. And you wander into the Hudson news, you know, and you're like, I'll get some peanuts for the or some non-allergenic snack for the, for the flight, right? And a bottle of water. And you look over the bookshelf and they're facing you is your book. I know. It's a great feeling. And it's, you know, it's even better feeling when like they're clearly word more, but a couple of them have been taken, you know, it's great. It's great. You look over. You see like the economist magazine and then you see like Stephen Pinker's latest book and then yours. You know, and it's just like this is fantastic. That makes it all worthwhile right there. And then the other thing I like to do is, you know, like you said, you want to make it appear that it's it's probably you buy all the copies. You know, it ends up costing like a thousand dollars, but you buy all of them. And that way, or the other trick is on Amazon after the book's been out for a little while, you buy all the used copies. So people have to buy new ones, thus inflating your numbers, which is, which is what you want. So the little tricks of the trade that you learned over the years. And you also Jeff were in the UK, right? So I was in the UK talking about AI governance in Oxford. It was very interesting visit, talked to a lot of interesting people about some of these challenging problems in addressing risk from from AI. And I'll talk more about that. I'm sure on this podcast. Before we get into AI stuff, I do want to talk a little bit about AI stuff today. Before we get into it, I want to talk about Russia Ukraine, because stuff's happening in that conflict as usual. We had some big news of last week. So President Trump has decided to basically import directly Russian diesel fuel. I, you know, the US, as you may know, markets having some high energy costs here. And if you ask the Trump administration, what is causing those high energy costs? They will say, I don't know, Ukrainian attacks on Russian oil, which were not supposed to be importing anyway, I guess. But whatever. So it's clearly the issue is Russia somehow. So in order to solve the problem of very expensive diesel fuel, it says nothing to do with the Middle East. In order to solve this problem, we have to directly import or at least remove sanctions on Russian diesel, which will reduce the overall price of diesel generally. And then also we're going to import some directly ourselves, thus funneling US dollars directly to Russia, right? This is like, let's just write them a check kind of a situation. And so that happened. Congress people on both sides of the aisle have, I think, widely condemned this idea as a bad idea. And then at the same time, President Trump is criticizing Ukraine for having been attacking Russian energy infrastructure. President Trump said Ukraine needs to get a new leader, someone who's willing to make a deal. This comes, of course, as Russia is escalating its energy infrastructure attacks on Ukraine. There was big strike last night as we're recording, but I think maybe the largest ballistic missile attack on Kiev since the war began. Plunging much of Kiev into darkness, no power, there's these are attacks on energy infrastructure primarily as we get into the winter months and where we can start to move toward the winter months in the region. At the same time, Marcus issues with Starlink. Did you see this? No, I missed that one. What's happening with Starlink? So apparently there's a new restriction. So as of this record is not off the presses stuff. So as of this recording, StarShield access. So StarShield is the military national security version of Starlink. Starlink is the satellite internet provider from SpaceX. And that's what the military relies on the StarShield system to guide primarily to guide drones. Because drones, these drones need internet access. They need to know where they are. They need to know what they're supposed to be doing. And they need that both to attack Russian targets and to intercept Russian stuff that's coming toward Ukraine. So it's both on the offense and on the defense. And apparently StarShield has implemented a speed limit for items that are accessing its internet. So if you're going more than 100 miles an hour, so you are a drone, then you will lose access to StarShield, which is like a big deal. Yes, that's happening. That is a new thing. Nobody's willing to comment on this. In the past, when there have been issues with StarShield, SpaceX has said, well, that's a US government decision. This is a US government thing. And the US government is giving Ukraine access to this thing. When interesting layer here is that internet access through Starlink doesn't work in Russia at all. So this has no effect on Ukrainian operations in Russia. So if Ukraine were going to try to attack an oil facility in Russia, which they are not doing because they told President Trump, we're not going to attack any energy infrastructure. Hopefully Russia will abide by that too. Russia has not. But this wouldn't help. This internet access wouldn't help Ukraine target Russia because it doesn't work in Russia. So it's only for stuff on Ukrainian territory where this is useful at the same time, Marcus. This is not enough. Some worrying reports about a lack of intelligence sharing in advance of the latest Russian attacks. So we've seen some sketchy reports that at least some of these attacks came without the usual heads up from the US, which has been a major intelligence provider to Ukraine throughout the conflict with the exception of a brief period last year. U.S. Cut off! Ukrainian access to US intelligence to try to leverage Ukraine into an agreement that they didn't want. So there's some worry that since the rhetoric has kind of escalated around President Trump's support for Russia in terms of this diesel deal and criticizing Ukraine's leadership for not already reaching a deal and criticizing Ukraine for attacking Russian oil infrastructure and not condemning Russian attacks on Ukraine's energy infrastructure and maybe putting new restrictions on Internet access that also the US is sending a message by preventing intelligence sharing leading to these attacks. So one of the things that Ukraine needs to heads up on these attacks to do is so that it can scramble interceptors and try to prevent some of the damage that these missiles and drones are causing and without that intelligence that becomes much more difficult because Ukraine doesn't have its own capability to see these attacks coming far enough away to take action to stop them. Anyway, that's my sense of the state of play, optimistic. What do you make of all this? Oh boy. Well, I go back and forth Jeff on whether what we're seeing is a master stroke of genius from Trump and it has ability to negotiate a deal and get to yes and just like resolve all of these problems. Or if this is just a tremendous like blunder on the international stage because he's worried about domestic politics so I think one of the things that Unites many of the different pieces of news that you just you know gave us is that it seems like Trump is is trying to. Recast the problem in the United States with of high gas costs and high diesel costs by sort of doing a magic sort of slight of hand move to suggest that this is not actually about the Iran war. This is like the fault of Ukraine so like a lot of a lot of the things the news items are about sort of you know telling Ukraine stop attacking this energy infrastructure like you are making diesel prices more expensive like your. Ukraine's attacks on Russia is making it more difficult for for US consumers to be able to fill up their car in a cost effect the way so the so the idea I think of that Trump has is that you can get American voters and Americans more generally to kind of understand or believe that the problem with diesel is in fact Russia and not the straight of our moves. That might be a stroke of genius and that might be a masterful kind of political ploy with the midterm elections coming up in a few weeks. So the problem with that Jeff is I don't think anybody buys it I think everybody that that sort of has analyzed this including Zelensky himself who pointed out like the problem with diesel prices has nothing to do with the US the Ukrainian attacks on Russia and everything to do with the Iran war which the United States started. Like analysts and everyday citizens I don't think are buying the kind of bait and switch move that Trump is trying to make here and blame Ukraine for high diesel diesel costs. The reason this seems like a huge political blunder as well is because it goes against sort of all of the logic I think that the sanctions normally have right so like the idea with sanctions it seems to me as a you sanction somebody you put sanctions on Russia let's say in order to get them to make a concession and when they make that concession then you can say okay well now we're going to maybe like you know loosen the the sanctions a little bit give you some relief or whatever in that kind of like quid pro quo negotiation that takes place. Here it seems like Trump is basically saying we're going to give you the the sanctions relief and in fact just buy you know diesel from you without getting a concession first so that means that the United States is at the same time trying to pressure Russia into some type of agreement some type of like way to end the war. And at the same time relaxing the very pressure that they're putting on Russia to get to that agreement like it's it's sort of makes your mind sort of like hurt because it's like we're trying to do two things at once and they're moving in opposite directions even though our goal supposedly is to end the war. So the question that I have you know for thinking about this as Trump sort of like diplomatic genius or negotiating and bargaining genius like what incentive does Putin now have to make any concessions when he knows he can just wait it out and Trump will just give him you know some type of of you know relief without having to make a concession like it's it's very odd. So for that reason I think that actually this is all about domestic politics and I don't think that Trump is really thinking about you know the deal with Russia or you know sort of talking about Selensky and attacking you know Russian energy reserves or anything like that in terms of ending the Russia Ukraine war. I think this boils down as is often the case with Trump to domestic politics and he's trying to get some momentum behind the idea that you know the problem that Americans see at the gas station and the grocery store and all that kind of is not his fault but rather the fault of somebody else namely Ukraine. I think it's working but I think that's kind of the strategy that he has. Yeah maybe but you know from Ukraine's perspective it doesn't have the luxury of caring about the domestic political environment in the United States. I mean you got to protect the people of Ukraine and try to make sure that they have heat in the winter and electricity and this is a sign of what's to come for them then that's really bad and I mean you know I guess one hopeful story might be that maybe after the midterms. Trump doesn't care as much about this I guess that's a possibility. I want to say like I think there's some uncertainty about all of this. Well there's not uncertainty about the diesel thing. That's pretty clear but there is uncertainty I think about the Starshield access and like what this is really intended to do and also the intelligence sharing thing because Trump made a statement or in response to a question said we are still sharing intelligence and it sounds like there was one volume of missiles. That they didn't get intelligence for but they still got a warning about others so it's not clear to me how this could just be like a one off kind of situation and a couple of in a couple of ways. But you know this is coming today I think Zelensky is in the UK as we were recently meeting with UK leaders as we as I'm sure both you and I were meeting with UK leaders while we were there. Depends on what you mean by leader but yes right thought leaders right not actual leaders and you know it's kind of interesting to think about what Ukraine's options are given the way Russia has prosecuted this war seems the plan for the winter that that seems to be coming from Russia. Also given what the US is doing and I mean I think the European allies are very strongly in support of Ukraine they're remain committed to supporting Ukraine they are upset with US treatment of Russia and Ukraine in this conflict currently. And the question is what can be done and I think there is an extent to which Europe can supplement the intelligence sharing that the US has been providing and backfill it a little bit so that it won't be like Ukraine. Like Ukraine's intelligence apparatus goes completely dark when it comes to advanced warning of missiles and drones Europe can help with that Ukraine does have some domestic capability but it is kind of evolved to handle particular kinds of threats because the US was filling this other piece. So when the US removes its ability to detect certain kinds of incoming threats Ukraine doesn't have the ability to just fill that and doesn't have the capability to fill that Europe has some capability. But the US is by far the most capable player in that space and so their loss will certainly be felt. So if it remains the case that the US is unwilling to provide Ukraine with advanced warning of Russian attacks that will be a really big deal as the war continues. I'm hopeful that that isn't the case and that this was just a blip. On the Starsheel thing here Europe might have more leverage because Europe is a pretty big customer for SpaceX and for other Elon Musk companies. And so Europe can say hey like we are customers for Starlink and we don't want to support a company that is using internet access in a war zone as leverage against a country that's been invaded by an aggressive Russia. You know like the optics of this are very poor. I mean it's possible that Europe has the ability to make sure that at least some element of internet access is available for Ukraine and prosecuting this conflict. But I think there's a broader issue here with relying on a single US company for internet access in an age of the interconnected warfare where if you do not have internet access on your battlefield you are to severe disadvantage. And I think like everyone has to take a second and think is it okay that this is our single point of failure when it comes to defending Ukrainian targets against Russian drone strikes. It shouldn't be right and if you're Europe you might want to think about having an alternative that is effective here that you can put into play in case the US decides in your later conflict that they don't want to support you with with internet access. Okay when I was over there and they were talking about is it you tell sat like you tell sat or something like that so Europe has like a sort of fledgling. I'm like I don't know the fledgings right word because I don't know the level of technology but it's a smaller kind of version of Starlink so Starlink has you know tons of satellites that help you know sort of provide internet access. The EU has a version of this but I think it's my understanding is that it's much smaller perhaps not quite as robust from a technological perspective and maybe not as reliable and so it's not like a replacement necessarily but it could be one of the things that Europe develops and maybe they can put some some effort into it in the short term to sort of get it kind of ready for what it needs to do. So there is that alternative that does technically exist. It just seems like it's not really ready for prime time as a full-on replacement. But I remember the folks in the BBC were kind of talking about this as something that might become one of the priorities in the next weeks to months to years to develop because as you point out, if you're relying on one sort of point of failure, that can be pretty problematic. And the other thing, Jeff, that was under discussion, what we were over there was this sort of question about what the EU can do specifically on the weapons front. So one of the things that Ukraine has talked a lot about is eating more of these patriot missile interceptors and air defense systems and different ammunition to protect themselves. The EU doesn't necessarily have the direct ability to provide a lot more of that type of armament. But one of the interesting things that they're doing is they're actually buying some of these interceptors from the United States and transferring them to Ukraine. So it's like it's this weird kind of third middleman party or third party kind of way of getting more missiles to Ukraine, not directly from the United States, but through the intermediary of the EU, which I thought was kind of interesting. And the EU does have money. And so they have to the extent to which that there are weapons available to buy from the United States to transfer over or just, you know, provide Ukraine with as much money as possible. They do have, you know, financial resources to put towards this problem. But it does seem like the weapons themselves. It's kind of like one of the other bottlenecks here. It's like they're just not enough of them to get to Ukraine in a timely manner. So first of all, you're right that Europe does have its own thing, right? It says this is called OneWeb or Utilsat is the name of the company. It's like a French company that, you know, provides this capability and the European like Starlink alternative is called OneWeb. But the issue with it is that it's just much, much smaller. I was just pulling up the numbers. So Starlink has more than 10,000 satellites in its constellation. And OneWeb has about 650. And there are plans for an expansion of that capability. But what is on the drawing board for future OneWeb capability is replacing existing capability. Part of what's going on here is that this is conceived as like a European focused capability. And so it doesn't need to have the global scope that Starlink has, because Starlink has these ambitions to provide it in excess throughout the world, right? And OneWeb is narrower. But it also is like there's an issue with how much bandwidth can be provided. So it's not just like geographic coverage. It's how much information can you communicate via this number of satellites. And where they're located. They're also like in a different part of the world. But anyway, the, I guess the point is like there is a capability that could be scaled up if there was the political will and the money to do so. But this is a long term kind of enterprise. It's not like, it's not like Starlink could be replaced imminently with this alternative and provide the same capabilities for, for militaries in the region. And then you're, you know, you're right to about the kind of military, other military capability and, and whether Europe can, can fill the gap. There are some things that we just need the US for in our, in our current, in our current system. So it's, it's worrying to think what will happen if the US continues down this road. And it might be a difficult winner for the people of Ukraine. Yeah. I mean, I guess, I guess the one bright spot is that there seems to be this recognition that Europe can't replace the United States in terms of support for Ukraine, right? So like they're not going to be able to do everything that the United States has historically done with intelligence or weapons or Starlink. If you think that this blue wave is coming with the, with the mid-term elections, then that might, you know, bode well for Ukraine and for changes in policy at, you know, to a certain degree. But it does seem like Europe is testing the boundaries of where they're able to continue to support Ukraine without depending on Washington. So like that, that in and of itself, I think, is a bright spot in the sense that at least, you know, politically, rhetorically, economically, from a defense perspective, they seem to be doing everything that they believe that they can, which I think is, is good. And the other thing is cursing me, Jeff, you know, it's like at some point, you know, when this war ends and we don't know how it's going to end, we don't know when it's going to end, we don't know what the settlement's going to look like or if there'll be a settlement. But I think Europe is going to be in a position where they're going to be, I think, taking a large role in the reconstruction of what has happened to Ukraine and potentially providing even the security guarantees for Ukraine moving forward. And so I think for that reason, I mean, they are, you know, both, you know, so I have an interest in making sure that Ukraine is supported, but also thinking about the next step of this is like, what is Ukraine going to look like after the war? And they're going to have to be, you know, sort of on the ground being in charge, I think, of a lot of the reconstruction. If there are a democratic administration in the US, I'm confident that there will be money and resources put towards that reconstruction. But I think fundamentally this is going to be a European effort or European led effort. And so that's all the more reason why Ukraine is, you know, front and center in terms of security thinking in Europe right now because they see both the present and like ending the war on favorable terms, so Ukraine, if possible. And then what comes next is ultimately going to be their responsibility. Let's talk about AI and math. I love AI. I don't really love math, but let's let's do it. Let's see what AI and math got cooking. So just a disclaimer, I am not a mathematician, but I am pro math. I support math. What does it mean to, so when you say support, like, there's some people that don't support math. You know, I saw bumper sticker the other day. It said, you may not believe in science, but science believes in you. Right. I feel like math just exists. Yeah, like, it doesn't mean my support necessarily. No, although maybe at this difficult time, mathematicians need our support. I did read that math and mathematicians are quite upset at the moment. Yeah, let's let's talk about the math world. Yeah, so everyone saw the story, like back in September, that open AI released a solution to like a big math problem. You can tell we're going to really get into the nitty-gritty already of technical term. It was a big math problem. Well, to be fair, I mean, these problems are like, I was once talking to a PhD student in math at Ohio State. That's where I went to graduate school, and I had a friend who was in the math, you know, PhD program. And he was explaining how in math, at the PhD level, it takes about a year to like appreciate a problem. So in other words, you can't just like, like, read it and say, oh, I understand what the problem is. You need to like, read about it and do theorems and work and all this kind of stuff. For approximately a year, just even like understand what you're trying to do. So we can't do that on here on the pod. We don't have a year to dive into these problems. So we use shorthand. And it's so one of the shorthands is this is a big problem in math. So in in September, open AI released a solution to the Navier Stokes problem, which I don't know if you know if I'm saying that right, because this might be the first time I've ever said that out loud. But this, you know, open AI spent millions and millions of dollars of compute power to solve this problem, which is one of these big math challenges that's been out there. And, you know, this was met with some consternation among mathematicians, right? Because this was a problem that many people had spent their whole careers working on. And also there was a particular set of mathematicians who were using open AI tools to solve the problem. It appears that open AI may have used some of their work in its solution. So there's a whole complicated story around that, but that's not the math thing I want to talk about. So everyone's kind of heard that story. But then on October 6th, so like a week ago, open AI posted a GitHub repository of math results that had been produced by this unreleased model that they're running internally. It contained 722 manuscripts grouped into 372 families of results drawn from roughly 4,000 open problems that the model attempted to solve. So they solved in the hundreds of math problems that are among the set of difficult problems that had been unsolved in mathematics. The average result took about three hours of chat GPT pro-level compute. So per problem, they spent about three hours of chat GPT pro. Now, chat GPT pro is not, that's the $200 a month one, right? So it's not like free, but it's still, you know, that's pretty quick, pretty quick. There are a lot of like interesting problems that are among the set, but this is a big deal for mathematicians, right? Because many mathematicians have spent their careers trying to address some of these problems. So Scott Aronson, who's a quantum computing guy, a computer scientist at UT Austin called it "mathpocalypse", and so that's kind of stuck a little bit. And, you know, the stories that have been written about all these solutions have really focused a lot on the mathematicians and how they're feeling right now. Which is fair. I mean, can you just just, as an aside, imagine spending your whole career, like your 45-year-old professor, and even working on this one problem for, you know, let's say a decade or multiple decades, and then overnight in three hours, you wake up and chat GPT has solved the problem for you. That's kind of be pretty devastating. Like, I can't imagine that feels too good. Especially since we didn't, you know, when you get into this field of math, you know, you got your PhD in like 2000 or 2010 or even 2020, AI wasn't even on the radar. You might have thought hypothetically like computers would be able to solve this problem at some point, but computers in 2020 were kind of useless for a lot of these things. You needed humans to do it. And now, just like that, that need is gone. That's kind of be tricky. I'm very sympathetic and empathetic to the folks who work on these problems. One of these fields, Madolas, Hugo, Nathaniel Copin reportedly told at French, "mathematical society that he'd cried so hard one night that it woke his daughter." The Scott Erinzen's wife is a complexity theorist who's worked on this unique games conjecture for much of her career and he wrote that his nine-year-old son taunted her. He said, "Mommy, I heard you got cooked. I heard a robot solve the math problem you worked on for your whole career." So this is the kind of stuff that people are dealing with. Now I should say, what opening I release, are these solutions? It's hard to know. Right? Now the work begins of trying to figure out whether opening I really solved the things that it says it solved. So a lot of these manuscripts that it's released are not well structured or thought out and they may contain the seeds of the solution, but to really understand them is going to require some additional work. There are some backlash to this, mathematicians calling for a boycott of open AI. I mean, some of the people who are upset about this are saying, "Well, this isn't really math. If humans aren't solving it, but I don't think that is a hard time getting behind that." If solving math problems isn't like what mathematicians are about, then we might need to have a rethink. Well, you can make the argument just a more sympathetic version of that is that the AI is able to solve the problem, but you might still require humans to understand how that problem was solved. I could see that. I could see mathematicians saying, "ChatGBT has a brute force method." In the same way that chess, the computers have figured out chess in the best moves, but sometimes just because you know what's the best move, you don't understand why. It might be that the humans can still provide value in translating what the AI is able to do into some sort of human level understanding of how that problem actually was solved. I think I could see some wiggle room there, but I fundamentally agree with you. It's like if there's math problems to be solved, and AI has solved them, then that is sort of the, that's the thing. That's what math is trying to do, solve problems. Yeah. In the fullness of time, maybe humans working together with AI can solve even bigger problems. So I'm not like- Before I uncover new ones. Exactly. I'm not condemning math as an over, right, math is still a thing, and what's interesting to me and what hasn't been talked about as much, at least I haven't seen it talked about as much, and what I want to talk about with you, Marcus, is the fact that math is the basis of a lot of stuff that we care about in real life, as if math not really. Let me try that again. Like math underpins a lot of what happens in, say, our banking system, our computer systems, our cyber security, cryptography, which is kind of based on math, in many cases, is what protects our secrets, is what protects government secrets. And so there is a potential, interesting angle here that I haven't seen discussed about how math apocalypse matters for the risk of future cyber security issues, and even for the security of our financial system, which is kind of important. So I just want to talk a little bit about that. I should say that none of the things that were solved by the AI companies in this latest project, and the anthropic has been in on this too, although they've been doing it differently. So anthropic has given some results to some prominent mathematicians who have then published based on the stuff that anthropic has provided, which is a different approach to doing this and making it nicer, but it's still AI working on math problems. But none of these solutions really directly impact the field of cryptography, the codes and secrets, right? And it's almost a little surprising than none of them do until you think AI companies don't want to mess with that, right? Like there's an area here where even if your AI tool could make inroads in solving code breaking like problems, you might not want to put that on the GitHub repo with the rest of the stuff, because it is potentially national security issue and worrying and bad for financial markets and stuff like that. So the fact that none of these math problems are really cryptography adjacent is interesting in and of itself, and they reflect a selection issue in how the companies go about choosing the problems that they want to solve. But one risk here is that there's a sudden drop of new math problems that have been solved that breaks some of how we do codes in the world. Just a little bit of background. The way we secure our internet access, HTTPS, you know, the S at the end. I've always wondered about that S. Is it necessary, Jeff? Do we need the S? It is. You should use that. The way online banking works, card payments, bank-to-bank messaging, it overlies on what's called public key encryption or public key cryptography. The way this works is it's basically around a set of math problems that are too difficult for us to solve currently. So you can multiply two prime numbers together markets almost instantly. Even very big prime numbers can be multiplied together instantly by the amount of compute we have available, but if you try to factor out the result of that multiplication problem, it could take longer than the lifespan of the universe with current computing power. And we rely on that difference, the fact that you can build up this code, but you can't back out of this code to make public key cryptography work. And so one concern at the heart of something like math pocklebs is, what if we suddenly got a math result that threatened the foundations of public key cryptography or even another form of cryptography? So, you know, right now, a lot of our codes, a lot of this, a lot of the financial systems is based on this way of doing things, but there are other. So there's the blockchain, for example, where I keep all my money in Bitcoin and Ethereum right in cryptocurrency is based on elliptic curve cryptography, which is kind of a different kind of an operation, but in is a similar concept, like you are given a curve and you need to like figure out how many hops forward on a curve it takes to get to a particular point. And all of this is like relies on a math problem that is like too hard to do with the time we have available in order to be secure. So what if we got a math result dropped on us by open AI that made that problem a lot easier, tractable, something we could do relatively soon? That would be potentially worrisome. Companies that do this kind of work are already thinking about this because of quantum computing. Do you know about quantum computing? Well, I'm not an expert in it Jeff, but I've heard of the concept. Yeah, so quantum, we're not going to get into the technical detail. Don't worry. I can see the look on your face, but the real thing that quantum computing can do that makes people worry about cryptography and codes is that it can solve particular kinds of math problems that have a particular kind of underlying pattern very quickly. So quickly that these kinds of public key cryptography schemes based on prime numbers or these elliptical jumps, they wouldn't hold in a quantum computing environment. Quantum computing is real. It's a thing that's being advanced. Companies are making a lot of investments and then big advances are happening. And so a lot of firms that rely on codes being unbreakable are worried about advances in quantum computing and have been trying to transition the cryptography they use internally to oppose quantum solution, something that is more resistant to quantum computing solving. For example, a lot of the financial systems are kind of slowly moving their codes to something called lattice based cryptography, which is resistant to quantum computing approaches. It's a kind of problem that quantum computing doesn't give you an advantage in solving. Anyway, governments are doing this quickly to try to get to the point where stuff that we're trying to keep secret is protected from quantum computing. At the same time, adversaries knowing that quantum computing is down the road have been adopting an approach called harvest now to crypt later. This is where you're China or the US and you're spying on some other country and you hadn't you see some secrets and you don't have the capability now to decrypt that stuff and read it. But you might someday win quantum computing comes around, so we'll just download all that stuff, save it and wait until we have the capability to decrypt it down the road. So we're going to harvest now and decrypt it later. You could imagine a situation where we don't need to wait for quantum computing to come before we have some additional leverage over some of these math problems that underlie the cryptographic strength of our secrets. So you know, this harvest now decrypt later approach later might be like next year, six months. I don't know because we got these kind of like sudden drops of these new math solutions. Maybe one of these math solutions in the next round is on something that matters for cryptography or maybe some mathematicians somewhere uses their three hours of ChatGPT Pro to solve even though open AI doesn't want to do this because it seems to cryptography adjacent for them. Maybe some mathematician or some state wants to use one of these tools to make a headway on a problem like this. So there are people who worry about this stuff and who are kind of sounding the alarm. So there's somebody who's like associated with founding it creating Ethereum, which is one of these cryptocurrencies, has recommended that people who store a lot of money and cryptocurrency start moving their currency to systems that are less susceptible to approaches like this, which is you know exactly what you want to hear from the guy who's like who's a big proponent of cryptocurrency which was supposed to be this like very secure thing right some people are worried about this others are thinking the kinds of math problems were likely to see solutions to you from these AI attempts are not the ones that underlie cryptography but that seems like it's a bet you could make it's also like wishful thinking right it sounds like wishful thinking right like like why not right like why isn't it possible to make inroads into these particular kinds of problems that underlie cryptography so i don't know that's i don't find that particularly reassuring either what is maybe more reassuring though is that these same tools might help us create new systems for protecting our secrets that are more robust so we've already seen this happen a little bit so in the world of post quantum cryptography identifying codes that would be resistant to quantum computing there have been a bunch of proposals for new schemes that are difficult for quantum computers to break and one of those is called hawk and very recently anthropic used AI to determine that there was a vulnerability in this hawk algorithm and so it was withdrawn from consideration as one of the possible solutions to this problem so we can use AI to help create more robust cryptographic schemes on the front end right so like we have a defensive use of AI as well that we can test out these schemes and make sure that they're resistant we can maybe come up with new ideas for protecting our secrets that will be more diversified against potential mathematical breakthroughs on the other end so there is a defensive story as well that i think is important and all of this margues reminds me of cyber of the offense defense balance in cyber where we remember we talked about this a bunch of times on the podcast where like we we start with this AI stuff mythos the new anthropic model that was going to break down all the security on all the websites you know one of the things you said was like well maybe this is really good for the offense and it makes all these computer system vulnerable but then what happened was anthropic released a version of it to all the banks and said hey use this to defend yourself and we can find all the vulnerabilities in advance and fix them so that our systems will be more well defended and so there really is this this offense defense story that you can tell about a lot of the advances in AI in the international relations sphere where here we have a situation where there could be a new new math problem that weakens our cryptography and that's really good for the offense but we can also use those same tools to create better defenses that protect us against these new solutions to these math problems in the future and the real question is like where do we end up what is going to be the stronger piece is it going to be easier to break through existing schemes or is it going to be easier to build up new ones yeah Jeff again as often as the case with you in AI this is somewhat terrifying i mean i think one of the one of the interesting things i was thinking about when you were talking about the on the more kind of political international relations and it seems like if what you're saying is true and i have no reason to doubt that it's that it's not that states will be in a situation where they won't really potentially know that their cryptography has been exploited right so you can imagine like you're the United States and you have either like Russia or China or some other actor figures out the quantum computing or cryptography or whatever but basically get access to classified information or communications or financial information because it can decrypt what was previously encrypted the United States might not know about it and then the state that's doing the the the looking and the viewing and the and the sort of reconnaissance really doesn't have it much incentive to tell the state that it's viewing that it has this capability unless it wants to put controls you know on the system so that they won't then be the the next victim of this type of of attack so it creates this kind of weird incentives where like states will have this potential capability to kind of spy on on others but they might not want to be spied on and so there may be an incentive to actually cooperate in some way with with some of this stuff so i guess that's kind of like more of the the optimistic take but you're all of this does seem very similar to the old school like the 1990s you know in 2000s with the internet coming up in cyber security like the same sort of core principles like offense defense balance does seem to be in play here it's just a whole lot more complicated and it's like moving like at light speed you know whereas you know it was like just very very quick in you know 1998 now we're you know things are changing you know somewhat day-to-day um but yeah i mean in a way you can make the argument that Jervis kind of identified like a version of this in the 1970s and we're we're just seeing it play out with different technologies over time but i how worried are you Jeff like what is the what's the level of capital concern that's an index that we sometimes we'll we'll use on the capital concern index i'm not i'm not too freaked out about it i just thought this is a really cool demonstration of how things can change really suddenly yeah and in an area that is not something you might think of as an international security concern like mostly we're just all you know weeping for the mathematicians in our lives but if you think about it math is really important to a lot of things i don't know i don't know if your high school math teacher ever told you like this will be important someday i mean this is stuff like undergirding so much of how we function in our society and we don't really realize it and so you get this kind of sudden advance in this area and i i'm just gonna like assuming it's an advance let's just put aside all the controversy if this really did solve a bunch of problems and we get another drop that solves a bunch of different problems eventually some of these might have implications for these international security concerns and i'm also like you know we don't know what everybody is doing so so opening i just dropped all this stuff at once who else out there is working on this if it's a government they're unlikely to publish their solution right right exactly the the enigma uh codes in world were two exactly which which were cracked uh by the uk in the in the ultra project Alan Turing at the lead of that one right there's a good movie about this yep we went to great lengths not to show Germany that this code's been cracked i think that there is a strong incentive for states to be involved in this kind of AI facilitated cryptographic research that breaks these forms of encryption and they don't say anything and then clues our own state right so like if u_s_ had the ability to break the htps public key encryption and read everybody's internet transmissions would you think they would tell us Marcus i don't have a lot of confidence that this is something that the u_s_ would publicize i think they would use it you know in it for intelligence purposes and potentially for domestic surveillance purposes the same is true of China certainly so i'm not comforted by the fact that nobody has put this out there in the world yet because we might never know if it's a state actor or someone who hopes to gain from this work that actually solves it i think it's quite worrying on the offense defense side you know there arguments about like which would be stronger that i think are kind of interesting on the offense side on in cyber there's this argument that you only need one bug to break in and there are always bugs and the defender can't possibly identify every possible gap in their defenses and the cryptography argument is similar right that like creating a cryptographic standard that the world uses that's not a quick thing to do if we were to break one of these standards it would be out there would be no way to fix it quickly right like there's no patch ability in the world of cryptography you just need to replace it with a different scheme and it's not clear how quickly that could happen in any kind of real world environment and so you know there's this idea of like harvest now decrypt later like everything that anyone is collected up to this point then would be compromised and so from the defensive perspective it's hard to know what you can do about that besides like preemptively moving all your stuff into a better cryptographic standard but you don't know whether that's the one that's going to end up being secure as we've seen like already some of these proposals for post quantum cryptography not holed up to scrutiny so like on the defense side it's really it's really hard to think about whether you have much of an advantage here except that you can throw everything you have in AI against your potential solution to try to battle test it make sure it's it's hardened before you put it into place but that may be the one advantage that defense has in this in this realm yeah anyway that's just something to for everyone to contemplate maybe you should have some gold bars or cash money I don't know it gets back to math it's true like if you ever think of physics class you go you can take a physics class you think you're you know you're learned about like different atoms moving around it turns out it's just math you know there's physics is math like everything's just math computer science is math everything just math yeah cryptography's math AI is all around us all around us Marcus but are you going to are you going to go get some extra cash just to have on hand I was at the baking system I was at an event recently where this was the the topic of conversation it's like what what happens when you wake up in the morning and you load up your like Bank of America dot com or whatever and it's errors all zeroed out like you can't you go the ATM card doesn't you know it's not reading it yeah that's a little scary I'm not sure what happens this is a chance this is a chance for me to recommend one of my favorite movies of all times sneakers let's the late 80s Robert Redford that is a star-studded movie Dan Akrod is in that movie River Phoenix is in that movie Sydney Poetiers in that movie well that's good yep Ben Kingsley and it's like it's fun like for the for the younger viewers out there who might not been alive in 1992 I think it holds up it's worth watching yeah it's like a heist movie at the heart but but the thing that they're working on is the is the ability to see all the secrets in the world and which I guess is the promise of quantum computing in the world so yeah good chance to recommend sneakers everyone should go watch sneakers we should talk about we should have a whole podcast on on early there is there is that what's the pod like rewatchables where they watch yeah yeah so we could do something similar they could these could be AI slash IR like wargames like that was a that's sort of like an early glimpse into the future in some way because well how all that holds up this can be a whole new segment they should invite us on the rewatchables to should to rewatch those movies bring our international relations expertise to bear let's leave it there thank you so much for joining me today Marcus it's great Jeff thanksgiving is just a couple days away we're gonna start our preparations here soon favorite time of year wrong month we could try that again next month please send us an email we're cheap talk pod at gmail.com we'd love to hear from you you can go to our online store at cheaptalk.shop where you can purchase some merchandise we need more AI maybe a math pocket lip shirt now that I think about it I do I I will say the the our mugs are flying off the shelf I noticed that lots of lots of mug purchases recently that's true I think the panda diplomacy mugs in particular have been spurred on by the recent Trump she interaction I think that's right I also think people's anxiety is up over all his AI stuff is sort of drinking more coffee and you know I think that's probably it yeah you should subscribe to this podcast wherever you get it it could be Apple podcasts it could be Spotify could be overcast wherever you get your podcast we are there unless it's YouTube or not there but everywhere else you can get us so subscribe like leave a review we appreciate that Marcus thanks again folks we'll see you next time that's the other thing on these trips I feel like out of the loop although you turn the BBC and it's like they're covering all these countries I've never even heard of before what's going on in Guadalupe like the American media just I know like you know that like France is like burning like these students are very upset and they're like taken to the streets and BBC's like they're with like reporters here it's just like a little blip you know all protests in Paris you know there's a protest in Paris the whole city's burning down for crying out loud my favorite part of British TV is like if you listen to the news or what or if you're watching the news day they give the weather in various countries the weather in the US is 60 you know first of all it's always like 26 degrees whatever that is right but like I know but what does that mean it's the weather in the US it's a big country I it does it's still like aspirational like you know it's like oh the weather in Hong Kong today is blah blah blah are Tokyo you think you're sound like one day I might go to Hong Kong or Tokyo and like this will be relevant for me you know but it's it's a BBC worldwide service that's why so if you happen to be in Hong Kong and you're getting your weather for some reason from the BBC you have it you know that's pretty good though weather in the US I also like the the in the BBC when you watch the television part the way that they depict clouds and like wind it's very nice like there's like this little animations that look like swiggly lines that show you which way the wind is blowing I find those to be helpful oh the other the other thing we should do at some point is have a midterm elections I know I know international relations as our specialty we're both experts in that that's right but we do have elections coming up and I don't know about you but I have students asking me like what I think is going to happen and then based on what happens like what they what we can expect in terms of like international politics so that might be a good we can do like a midterm elections special what do you think is going to happen so I'm a I'm a as you know Jeff I'm a well-known polling skeptic like after 2016 people like Nate Silver are dead to me and so I I don't I don't believe the polls I don't listen to them and I try to like pay no attention whatsoever but what I do pay attention to is like the way people talk and and where the ways in which politicians kind of like spend their money and and the parties and like who what what races are pouring money into because I think they probably have a sense they have their own polling of course which is probably better than the public stuff and then they also I think just on the ground more so they have a better sense of what's what's happening and it seems like the Republicans think they're in trouble you know certainly in the house which I think everybody believes is going to go to Democrats but also like the Senate like it just seems like these tight races you know Maine Alaska like it's just some of these could flip and the the Democrats are you what for and it looks like that there's a decent shot that that happens so if the if the Democrats take over both the House and Senate I think it makes Trump's remaining two years of his presidency very uncomfortable at best and he's going to be plagued by all kinds of different investigations and impeachment and things like that and I think that the Democrats will try to put an end to the Iran war as quickly as possible they have like some tools that are supposed to do that but so I don't I kind of think the Democrats are poised for a pretty decent time in a few weeks in the midterms what do you think Jeff? I think that the Democrats are likely to take control of Congress and so you know that obviously has some national security implications but what's bum bum me out a little bit is that I there these prediction markets you're familiar with these things right call sheet and polymarket and predict it so these used to be a pretty good way to make some extra money in election years because the the markets these prediction markets were so skewed in one direction or the other that you could bet on like pretty much sure things and still do better than the fees and everything and and make out pretty well but I think that now that prediction markets have gone mainstream and everybody's betting on all these sports on these prediction markets the election side of the house has also gotten more mainstream and this arbitrage opportunity no longer exists so now the prediction markets seem to pretty closely mirror the election models that are out there that's unfortunate yeah so my opportunities for you know supporting my family through additional election betting have kind of fallen away like right now I'm looking at call sheet it says Democrats have a 62% chance to control the Senate if you go to split ticket which is a really good election model that was that was pretty good in the last elections it has 65% chance of Democrats winning the Senate 50 plus one which is kind of my favorite right now of these election models has Democrats with a 71% chance 71 wow so that's a little bit higher there's some interesting reasons for why split ticket is is a little more optimistic for Democrats than than some of the others but even that that arbitrage opportunity is limited right so if you wanted to make money off of these models I think that would be harder to do so in other words the models are suggesting a blue wave as as you know sort of locally people are talking about this they certainly are I mean that and that's because the polls are right I mean it the the models are just taking the polls and well it's everything though it's like what people think right so that's largely formed by polls but it could be formed by other other variables as well yeah I mean what I'm trying to say is everything is pointing in the same direction right so like looking at the fundamentals if you want to look at like the presidential approval rating economic situation whether people feel that they're able to make ends meet all of these things are pointing in that direction and so are the polls and also just just history yeah like demographic shifts all of this stuff is the the weight of midterm history is weighing on this on this election so it's not as if they're pointing in different directions we have to kind of adjudicate things it really is looking like a kind of blue wave year but you know the polls been wrong so it's hard to hard to know whether they'll be wrong again this time maybe we can in a future podcast talk a little bit about non-response bias and whether that's affecting our assessment of the poll results as they come in well it's historically been one of the things that under under-reporting of Trump support remember the shy Trump voter from 2016 I do I do I have strong recollections of that the shy jump yeah apparently they exist