← Intelligence Squared
Intelligence Squared
Why Does It Sometimes Pay to Be a Chicken? With Professor Michael Wooldridge
00:42:42
Loading summary
Transcript41 lines
- [00:01]
Progressive Insurance Narrator
Insurance isn't one size fits all, and shopping for it shouldn't feel like squeezing into something that just doesn't fit. That's why drivers have enjoyed Progressive's name your price tool for years. With the name your price tool, you tell them what you want to pay and they show you options that fit your budget enough. Hunting for discounts, trying to calculate rates, and tinkering with coverages. Maybe you're picking out your very first policy, or maybe you're just looking for something that works better for you and your family. Either way, they make it simple to see your options. No guesswork, no surprises. Ready to see how easy and fun shopping for car insurance can be? Visit progressive.com and give the name your price tool a try. Take the stress out of shopping and find coverage that fits your life on your terms. Progressive Casualty Insurance company and affiliates Price and coverage match limited by state law.
- [00:54]
Professor Michael Wooldridge
So good, so good, so good.
- [00:57]
Nordstrom Rack Advertiser
New markdowns up to 70% off are at Nordstrom rack stores now. Stock up and save big on shoes, tops, dresses, accessories and more must haves for summer. Join the NordicLub to unlock exclusive discounts. Shop new arrivals first and more. Plus buy online and pick up at your favorite rack store for free. Great brands, great prices. That's why you rack.
- [01:21]
Mia Sorrenti
Welcome to Intelligence Squared, where great minds meet. I'm producer Mia Sorrenti. Why does it sometimes make sense to swerve, compromise or cooperate? And at other times to stand your ground? From international crises to everyday decisions, how can we think more strategically about the choices we make? On today's episode, Professor Michael Wooldridge Aschell professor of the Foundations of Artificial Intelligence at the University of Oxford, sheds light on how game theory, one of mathematics most influential fields, can offer some surprising life lessons when it comes to conflict, cooperation and human behavior. Let's join our host, Carl Miller, now with more.
- [02:00]
Carl Miller
Welcome to Intelligence Squared, everyone. I'm Carl Miller and our returning guest today is Professor Mike Wooldridge. Michael Wooldridge is the ashore professor of the Foundations of Artificial Intelligence in the Department of Computer Science at the University of Oxford. A senior research fellow at Hertford College, he's been an AI researcher since 1989, before AI was Cool Mike and has published more than 400 scientific articles on the subject. And his latest book is Life Lessons from Game Theory the Art of Thinking strategically in a Complex World. Michael, you've been on tentative quote before. Very warm welcome back to you.
- [02:37]
Professor Michael Wooldridge
Great to be back.
- [02:38]
Carl Miller
So I'm a firm believer in beginning at the beginning when it comes to these sorts of subjects. MICHAEL so the first question might be a very simple one, but can you just explain what game theory actually is for people that might not have ever heard of it before or of course used it?
- [02:56]
Professor Michael Wooldridge
So game theory is the mathematical theory that tries to understand every situation in which self interested parties interact. And those parties could be me and you, they could be governments, they could be nations engaged in a bloody war that are trying to navigate their way through it. They could be anything which is perceived of as having choices that can take actions. And it also has self interest in the sense that it has preferences about how the world is. So game theory was actually invented as far back as the 1920s and one of the main inventors was the Hungarian American mathematician John von Neumann. And von Neumann was a mathematician who was interested in games. And he asked himself a very, very, very natural question for a mathematician. He said, could game theory teach me how to play poker? Well now von Neumann solved that problem mathematically and in doing so he kind of set off in train the whole of the field of game theory. He didn't actually tell us how to play poker well, but he established the main theory that was underpinning it. The problem is when people hear game theory as a consequence of that, they think it's only to do with like parlour games, games like chess or poker or checkers and so on. But actually games, game theory studies every scenario where self interested parties interact with one another. And that could be people playing a game of poker. In a game of poker it's obvious what their self interest is. They want to win. You know, if I'm playing with you, I want to win the game, you also want to win the game. So our preferences there are not aligned with one another. And so the theory tells you how you might behave in such a game. The players could be, it could be me and my wife negotiating about who's going to wash up the dishes and who gets to choose what we watch on t. They could be a parent negotiating with a child about how much screen time they get before they go to bed. They could be you and your boss negotiating about your pay rise, or two companies trying to reach a deal in the market. So although it has its origins in parlor games, the scope of game theory is very, very, very broad indeed.
- [05:15]
Carl Miller
And so if we just stay with a poker example and von Neumann just for one more moment, what does that look like? So if one is applying kind of mathem principles of kind of analysis or conjecture to a problem like poker, like what actually are you doing? How does Maths help you think about decision making in that particular game.
- [05:36]
Professor Michael Wooldridge
So there is a big branch of mathematics called optimization, which is about trying to compute the optimal choice given a range of possibilities and your preferences over this. And it's a very big, very, very well established theory. And what game theory does is try to say, how do we do that in a game of poker? What's the choice that I could make, the move I could make in this game, which would result in the best outcome for me. Now what von Neumann did is, I mean, he had to greatly simplify things because poker's a complex game. But he came up with a simple model of what are called zero sum games. And a zero sum game is basically a win lose game, and a game of poker is a zero sum game in some sense. Because if I win, you lose.
- [06:19]
Carl Miller
Right, Which I frequently do when it comes to poker.
- [06:23]
Professor Michael Wooldridge
And me too, don't worry about it. I'm not going anywhere near poker, not ever again, not after the last time. But anyway, what von Neumann did is he showed what you need to do in a zero sum encounter. But there was a problem with that because the theory that he first established is about essentially games of conflict where there can be one winner and everybody else must be a loser. Zero sum games. And, and for the next 30 years, 25 years, 30 years, that was more or less it in game theory. And so game theory got kind of identified as the science of conflict, whereas in practice, game theory also deals with cooperation. When can cooperation happen? When can mutually sustained rational cooperation happen, when can win, when outcomes occur? Also concerned with those things, but in the public perception, it was the science of conflict. So part of the reason for writing the book is to continue the rehabilitation of game theory and to show people that it's as much about subjects like cooperation as it is about the kind of conflict scenarios that John von Neumann was the first to study.
- [07:34]
Carl Miller
That's an interesting word you just used, Mike, rehabilitation of game theory. Because I guess if people have heard of game theory before, they've probably heard it, I'm guessing in probably one of two contexts. Either, of course, AI and we'll talk about that for sure, everyone. But then probably also, if not there, then the Cold War, you know, and lots of nuclear missiles pointing at each other. So. So has kind of game theory, you think, kind of been exiled for a while or at least kind of carries with it this slightly kind of dangerous connotation.
- [08:06]
Professor Michael Wooldridge
Yeah, and it was problematic. It was, it was, it had a problematic reputation. And part of that was that a Lot of the early game theorists, extremely eminent game theorists, were looking at scenarios like nuclear war, saying, how can we play a nuclear war? Well, you can imagine an enormous number of people. A nuclear war is not a game. And these game theorists are reducing the ultimate nightmare of global thermonuclear conflict to a game. And the stereotype was that these were kind of sociopathic robots that divorced from any human understanding. How can you think about war as a game for so many people? That was just appalling. That's an oversimplification of what went on. But the fact that a lot of early work in game theory was at the height of the Cold War and was influential in Cold War thinking came to be kind of identified with the subject. But I say it's. The other problem was that the earliest work on game theory was about essentially games of conflict, zero sum games where anything good for me is bad for you, where I am motivated to make you do as badly as I possibly can do, because that's how I to get the best for myself. And this kind of got wrapped up in the reputation of game theory and it took quite a lot of people to kind of rescue it. And of course, one of the people that rescued it from that was John Nash, the very famous mathematician, Princeton mathematician, who was immortalized in the book A Beautiful Mind and the Hollywood movie of the same name. And what Nash did was he showed that actually game theory was applicable to a much, much wider range of scenarios than just games. And that led to an explosion in the applications of game theory.
- [09:57]
Carl Miller
And your book, of course, is the Life Lessons to Game Theory. So I suppose again, you're extending it beyond say, Nash's economics to an even wider variety of different contexts. Before we jump into the lessons just writ large. Mike, did you write the book because you think that a kind of world fuller of game theory is a kind of better world? Do people kind of make mistakes when they don't use game theory? In certain ways?
- [10:23]
Professor Michael Wooldridge
So I think there are a couple of things to take away from the book. And the first is that the first lesson in the book is it isn't all about greed and it isn't all about money. And there is this perception that game theory advocates some model of human behavior which is selfish in the kind of the cruel sense of selfish, that it's just me scrabbling to get the best outcome for myself and stomping on everybody else. Actually, game theory admits saints as well as sinners. It copes with saints just as. As sinners. It doesn't advocate some model where I am scrabbling to get ahead and trying to kick everybody else down and make them do as badly as possible. So that's the first part of what I wanted to say in the book. And I think that's an important lesson. It isn't just applicable to scenarios of pure greed and selfishness. It really is just as much about cooperation as it is about conflict. But secondly, I think an important lesson in the book is that trying to get to good outcomes, to get to cooperative outcomes, to outcomes that are satisfactory from the point of view of society, you can't do that naively. Game theory tells you what's required to get there rationally. And some people find the advice unpalatable, but the advice is very, very robust advice. And I think it's advice that we could all be advised to listen. But the fundamental reason I wrote the book is I teach this as part of a computer science and AI course here at the University of Oxford. I've been teaching it for more than a decade now. And my experience of teaching it with students has been very different to anything else that I've taught. I mean, I've taught pure mathematics courses and very often you get very disinterested students with pure mathematics courses. I've taught hands on robotics, where the students get excited at the beginning but then end up getting disappointed because robotics is really quite hard to get right. But actually, the reaction with students on this course has been very, very different. They engage with it. They engage with it in a way that they don't engage with other courses. And this is what got me thinking. Perhaps there's a book here where I can explain the kind of ideas that I explain in my course, but obviously without mathematics and without computer programs and so on. So that's what the book's an attempt to do.
- [12:50]
Carl Miller
All right, well, enough beating about the bush. Let's dive into some of these lessons. I think there's 21 in total. So we're unfortunately not going to able to cover the full landscape of Mike's book in the time that we have today. But why don't we start with sometimes it pays to be a chicken. What is the game of chicken, Mike, and how should you play it?
- [13:11]
Professor Michael Wooldridge
Well, so the game of chicken, the name comes from this game that was supposedly played by teenagers in 1950s America, that was immortalized in the movie Rebel Without a Cause, where James Dean's character Jim is playing a game against his nemesis, Buzz, and they're trying to prove who's brave and who's Chicken. And the way they do this is they drive their cars both towards a cliff edge. And basically the idea is that the chicken is the first person to jump out of the car, and the bravest is the last person to jump out. The problem is if they're both brave or they think the other person's chicken, they're looking over the shoulder waiting for them to jump out, then they both go over the cliff, and that's the worst outcome for both of them. So in a game of chicken, it pays to think very, very carefully about whether your opponent is really brave or really chicken. If you really believe they're chicken, then you should drive straight. You should hold on for that second longer. But you better believe, you better get that right. If you think that they're brave, you know, if you think they're just going to carry on driving straight, you don't want to go over that cliff, you may as well jump clear. I mean, it's not the best outcome for you. The best outcome is where you look brave, you know, where you're the last to jump out. But it's better than going over the cliff. So the point is, in the game of chicken, you want to judge what the other player is doing and then do the opposite. If you think they're going to be brave and drive straight, you should jump clear. If you think they're chicken, they're going to jump clear. You should drive straight a moment longer. But you need to think very, very carefully about your opponent. Now, the problem is in the game of chicken that if you misjudge that, you can both go over the cliff and then both end up with the worst possible scenario. Now, the game of chicken is fascinating because of the behaviors that we see in a game of chicken when people actually play the game of chicken in the real world. And we see lots of these. For example, I talk about how Brexit, the game played between the UK and the eu, is a game of chicken where basically both sides are saying, we are going to be brave, that is, we're going to push for a hard Brexit, or from the EU EU point of view, no access to our markets, without freedom of movement and so on. The worst outcome there is if nobody backs down, that actually if both players play a hard Brexit game, then the EU doesn't get to sell their cheese and wine and cars to us and we don't get to sell our umbrellas and warm beer to them. So what you see in the game of chicken play out is signals that the players make to each other about where they are heading. So when Jim and Buz play this game, the signals might be, you know, ah, you're chicken, I'm brave, I'm going to head straight. No matter what you do, you're going to lose this game. I'm the. I'm the bravest here. That's signaling. That's classic signaling in a game of chicken. And we saw exactly that in the Brexit game. You know, the announcements that came out of 10 Downing street about, you know, we will not back down, you know, we demand, and so on and so forth. And exactly the same from the eu. They're sending signals about which of these two outcomes they're going to go for. The danger, the classic danger in a game of chicken is escalation. That is, you send out these signals and these signals get stronger and stronger until both sides become irrevocably committed. So in this case, where they both pass a threshold from just sending signals to making commitments, like if both sides in the EU Brexit game of chicken, both sides commit to play hardball, then that ends up with a very bad. Metaphorically, they've both gone over the cliff. In Jim and Buzz's case, how might they commit themselves? They can throw out their steering wheel, they can sellotape a brick down on the accelerator, and so on. And these are then commitments which force them into a particular course of action. Commitments can be really powerful, but if you're gonna use a commitment, you have to get in first and make sure that the other player has seen. And. And the danger of continual escalation is that both sides inadvertently end up committing. They both go over the cliff. So the game of chicken is a classic game theory game. And the point is it happens all around us. So what's happened In Ukraine since 2022 in particular, the beginning of that looked like a game of chicken between NATO and Russia where Russia was sending out very strong signals, making thinly disguised threats about the use of nuclear weap weapons and so on. Classic signaling. And actually the invasion itself was a very, very strong commitment move. Up until then, it had been a game of signaling where we believed there was going to be an invasion and both sides were sending signals to one another. But Russia invading became an irrevocable commitment at that point. So one of the life lessons here is understanding the dangers when you see a game of chicken play out around you. And the classic danger in a game of chicken is mutual escalation until you're both fully committed and you both go off the Cliff.
- [18:37]
Carl Miller
But I suppose a big danger, Mike, there is that there's every incentive for your counterparty to appear to be as to try and appear to be as brave as possible. So poker's a great example. I mean, it's a bluff Cuban Missile crisis. I mean, students there will be pouring over the White House transcripts where JFK and his team are desperately trying to work out what the signals from the Russians really are. Were the back channels really actually telling them something from the saber rattling that's happening formally coming from the Russian embassy? How do, especially when it's in say, a life changing decision or a nuclear brinksmanship scenario, how can people really sort real signals from bluffs? Like, is there any way of doing that? Is it really just about kind of sorting the genuine commitments where you throw the steering wheel out the window from less meaningful sorts of displays of strength?
- [19:39]
Professor Michael Wooldridge
Yeah, and the mathematics of game theory doesn't have too much to say about that. I mean, but yeah, in reality that's exactly what you're gonna do. What is the serious, you know, what are their options? Seriously, what, what are the consequences are they really committing at this point? For example? That's a really important point. Is there a commitment here or is it just a signal? And a signal carries no weight other than what you're trying to do is send a message to the other, but there's no commitment there. There's nothing which irrevocably commits you. Commitments can be very powerful, but also very dangerous. And if you make those commitments which don't get seen by the other party, then the outcome can be very, very bad indeed. But of course, the Cuban Missile Crisis, probably the most famous game of chicken in human history, was exactly that. And what you describe exactly happened. JFK and his team busy working out, is this really a commitment or is this a signal? And so on. And it turns out in the Cuban Missile Crisis, it seems that both players actually backed down, although it didn't quite look like it at the time, but actually both players made concessions where they stepped back away from the brink. And it was the biggest brink in that they were on the edge of. But there were many ways in which that game could have been misplayed. So, yeah, game theory can flag up these dangers for you and tell you what the indications are that you're in a game of chicken, or that you're in one of the other games that we talk about in the book and help you identify those and the dangers. And I think that's one of the chief values of it.
- [21:24]
Shopify Advertiser
Starting or growing your own business can be intimidating and lonely at times. Your to do list may feel endless with new tasks and lists can easily begin to overrun your life. So finding the right tool that not only helps you out but simplifies everything as a built in business partner can be a game changer for millions of businesses. That tool is Shopify Shopify is the commerce platform behind millions of businesses around the world and 10% of all e commerce in the US from household names like Gymshark, Rare Beauty and Heinz to brands just getting started. Shopify has hundreds of ready to use templates that can help you build a beautiful online store that matches your brand style and you can tackle all the important tasks in one place, from inventory to payments to analytics and more. No need to save multiple websites or try to figure out what platform is hosting the tool that you need. And if people haven't heard about your brand, you can get the word out like you have a marketing team behind you with easy to run email and social media campaigns to reach customers wherever they're scrolling or strolling. Start your business today with the industry's best business partner, Shopify and start hearing. Sign up for your $1 per month trial today at shopify.com realm. Go to shopify.com realm that's Shopify.
- [22:49]
Windows 11 Advertiser
Study and play come together on a Windows 11 PC and for a limited time, college students get the best of both worlds. Get the unreal college deal everything you need to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 Premium and a year of Xbox game. Pass ultimate with a custom color Xbox wireless controller. Learn more@windows.com studentoffers while supplies last ends June 30th terms at aka mscollegepc.
- [23:20]
Red Bull Advertiser
Ready to soundtrack your summer with Red Bull Summer All Day Play? You choose a playlist that fits your summer vibe the best. Are you a festival fanatic, a deep end dj, a road dog, or a trail mixer? Just add a song to your chosen playlist and put your summer on track. Red Bull Summer All Day Play Red Bull gives you wings. Visit red bull.com brightsummerahead to learn more. See you this summer
- [23:47]
Indeed Advertiser
when you need to build up your team to handle the growing chaos at work. Use indeed sponsored jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsored job credit@ Indeed.com podcast. That's Indeed.com podcast. Terms and conditions apply. Need a hiring hero? This is a job for indeed sponsored
- [24:15]
Home Depot Advertiser
jobs this Father's Day. Do more with dad and spend less. With low prices guaranteed at the Home Depot, get him fired up with a new grill and accessories like the next Grill 5 burner for just $299 so you can spend more time together while he becomes the grill master he was always meant to be. Or build memories with savings on top brand power tools so you can tackle projects side by side, gift more and do more together this Father's Day with help from the Home Depot. Exclusions apply to home depot.com pricematch for details.
- [24:52]
Carl Miller
Well, let's move to another game. Let's move to the prisoner's dilemma, a famous one, one I think that people might have heard of. You know, you argue, Mike, that, you know, they're everywhere, these prisoners dilemmas. You know, climate agreements, fishing quotas, arms treaties, Coke versus Pepsi advertising. How can people recognize when there is a prisoner's dilemma? And how do you use that? You know, how do you use that for your advantage?
- [25:19]
Professor Michael Wooldridge
Well, so the classic characteristic of a prisoner's dilemma is that there is a temptation to cheat on an agreement where you and I can reach an agreement, but it's not a binding agreement. It's just me saying to you, okay, let's do this, but where I have a temptation to cheat, that is where I would strictly do better off by cheating, no matter what you did. But in a prisoner's dilemma, you have exactly the same temptation. Now, the problem with the prisoner's dilemma, if everybody is faced with that temptation to cheat, then actually everybody ends up cheating and everybody ends up worse off as a consequence. Now, one of the big lessons of game theory is that if you really are playing a prisoner's dilemma, then you're not going to get to an outcome where nobody cheats without a bit of help. That is, we would be foolish to sign agreements around, say, reducing fishing in the North Atlantic without actually, you know, enforcing that with spot checks and the kind of other tests to make sure that people are actually adhering to the agreement that's been made. Because without that, it's just noise. If there are no teeth there, it's just noise. It doesn't cost me to anything to promise something to you. And if I know that if you cheerfully cooperate while I cheat, I'm going to be better off as a consequence. And in the prisoner's dilemma, that is the choice that the players could make. So the temptation to cheat is the characteristic, the really characteristic of a prisoner's dilemma. And the same thing happens in public transport. The best outcome for me in public transport is if you all buy your tickets and I cheat, then you have all bought your tickets. We have a well funded public transport system, but if we all cheat, then we end up with a poorly funded public transport system. TV licenses in the uk, same story. Best outcome for me, you all buy your TV license while I quietly cheat. By the way, I have got a TV license. So that is the characteristic of a prisoner's dilemma. If everybody cheats in the television licenses game, then we end up with poorly funded public television services. In advertising to Coke and Pepsi, why do they invest in advertising their products? Everybody knows Coke, everybody knows Pepsi. The danger is that if one side decides, well, I'm not going to advertise this year, the other does and the other gets a huge market advantage. So they both end up spending money on advertising without actually advancing their market position at all. They've got the same market position they started out with before they spent a penny, but now they've paid advertising executives a huge amount of money. They're both worse off. But if I'm Coke and I think you're not going to spend any money on advertising, the best thing for me is to spend money on advertising. So the prisoner's dilemma arises time and time again. And actually we mentioned AI at the beginning. There's a kind of a prisoner's dilemma in AI right now that we've got some of the richest companies in the world that are racing to develop AI that apparently they don't believe is safe. Why are they doing that? Because if they believe, if they didn't do it, somebody else would cheat, somebody else would go ahead and do it. And it's been described as a prisoner's dilemma. But the characteristic of the prisoner's dilemma is where there is that temptation to cheat. Everybody and everybody has that temptation. Everybody cheats and everybody ends up worse off as a consequence. The problem is, if we are playing this game without binding agreements, without some mechanism to enforce cooperation, to enforce the agreement that we make, then game theory says you're not rationally going to get to cooperation. You need some other mechanism there to help you get to cooperation. But actually that I think is an important lesson of game theory. Don't be a sucker. Sucker.
- [29:10]
Carl Miller
Don't be a sucker. Indeed. Well, there are two life lessons that are somewhat mirrored. Sometimes it pays to lie, sometimes it pays to tell the truth. And I guess that begs the question Mike, which is a general one, I guess lots of these. How. How do you know which life lesson you're in that you should be paying attention to?
- [29:29]
Professor Michael Wooldridge
Yeah, well, sometimes it pays to lie is really about the question of strategic voting. And we saw strategic voting in the UK in a very big way in the last general election. I mean, I think that was one of the big factors that we saw lots of websites advising people, if you want to get rid of this particular party in this constituency, then this is who you should vote for. And that strategic voting is essentially voting, but where you're not voting truthfully, where you're casting a vote for somebody who is not actually your most preferred candidate. Now, there is a sense in which you're lying there. And this problem has been studied. The following problem has been studied. Are there voting systems like the one that we use in the uk? Single. The one we use in the uk, Plurality or single transferable vote, which is used in Ireland and some other countries. There are many different voting systems. Is there a perfect system in the sense that you would never benefit by doing anything other than voting truthfully, voting for your most preferred candidate. Now, without spoiling the book for your listeners, there is a result which says no. Unfortunately, whatever voting system you come up with, there will always be circumstances in which tactical voting is the best thing for you to do. In other words, voting really is a game. Now, we see the large parties tell us, no, no, you shouldn't vote tactically, you should always vote effectively, truthfully, because they know they benefit from that, because most votes in that way are going to go, you know, for one of the big parties. But the lesson from game theory is voting really is a game. It makes sense to think strategically about how to cast your vote to get the best outcome for yourself. And that means sometimes the big lesson there is, there are always situations where it makes sense to vote not truthfully, not for your most preferred candidate, but for somebody else, because that's the way you stand the best chance of getting this out. Outcome. And then sometimes it always pays to tell the truth. It turns out that we use auctions nowadays all the time in our lives. Ebay is now a big part of many of our lives. We routinely buy and sell things on ebay and other online auction platforms. But it turns out there are some auctions, some particular types of auctions, where actually it turns out the best thing to do is to bid for something exactly the amount that you think it's worth. In other words, words, where you are strongly incentivized to Think exactly how much do I think this item is worth and bid not a penny more and not a penny less. And again, the striking thing about this is that's telling the truth about how much you think it's worth. You've got actually got a strong incentive to bid exactly what you think it's worth. In other words, to tell the truth about how much you think it's worth. Which is a kind of a remarkable result. And there are some very good reasons why actually we very often want auctions with exactly those kinds of properties, and I talk about those in the book. So these lessons are if you're in an auction like that, and they do occur, and actually auctions like that are used to decide which adverts we see when we do Google searches, for example, so that they really do occur. And we're all regularly faced with the problem of voting and how do we cast our vote. And game theory provides pretty clear advertising advice for that. So which game are we in? I hope that in the book what I do is I convey to you the range of different scenarios and the kind of the game theoretic advice in each of those. And as we mentioned, each of the 21 lessons starts with the classic story from game theory to illustrate those, like the game of chicken or the prisoner's dilemma or the voting, you know, the strategic voting problem and on. But remember, game theory is just a bunch of mathematics. It's neutral. It doesn't say how you should cast your vote, it just says if you want to get the best outcome for yourself, then I'm afraid sometimes you're going to have to vote strategically. If you don't vote strategically, then sometimes you will be casting a vote and potentially end up worse off as a consequence. You'll cast a vote truthfully and potentially end up worst off as a consequence. So again, a lot of people think game theory is providing kind of moral advice or advising them to be dishonest. That's not the point of it at all. All.
- [34:11]
Carl Miller
Well, speaking of bundles of mathematics, let's talk about AI. How foundational, Mike, are these 1920s ideas of John von Neumann to the AI that kind of people use every day and that touches people's lives every day.
- [34:30]
Professor Michael Wooldridge
It's absolutely pervasive throughout AI in a range of different ways. So one classic example was Deep blue from the 1970s, the first chess playing program from IBM that beat a world champion chess player, a grandmaster, under controlled conditions. And it was really, it was one of the landmark breakthrough moments in AI. And why was it a Landmark moment in AI, because chess is a game that human beings find hard to play well, and we associate playing well with intelligence. And so the fact that we had an AI program that could be a human grandmaster was a big moment in AI. But actually, underneath the hood of that chess playing programs are techniques that were developed within game theory, which I talk about in the book. There's a technique called backward induction, which is used for analyzing games. And if ever you've played a game of chess or checkers or even a game of Tic Tac Toe on a computer, you will have encountered the. That algorithm. That technique, which was invented by, was invented by game theorists. And it's a very simple technique for analyzing board games like that. Now, you need to do a bit more work when you want to play a game as difficult as chess, but it's absolutely central, absolutely central to how those programs, those AI programs, will play a game of chess. And the other big idea which John von Neumann invented almost quite casually, almost as a side product of what he wanted to do, was an idea called expected utility. And what John did in that, what von Neumann did in when he invented expected utility, is he told you the best way to play when you are uncertain about what the outcomes are going to be. And what he said is, roughly speaking, this is simplifying. You'll have to read the book to find the full story. But roughly speaking, what he said is the way to play optimally is to make the choice which would give you the best outcome on average, if you played, made that choice a large number of time. And that's called the expected utility of a choice. The average amount that you could expect to earn from making that choice. But that idea, expected utility, I'm looking in my bookshelves, I'm looking at a couple of hundred books on my bookshelves. I don't believe there is one of those books that doesn't have that idea, expected utility. At its core, it is completely ubiquitous, not just in economics, but within AI. And when you play a chess program or some other AI program, what it's doing is it's trying to figure out what would be the best outcome on average for you. The outcome which maximizes expected utility, in von Neumann's terminology. So game theory is embedded completely throughout ideas. And it's remarkable how many of those ideas, ideas go right back to John von neumann in the 1920s.
- [37:40]
Carl Miller
You said earlier that game theory isn't a kind of moral compass. It's a kind of framework for optimizing decision making. Do you think that's why laypeople, kind of non AI researchers, when they kind of contemplate AI, there's so much nervousness. I mean, obviously the kind of speed and capability of AI AI systems is clearly part of this. But the idea that you've got at the heart of many of these systems a theoretic structure that isn't really interested in moral values.
- [38:19]
Professor Michael Wooldridge
I'm thinking here, Michael, a choice to maximize its expected utility. And the big problem with that is if you've given it the wrong model of utility to describe the technical terms, if you haven't described what you want appropriately, then things can go very badly wrong. One of the ways in which they can go wrong is what's called when we set up, when we have a machine that learns what to do and we give it feedback through what's called a reward mechanism. And the reward mechanism basically says, that was good or that's bad, thumbs up or thumbs down, you did a good thing, you did a bad thing. And this is in an area of AI called reinforcement learning. That's how AI programs learn. The problem is if we set up the wrong reward mechanism, it can learn to do things that we didn't actually want it to do. It can learn to what's called reward hack. It can learn to get a reward according to this reward mechanism, but in ways that we didn't expect it to do. And this is one of the fundamental ways in which AI can go wrong. But any we have to communicate to these machines what we want them to do. And that bottleneck for that human machine bottleneck is really one of the big bottlenecks in AI generally and how much
- [39:42]
Carl Miller
progress is being made there. So when we look at the different threads of AI development, kind of values alignment and kind of helping machines understand our own kind of moral systems better, how does that stack up in your eyes versus the other kind of areas of kind of red hot technical progress? Progress.
- [40:02]
Professor Michael Wooldridge
It's been a very, very active area for decade or so of. It's sometimes called the value alignment problem. Ensuring that the AI that we build has values which align with human values. That raises other questions, by the way, exactly whose values do you want it to align with? Mine? Yours? Elon Musk's? Donald Trump's? Whose values? And actually, it's not obvious that there is a consensus on what those values are, but never nevertheless, this problem of value alignment is a very, very well recognized topic. One of the challenges that we've got at the moment is some of the most successful examples of AI that we have right now, large language models like ChatGPT and the like were built by, just by training them on essentially all of the digital text that's available in the world, the whole of the World Wide Web, which includes, by the way, all of Reddit, all of Facebook, all of Twitter and so on. And along the way, those AI programs are going to learn every kind of toxicity that you and I can imagine. Every kind of human bias is present on Reddit, every kind of obnoxiousness is out there on Twitter. And the AI doesn't necessarily. It's just learning using this as raw training data. Once you built that model, all that stuff is essentially latent within the model, and all it requires is the right prompt for that to be exposed. So you have to imagine that you have these sort of core models which are then being wrapped in what are called guardrails by the big tech companies. And the point of a guardrail is to prevent any of that nastiness coming back to the surface. And the fundamental problem is that guardrails at the moment are rather flimsy and they're far, far, far from perfect. So this, this is an ongoing challenge, and I think it will be for the foreseeable future, building machines which align with human values and human preferences. And that machine, human bottleneck. We just take so much for granted because we've spent our entire lives growing up in a social system which trains us about what's acceptable, what's not acceptable, types of behavior. And before we came into this world, there was also billions of years of evolution which were also doing a similar kind of, kind of process to build apes, great apes, us, me and you, that have certain social skills, the ability to operate in a society of other great apes. And we have certain hardwired social skills as a consequence of that. You know, the fact that we groom each other, we spend a certain amount of time grooming each other and so on, and looking after each other. Seems like we're born with certain fairness norms, hardwired even to us. Exactly what those are is ongoing research. Machines don't have any of that. And just the idea that you could throw a bunch of text at a machine and it would pick all of that up magically, I think is very naive. So this is going to be a big ongoing area for the foreseeable future.
- [43:08]
Carl Miller
Well, timekeeping, sadly, has to be one of the things that's hardwired into me, Mike, as the host here, and we sadly are out of time, but thank you so much. What an interesting discussion and amazing to think that these ideas from the 1920s and from that great genius John von Neumann are still shaping the world in ever more powerful ways today. So that was Professor Michael Wooldridge, everyone. He's the author of Life Lessons From Game Theory, the Art of Thinking Strategically in a Complex World, which is available now online or at a bookshop near you. I've been Carl Miller, you've been listening to Intelligence Squared. Thank you so much as ever for joining us.
- [43:47]
Mia Sorrenti
Thanks for listening to Intelligence Squared. This episode was produced by Margarita Volpatto and it was edited by Mark Roberts. For ad free episodes and full length recordings. You can become a member@intelligencesquared.com membership and if you'd like to join us at future live events, you can head over to intelligencesquared.com attend to see our full events program and buy tickets. You've been listening to Intelligence Squared. Thanks for joining us.
- [44:26]
Carrington College Advertiser
Your next chapter in healthcare starts at Carrington College's School of Nursing in Portland. Join us for our open house on Tuesday, January 13th from 4 to 7pm you'll tour our campus, see live demos, meet instructors, and learn about our Associate Degree in Nursing program that prepares you to become a registered nurse. Take the first step toward your nursing career. Save your spot now at Carrington Edu Events. For information on program outcomes, visit carrington.
- [44:53]
Sweetgreen Advertiser
Edu Sci There's a new way to sweetgreen Meet Wraps Handheld, hearty and made for life on the move. With bold, chef crafted flavors, fresh ingredients and over 40 grams of protein, they're built to satisfy without slowing you down. Try wraps today in the app or@order.sweetgreen.com available at all participating locations.
- [45:26]
Shopify Advertiser
Starting a business can seem like a daunting task unless you have a partner like Shopify. They have the tools you need to start and grow your business. From designing a website to marketing to selling and beyond, Shopify can help with everything you need. There's a reason millions of companies like Mattel, Heinz and Allbirds continue to trust and use them. With Shopify on your side, turn your big business idea into sign up for your $1 per month trial@shopify.com specialoffer the
- [45:55]
Blinds.com Advertiser
right window treatments change everything. Your sleep, your privacy, the way every room looks and feel feels. @blinds.com We've spent 30 years making it surprisingly simple to get exactly what your home needs. We've covered over 25 million windows and have 50,000 five star reviews to prove we deliver. Whether you DIY it or want a pro to handle everything from measure to install, we have you covered. Real design professionals, free samples, zero pressure right now. Get up to 45% off site wide plus get a free professional measure@blinds.com rules and restrictions apply.