Liv Boeree: Poker, Game Theory, AI, Simulation, Aliens & Existential Risk | Lex Fridman Podcast #314

Evolutionarily, if we see a lion running at us, we didn't have time to sort of calculate the lion's kinetic energy. And is it optimal to go this way or that way? You just reacted. And physically, our bodies are well attuned to actually make right decisions. But when you're playing a game like poker, this is not something that you ever evolved to do, and yet you're in that same flight or fight response. And so that's a really important skill to be able to develop, to basically learn how to meditate in the moment and calm yourself so that you can think clearly.

The following is a conversation with Liv Burry, formerly one of the best poker players in the world, trained as an astrophysicist, and is now a philanthropist and an educator on topics of game theory, physics, complexity, and life. This is the Lex Friedman podcast to support it. Please check out our sponsors in the description. And now, dear friends, here's Liv Burry. What role do you think luck plays in poker? And in life, you can pick whichever one you want. Poker or life and. Or life.

The longer you play, the less influence luck has. You know, like with all things, the bigger your sample size, the more the quality of your decisions or your strategies matter. So to answer that question, yeah, in poker, it really depends. If you and I sat and played 10 hands, right now, I might only win 52% of the time, 53% maybe. But if we played 10,000 hands, then I'll probably win like over 98, 99% of the time. So it's a question of sample sizes.

And what are you figuring out over time, the betting strategy that this individual does, or literally it doesn't matter. Against any individual over time.

Against any individual over time, the better player, because they're making better decisions. So what does that mean, to make a better decision? Well, to get into the real nitty gritty already, basically, poker is a game of math. There are these strategies familiar with Nash Equilibria. That's. So there are these game theory optimal strategies that you can adopt, and the closer you play to them, the less exploitable you are. So because I've studied the game a bunch, although admittedly not for a few years, but back in, you know, when I was playing all the time, I would study these game theory optimal solutions and try and then adopt those strategies when I go and play. So I'd play against you and I would do that. And because the objective, when you're playing game theory optimal, it's actually, it's a loss minimization thing. That you're trying to do. Your best bet is to try and play a sort of similar style. You also need to try and adopt this loss minimization. But because I've been playing much longer than you, I'll be better at that. So first of all, you're not taking advantage of my mistakes. But then on top of that, I'll be better at recognizing when you are playing suboptimally and then deviating from this. Game theory optimal strategy to exploit your.

Bad plays can you define game theory and Nash equilibria? Can we try to sneak up to it in a bunch of ways? What's a game theory framework of analyzing poker, analyzing any kind of situation?

So game theory is just basically the study of decisions within a competitive situation. I mean, it's technically a branch of economics, but it also applies to wider decision theory. And usually when you see it, it's these little payoff matrices and so on. That's how it's depicted. But it's essentially just like study of strategies under different competitive situations. And as it happens, certain games, in fact, many, many games have these things called Nash equilibria. And what that means is when you're in a Nash equilibrium, basically there is no strategy that you can take that would be more beneficial than the one you're currently taking, assuming your opponent is also doing the same thing. So it would be a bad idea if we're both playing in a game. 3. Optimal strategy if either of us deviate from that now, we're putting ourselves at a disadvantage. Rock Paper Scissors is actually a really great example of this. If we were to start playing rock paper, you know nothing about me and we're going to play for all our money. Let's play 10 rounds of it. What would your sort of optimal strategy be, do you think? What would you do?

Let's see. I would probably try to be as random as possible.

Exactly. Because you don't know anything about me. You don't want to give anything away about yourself. So ideally you'd have a little dice or somewhat, you know, you know, perfect randomizer that makes you randomize 33% of the time each of the three different things. And in response to that, well, actually I can kind of do anything, but I would probably just randomize back too. But actually it wouldn't matter because I know that you're playing randomly. So that would be us in a Nash equilibrium where we're both playing this unexploitable strategy. However, if after a while you then notice that I'm Playing rock a little bit more often than I should.

Yeah, you're the kind of person that would do that, wouldn't you?

Sure, yes, yes, yes. I'm more of a scissors girl, but anyway, you are. No, I'm as I said, randomizer. So you notice I'm throwing rock too much or something like that. Now you'd be making a mistake by continuing playing this game theory optimal strategy. Well, the previous one, because you are now I'm making a mistake and you're not deviating and exploiting my mistake. So you'd want to start throwing paper a bit more often in whatever you figure is the right sort of percentage of the time that I'm throwing rock too often. So that's basically an example of where what game 3 optimal strategy is in terms of loss minimization. But it's not always the maximally profitable thing if your opponent is doing stupid stuff, which in that example. So that's kind of then how it works in poker, but it's a lot more complex. And the way poker players typically nowadays they study. The game's changed so much and I think we should talk about how it sort of evolved. But nowadays the top pros basically spend all their time in between sessions running these simulators using software where they do basically Monte Carlo simulations, sort of doing billions of fictitious self play hands. You input a fictitious hand scenario like oh, what do I do with Jack 9 suited on a king 10, 42 spade board against this bet size. So you'd input that, press play, it'll run its billions of fake hands and then it will converge upon what the game theory optimal strategies are. And then you want to try and memorize what these are. Basically they're like ratios of how often, what types of hands you want to bluff and what percentage of the time. So then there's this additional layer of inbuilt randomization built in.

Yeah, those kind of simulations incorporate all the betting strategies and everything else like that. So as opposed to some kind of very crude mathematical model of what's the probability you win just based on the quality of the card, it's including everything else too. The game theory of it.

Yes, yeah, essentially. And what's interesting is that nowadays if you want to be a top pro and you go and play in these really like the super high stakes tournaments or tough cash games, if you don't know this stuff, you're going to get eaten alive in the long run. But of course you could get lucky over the short run. And that's where this like luck factor comes in. Because luck is both a blessing and a curse. If luck didn't, you know, if there wasn't this random element and there wasn't the ability for worst players to win sometimes, then poker would fall apart. You know, the same reason people don't play chess professionally for money against. You don't see people going and hustling chess, like, not knowing, trying to make a living from it because, you know, there's very little luck in chess, but there's quite a lot of luck in poker.

Have you seen Beautiful Mind? That movie?

Years ago.

Well, what do you think about the game theoretic formulation of. What is it? The hot blonde at the bar you remember, like. Oh, yeah. The way they illustrated it is they're trying to pick up a girl at a bar and there's multiple girls. They're like friend. It's like a friend group when you're trying to approach. I don't remember the details, but I remember.

Don't you, like, then speak to her friends like that? Feigned disinterest. I mean, it's classic pickup artist stuff. Right. You want to.

And they were trying to correlate that somehow that being an optimal strategy game theoretically. Why? What, What, Like, I don't think I remember.

I can't imagine that there is. I mean, there's probably an optimal strategy. Is it. Does that mean that there's an actual Nash equilibrium of, like, picking up girls?

Do you know the, the, the marriage problem? It's optimal stopping.

Yes.

So where. It's an optimal dating strategy where you. Do you remember what it was?

Yeah, I think it's like something like you, you, you know, you've got like a set of 100 people you're going to look through. And after how many do you. Now after that, after going on this many dates out of a hundred, at what point do you then go, okay, the next best person I see. Is that the right one? And I think it's like something like 37%.

It's one over E, whatever that is.

Right. Which I think is.

We're going to fact check that. Yeah. So. But it's funny, under those strict constraints, then. Yes. After that many people, as long as you have a fixed size pool, then you just pick the, the per. The next person that is better than anyone you know before.

Yeah. Have you, have you tried this? Have you incorporated it?

I'm one of those people. I'm. We're. And we're going to discuss this. I. And what do you mean those people? I try not to optimize stuff. I try to listen to the heart. I don't think I like. My mind immediately is attracted to optimizing everything. And I think that if you really give in to that kind of addiction that you lose the joy of the small things, the minutiae of life. I think. I don't know, I'm concerned about the addictive nature of my personality in that regard.

In some ways. While I think the on average people under try and quantify things or try under optimize, there are some people who, you know, it's like with all these things it's a, you know, it's a balancing act.

I've been on dating apps, but I've never used them. I'm sure they have data on this because they probably have the optimal stopping control problem because there aren't a lot of people that use social like dating apps on there for a long time. So the, the, the interesting, the interesting aspect is like, all right, how long before you stop looking before it actually starts affecting your mind negatively such that you see dating as a kind of.

A game.

A kind of game versus an actual process of finding somebody that's going to make you happy for the rest of your life. That's really interesting. They have the data. I wish they would be able to release that data. And I do wonder.

It's OkCupid, right? I think they ran a huge study on all of the.

Yeah, they're more data driven. I think OkCupid folks are. I think there's a lot of opportunity for dating apps. Even bigger than dating apps, people connecting on the Internet. I just hope they're more data driven. And it doesn't seem that way. I think like I always thought that Goodreads should be a dating app. Like the.

I've never used it.

Goodreads is just list like books that you've read.

Okay.

And it allows you to comment on the books you read and what books you're currently reading. But it's a giant social networks of people reading books. And that seems to be a much better database of like interests. Of course it constrained you to the books you're reading. But like that really reveals so much more about the person, allows you to discover shared interests because books are kind of window into the way you see the world. Also like the kind of places, people you're curious about, the kind of ideas you're curious about. Are you a romantic? Are you cold calculating rationalist? Are you, are you into Ayn Rand or are you into Bernie Sanders? Are you into whatever, right? And I feel like that reveals so much more than like a. A person trying to look hot from a certain angle and a Tinder profile.

Well, and it also be a really great filter in the first place for people. It selects for people who read books and are willing to go and rate them and give feedback on them and so on. So that's already a really strong filter of probably the type of people you'd be looking for.

Well, at least be able to fake reading books. I mean, the thing about books, you don't really need to read it.

You can just look at the book. Yeah. Game the dating App by feigning intellectualism.

Can I admit something very horrible about myself?

Go on.

The things that you know, I don't have many things in my closet, but this is one of them. I've never actually read Shakespeare. I've only read Cliff Notes and I got a 5 in the AP English exam.

Which book?

Which books have I read?

Well, yeah. Which was the exam on which book?

Oh, no, they include a lot of them, but Hamlet. I don't even know if you read Romeo and Juliet, Macbeth, I don't remember. But I don't understand it. It's really cryptic. It's hard, and it's not that pleasant to read. It's like ANC speak. I don't understand it anyway. Maybe I was too dumb. I'm still too dumb. But I did got a five, which is. Yeah.

Yeah. I don't know how the US grading system.

Oh, no. So AP English is a. There's kind of this advanced versions of courses in high school and you take a test that is like a broad test for that subject and includes a lot. It wasn't obviously just Shakespeare. I think a lot of it was also writing. Written. You have like AP Physics, AP Computer Science, AP Biology, AP Chemistry, and then AP English or AP Literature. I forget what it was, but I think Shakespeare was a part of that.

But I and you and you Game. The point is you gamified it.

Well, entirety. I was into getting A's. I saw it as a game. I don't think any. I don't think all the learning I've done has been outside of the. Outside of school. The deepest learning I've done has been outside of school, with a few exceptions, especially in grad school, like deep computer science courses. But that was still outside of school because it was outside of getting. Sorry, it was outside of getting the A for the course. The best stuff I've ever done is when you read the chapter and you do many of the problems at the end of the chapter, which is usually not what's required for the course, like the hardest stuff. In fact, textbooks are freaking incredible. If you go back now and you look at like biology textbook or any of the computer science textbooks on algorithms and data structures, those things are incredible. They have the best summary of a subject, plus they have practice problems of increasing difficulty that allows you to truly master the basic, like the fundamental ideas behind that.

I got through my entire physics degree with one textbook that was just this really comprehensive one that they told us at the beginning of the first year by buy this, but you're going to have to buy 15 other books for all your supplementary courses. And I was like, every time I would just check to see whether this book covered it and it did. And I think I only bought like two or three extra and thank God, because they're super expensive textbooks. It's a whole racket they've got going on. Yeah, they are. You get the right one, it's just like a manual for. But what's interesting though is this is the tyranny of, of having exams and metrics.

The tyranny of exams and metrics. Yes.

I loved them because I loved. I'm very competitive and I liked finding ways to gamify things and then sort of dust off my shoulders afterwards when I get a good grade or be annoyed at myself when I didn't. But yeah, you're absolutely right in that the actual, you know how much of that physics knowledge I've retained, like I learned how to cram and study and please an examiner. But did that give me the deep lasting knowledge that I needed? I mean, yes and no. But really, nothing makes you learn a topic better than when you actually then have to teach it yourself. I'm trying to wrap my teeth around this game theory Moloch stuff right now and there's no exam at the end of it that I can gamify. There's no way to gamify and sort of shortcut my way through it. I have to understand it so deeply from deep foundational levels that to then build upon it and then try and explain it to other people. And like, you know you're about to go and do some lectures, right? You can't sort of just like you presumably can't rely on the knowledge that you got through when you were studying for an exam to reteach that.

Yeah. And especially high level lectures, especially the kind of stuff you do on YouTube, you're not just regurgitating material. You have to think through what is the Core idea here. And when you do the lectures live, especially, you have to. There's no second takes. Yeah, that is a luxury you get if you're recording a video for YouTube or something like that. But it definitely is a luxury you shouldn't lean on. I've gotten to interact with a few youtubers that lean on that too much and you realize, oh, you're, you've gamified this system because you're not really thinking deeply about stuff. You're through the edit, both written and spoken, you're crafting an amazing video. But you yourself as a human being have not really deeply understood it. So live teaching, or at least recording video with very few takes is a different beast. And I think it's the most honest way of doing it, like as few takes as possible.

That's why I'm nervous about this.

Don't go back.

Don't like Alice. Do that.

Don't fuck this up. Live the tyranny of exams. I do think, you know, people talk about, you know, high school and college as a time to do drugs and drink and have fun and all this kind of stuff. But you know, looking back, of course I did a lot of those things. No. Yes. But it's also a time when you get to like read textbooks or read books or learn with all the time in the world. Like you don't have these responsibilities of like, you know, laundry and having to sort of pay for mortgage or all that kind of stuff, pay taxes, all this kind of stuff. In most cases, there's just so much time in the day for learning and you don't realize it at the time because at the time it seems like a chore, like why the hell there's so much homework. But you never get a chance to do this kind of learning, this kind of homework ever again in life. Unless later in life you really make a big effort out of it. You get like, you basically your knowledge gets solidified. You don't get to have fun and learn. Learning is really, is really fulfilling and really fun if you're that kind of person. Like some people like to, you know, like knowledge is not something that they think is fun. But if that's the kind of thing that you think is fun, that's the time to have fun and do the drugs and drink and all that kind of stuff. But the learning, just going back to those textbooks, the hours spent with the textbooks is really, really rewarding.

Do people even use textbooks anymore?

Yeah. Do you think because it's these days with their TikTok and well, not even.

That, but just like so much information, really high quality information, you know, is now in digital format online.

Yeah, but they're not there are using that. But, you know, college is still very. There's a curriculum. I mean, so much of school is about rigorous study of a subject. And still on YouTube, that's not there, right? YouTube has Grant Sanderson talks about this. Is this master? Yeah, three blue, one brown. He says, like, I'm not a math teacher. I just take really cool concepts and I inspire people. But if you want to really learn calculus, if you want to really learn linear algebra, you just, you should do the textbook. You should do that, you know, and there's still the, the textbook industrial complex that, that like charges like $200 for textbook and some somehow, I don't know, it's ridiculous.

Well, they're like, oh, sorry, new edition edition 14.6. Sorry, you can't use 14.5 anymore. It's like, what's different? We've got one paragraph different.

So we mentioned offline Daniel Negrano. I'm gonna get a chance to talk to him on this podcast. And he's somebody that I found fascinating in terms of the way he thinks about poker, verbalizes the way he thinks about poker, the way he plays poker. So. And he's still pretty damn good. He's been good for a long time. So you mentioned that people are running these kinds of simulations and the game of poker has changed. Do you think he's adapting in this way? Do you like the top pros, do they have to adopt this way or is there still, like, over the years, you basically develop this gut feeling about, like, you get to be like, good the way, like AlphaZero is good. You look at the board and somehow from the fog comes out the right answer. Like, this is likely what they have. This is likely the best way to move. And you don't really. You can't really put a finger on exactly why, but it just comes from your gut feeling or no.

Yes and no. So gut feelings are definitely very important that we've got our two modes. Well, you can distill it down to two modes of decision making, right. You've got your sort of logical, linear voice in your head. System two, as it's often called, and your system one, your gut intuition. And historically in poker, the very best players were playing almost entirely by their gut. You know, often they'd do some kind of inspired play and you'd ask them why they do it, and they wouldn't really be able to explain it. And that's not so much because their process was unintelligible, but it was more just because no one had the language with which to describe what optimal strategies were, because no one really understood how poker worked. This was before, I guess, some people would write down their hands in a little notebook. But there was no way to assimilate all this data and analyse it. But then with when computers became cheaper and software started emerging and then obviously online poker, where it would like, automatically save your hand histories, now all of a sudden you had this body of data that you could run analysis on. And so that's when people started to see these mathematical solutions. And so what that meant is the role of intuition essentially became smaller and it went more into, as we talked before about this game theory optimal style. But also, as I said, game theory optimal is about loss minimization and being unexploitable. But if you're playing against people who aren't, because no person, no human being can play perfectly, game theory optimal in poker, not Even the best AIs, they're still like, they're 99.99% of the way there or whatever, but it's kind of like the speed of light, you can't reach it perfectly.

So there's still a role for intuition.

Yes. So when you're playing this unexploitable style, but when your opponents start doing something suboptimal that you want to exploit, well, now that's where nothing.