How AI impacts chess and the game of World Chess Champion Magnus Carlsen
2:49PM May 28, 2020
BR Good morning. Good afternoon. Good evening, and welcome to the AI for Good Global Summit over here, always online. We hope that all of you, your friends, family and colleagues are staying safe and healthy. My name is Kseniia from them from the A to the International Telecommunication Union. And I have the privilege of introducing today's webinar on how AI impacts chess, and the game of Walters champion Magnus Carlsen. Now the ICU is the United Nations specialized agency for information and communication technologies. And we are also the organizer of the AI for Good Global Summit, alongside XPrize foundation and in partnership with 36 un sister agencies, ACM and Switzerland, our strategic partner. The goal of the summit is to identify practical applications of AI to advance the Sustainable Development Goals and scale those solutions for global impact. And like the most of the world, the AI for Good Summit has gone digital with the weekly online programming, allowing us to reach even more people throughout 2020. And before we start, we'd like to run a quick poll just to get a feel of the audience. So now on your screen, you will see the first question. Do you ever play chess online? And you can select yes or no. You can always use the chat and q&a functions to communicate. our moderator will select and read out the questions to the panelists. And we have cost accounting on your active participation to create a very interactive session. So you have few more seconds to respond. We hope that we have a lot of chess players. So now we can see the results. Wonderful 78% of the audience to play chess online. So I guess we have some rough monsters in the audience again as well. So please let us know through this chat function. And now the second question that you will see on your screens. Have you ever heard of any of the following chess computer problems, and you can select multiple answers stockfish alpha 00. And again, you have a few seconds to respond. And we also invite you to let us know from where you're calling from, for instance, I'm calling from Geneva. And actually, let me write it down in the chat. So please use the chat and the q&a functions to communicate. We also invite our young audience if you are under 20. Please let us know when you ask your question. Fantastic. So we have people from India, from Jack's from Barcelona, or now Saudi Arabia. Wow, impressive, everyday international audience from Russia previous France for Drew. Perfect. So I guess we have few more seconds to respond to now we can see the results. Great. So 71% of the audience heard about stockfish Alpha zero 82 Leela to zero 52. Thank you very much for participating. And now it's time to introduce our moderator cannot cook here who is a New York Times bestselling author, senior editor at the economist and the host of garbage from economists radio, a weekly podcast on technology and science. Ken, welcome.
Hello, Sr. And thank you very much. And thank you everyone for joining. It is a pleasure to have so many people online to discuss this incredibly interesting topic. You know, the world of AI is undergoing a revolution. Well, this revolution in the world of course, because of AI, but AI itself is undergoing a revolution going through lots of changes and what's really special about it is From the almost the very beginning of AI, chess was in some ways, if you will, the sort of fruit fly of AI. And the reason why it was the fruit fly of AI is because you could track its evolution like we tracked the genetic evolution of the fruit flies when we did experiments with it. And what was so special about it is that we could measure our progress in artificial intelligence through the quote unquote, fruit fly of chess, and how good we were at it. And of course, for a while, we weren't very good at it. The whole idea of machine learning this revolution that we're going through right now in society because of technology, machine learning, its origin was actually a boarded game, the same eight by eight squares. It was checkers though. Arthur Samuel created this in the in the 1950s. Working at IBM, that would actually look and create a prep probability table very intricate of what move was better with most likely To a winning board versus a losing board. Since then, of course, deep blue in the 90s Casper off, and then alpha zero just a few years ago from DeepMind. Today, we are very lucky to have two formidable people to think about these issues on our panel. Let me introduce them to you right now. The first one is Sebastian Kunar, and Sebastian is the CEO of chess x, which is the web's biggest chess ecosystem, and also of just 24, the top chess broadcasting site and the digital home of the world champion, Magnus Carlsen. Now in addition to Sebastian, we're very fortunate to have Peter Jaime Nelson. Now, Peter honey himself is a grandmaster. He earned that title in 1994. He competed a lot in the Danish chess championships in the 90s and 2000s. And, and won five times. But what's really special Is that he has been the coach of Magnus Carlsen since 2013. And before that he coached his Watson on end between 2002 and 2012. So we're really pleased that Peter Hynek can be with us as well. Most importantly, a lot of things to people the it You deserve our thanks, as well as you for turning up and showing. But you can then thank all of us by giving us lots of great questions to respond to in the chat function. And we're going to go on to questions very quickly. But before we do that, what we're going to do is hear from both our panelists and to start us off, I'm going to ask Peter, honey Nelson, to tell us a little bit about the history and technology of AI and chess. Peter. Hi, please.
Hi. Thanks for the very kind introduction.
You're right that
well, chess has very much been the sort of basket Round of AI is basically a perfect way to test, you know, the battle between human intuition and against the massive computing power of calculation power after after computers. And this was especially big, as you mentioned, with the Kasparov Deep Blue battles that sort of went back and forth. The first one was actually won by Kasparov. And the second one was won by Deep Blue. And this was an incredible prestige project, which IBM putting a lot of effort into, and they managed to beat them in 1997. But then, basically, we have some kind of I would call it a more landing effect in the sense of, well, the human the world champion was beaten by a machine. And that meant that well, there was no prestige in his left so there was still be sort of some kind of developments, but it basically came not to a standstill, but it was more connected to computers being faster some kind of development and in software, but basically we thought we had kind of cutoff And that chest was from an AI point of view, basically a solved project. But then in 2017 20 years later, something incredibly happened. And from our perspective completely out of the blue,
well, we, every chess professional, we won't be working with the chess program called stockfish, which is basically an incredible calculating machine, you can calculate millions of millions of moves, probably billions. So within a minute, this would be the tool of all, you know, chess professionals, and we considered it the highest authority in the world. But then suddenly, in a sort of completely isolated world from us, a technology company called Deep mind, which is owned by Google, make created an app program called alpha zero, and he played a test match, I can stop fish and completely crushed it. And, well, if we didn't see it, we wouldn't think it was possible. We simply come at that time, we thought Office is very close to perfection. It's basically like having a pocket calculator that can calculate things perfectly. We simply had it as our biggest authority. But it got completely wiped out in a match against alpha zero. And this was, well, obviously incredible moment for us. You can imagine that suddenly, well, everything we used to believe was true, we suddenly had to doubt and such. And also, we had to understand very new things because, well, they were talking about self learning. They were talking about neural networks, and they were talking about tabula rasa, which was terms we have never ever heard about before. Also, if you look at the picture on the left, you can see that well, they did chess, but like IBM's, Deep Blue was a solely a chess project, what deep mind was doing is that they've made a general algorithm that to solve in principle, any kind of games, so it learns to play chess better than anything we've seen before but also shogi a complex Japanese games and go another complicated board game. And this was basically sort of mind blowing to us. And the way they did it was also very shocking. The sort of story had always been, we're talking human intuitions against machines calculations. But what they actually created is a computer with intuition. It sounds completely insane at the time, at least to us from the chess world. But it's very much what it's done. And well, I've taken a slide here from a book called Deep Learning and the game of Go. And well, it's the example is a very stupid game. Well imagine that we can choose a number from one to 500 times and the one who gets the highest number in the end wins. Well, any human can, of course, do mathematics. So we will just choose five every time I will get to 500 and win. But for a computer that can do mathematics, but only self learning, you can see the curve. It starts in the beginning, where well it chooses randomly, which means that 0.2 of the time To choose five, but then you can see keeps playing against itself. And it becomes clever and clever. It develops it into tuition, you can see on a straight line, so it's experimenting, okay, this work that didn't work, and it's a bit back and forth. And you can see that steadily, it creates a complete intuition that understands, well, I have to choose a five, basically all the time, you can see how having played by itself to size thousand times, it's gotten a perfect integration in this kind of very simple game. And that's basically what they did by chess, not playing 2000 games, but I think billions of games by itself, it started from this is also a very important point of how they actually changed it, that there is absolutely no human input, no human knowledge. They wanted basically to sort of create something that learns just solely by itself, to make sure that there is no kind of human bias. And that's actually very, the very interesting part that they can create. Basically a vacuum, where they are rerunning a sort of, let's say, evolution of the game of chess, but with no kind of impact. And the interesting things is that we got verification from a lot of our understanding of chess. But also some of the strategies we thought was great, was maybe not as great as we saw. And that's basically has shown sort of the power of deep learning and of machine learning that we can actually will get rid of sort of the human bias and look very much at that sort of, well, what the data suggests and which kind of knowledge that brings us And will the game I've put them the right side is a game from Magnus played against mammoth era in 2019. And I think it's a great example of understanding of chess before alpha zero and after alpha zero, I mean, in this position, before we got all the knowledge from alpha zero on the neural networks, we would say that, well, black should never push his porn from h4 to H three because that's close to you. line will be a strategic blunder. And you can easily create a narrative where you explain that as being a very bad decision. The problem is that we have this narrative, but this narrative seems to be wrong. Now, we would say that x ray is a very deep movie creates a long term attack instead of the short term attack you're looking at earlier. And that was quite some examples in Chester, we have simply learned from this will gain some new knowledge and overruled old understanding. And this is basically all machine learning has shown us that they can create computers for intuition. And this has really moved the understanding of chess forward and becomes a very interesting tool for us.
Thank you very much. That's a really helpful summary. To understand the changes that are going on. I'm really struck by the idea that you're using the term intuition, which is a term that is usually reserved for what human beings can do. So let me press Little further on this before I turn to Sebastian and ask, Is it really intuition? Or are they just instant machine just really, really good at probability in a ways that we can understand?
Well, what is intuition? I think we will say, human intuition that I will have a strong chess intuition. But that's because I had a, you know, 35 years of studying a lot of chess and being exposed to a lot of chess positions and remembering the result of these chess positions. But this is exactly what they do to this machines. They show them a lot of chess positions. And, and when the point is, this neural network is probably the piece that has seen the most test position in their life. So what they do is that they learn from seeing them and evaluating the results. That's basically intuition. I mean, well, the classical example is to show them a picture of a dog or cat and it has to get guessed it has a dog of cat. I mean, it becomes some kind of intuition that is getting that and it's basically the same the I have seen so many chess positions, many more than any human is possible to see in their life, and they know how to evaluate it or remember the patterns. I cannot describe it with any other word than intuition. I understand that. It shouldn't be possible. But well, I mean, you asked him the questions. I felt exactly the same in December 2017. And if it wasn't, because I have seen it work, I would say it's not possible. But it seems to
me I'm sure we're going to come back to this for the moment, I'm going to embrace your idea that it actually is a more enlightened, sophisticated pattern recognition, rather than intuition, because I'm going to want to come back to the idea of what humans can do in chess and in life, versus what artificial intelligence can do in chess and in life. But let's start off first and bring Sebastian into it with the macro view, which is how is AI disrupting the game of chess? Sebastian, let's hear from you.
I kind of Yeah, thanks for having me. Yeah, having a big impact on chess, as we've already heard from from Peter hyena having a big impact on Magnus game, the top players games, it's putting fundamental questions about what the entertainment value of the human and chesses versus the the machine and what the purpose of learning chess is. Versus Yeah, just watching a perfect machine play it. And we can derive much larger things from it. Beyond just like your you've already insinuated, and like you can see many of the questions are going towards this also general applicability of what we derived from you. In interests right now. It's, it's affecting all areas, you're working on error on anti cheating in this area where AI can be a big help, where many of our algorithms already AI inspired in this area. And working on the next generation of that, then also in Cest e learning, you're using it in the area of image processing for deriving learning applications from there at our other companies festival. And yeah, in a way you could say that this our whole play Magnus group was also inspired in the very beginning by the idea of creating algorithms that in some way shape or form behave like magnets at different age levels kind of limited to to that intelligence at that age, you could say, but, of course, this is this is further away from AI. But for to the to the broad mass market, it might feel like AI.
Well, let me let me pick up on that and ask you a question because it leads to a very interesting subtle way in which AI and chess or anything else can evolve. Do we want the world's best chess player In general, or do we want the world's best version of Magnus, right? Because what we could do is we could actually have alpha zero, just sort of divine, the strategies of chess that have never been known before and can do it. Or you can study all of madnesses games, and just be the best version of Magnus. And that might be quite interesting because it can, it will only be Magnus, it will only maybe be a little bit limited. But it'll be the sort of instantiation of this interesting, interesting person who has his foibles, but also will aspire to an excellence that that, frankly, probably has never been a team before in chess. But certainly, we do things that are new and novel and great within the boundaries of the AI trained on only Magnus Carlson's game. What do you think?
I think it takes a bit of wave like you're saying you're saying it's it kind of takes away the visibility to go beyond What what he knows so far about him?
In some way? No, I think it means that players might you might have players who will say, Hey, I only want to play Magnus because I want to play Magnus. But there's other players who say, you know, I know the Kasparov would have hit the dust if he was if he was playing Holton, you know, in a five game or 50 game match, but I still love Casper offs game of play game of play, in particular in the sacrifices he made playing playing styles.
Yes, cutting things off.
Yeah. Yeah. Yeah. I mean, I personally at least think that exactly what makes humans human that they they make errors is the interesting thing to watch in chess. Of course, there's also the, the broad string of people that that would like to see these perfect versions of a certain playing style playing each other. So it depends a bit where you want to evolve the game and if you see it as a mathematical expression of beauty or as a human emotion and mind. So you could have both and I think in a way, the the engine World Championships, computer engine championships that are that also exist in chess are like representing one of those strings and then things like the Magnus Carlsen chess tour that we're running now is maybe more goes more towards the shorter timeframes, shorter time controls and is pushing more towards the human emotion side.
Right, Peter, Jaime, how do you see AI chess evolving? And maybe start by answering the question that many people are asking what does it tress a chess trainer or chess coach actually do?
Yeah, good question.
I think he does like in most other sports, I mean, while he helps the player develop within their fields and sort of sharpen their skill set. But of course, chess is a very specific sport in the sense that, as we debated the influence of the computers now the influence on AI, and well, from my perspective, it doesn't matter if I like it or not, it's a premise, it's there. So we have to be able to use it. And, well, I mean, that's more or less than the point that the players are playing against another human being. But we have computers, we can evaluate their play, you can find leaks in their play and suggest new ideas. So everybody is using computers and AI aggressively in order to try to get an edge towards your opponent or try to make the disadvantages. Lesser so and whenever some of Microsoft's opponents are playing a game. I will of course try to analyze it with my version of AI. Why wouldn't Microsoft playing all his competitors will try to see okay, maybe we can attack macros like this to the next time. So and especially before, there was a one Championship match, I mean, magnets team as well as a team of the opponent, they will look down in, you know, separate camps, they will train for half a year, they will try to map out their strategy, you know, basically becomes an arms race. I mean, you know, we try to develop strategies, we know that they're going to surprise us and we will try to surprise them, no surprises, and etc, etc. And we'll never use all this kind of electronical tools simply because they are so powerful. I mean, maybe it was more charming in the 60s, where you will get your, you know, opening ideas going for a walk or something like that, but, but that's how the world is in 2020. And we don't really have a choice. So that's what we do.
So I love the idea of going for a walk and getting ideas for a new opening or for for a classic way to go take the mid game to the end game. But it of course raises the question that as we play against artificial intelligence, and artificial intelligence does different things that we would have expected. It may affect how we think about chess and then how we strategize and play the chess So let me pose a very direct question to you, Peter, honey. How has Magnus Carlsen applied artificial intelligence to change the way he plays chess?
I mean, he has learned from watching games from it, right? I mean, we suddenly had this new technology with with a ton of knowledge. And we can see that the only played certain games. And well Magnus is incredibly intelligent and incredibly adapt. So he will see that Okay, I have there is this correction in my chest understanding, maybe I can use that. And he was very quick at the Well, I mean, when I showed the slide, there was a book called game changer, which I think was out in the beginning of 2019. And we had a very good period afterwards where he were using some of these ideas aggressively. I mean, basically, what has happened sometimes in chess, is that the defensive resources becomes too strong, if well, you know, with the previous technology, we thought we were solving more and more problems. And basically, it was difficult to get the game going, suddenly there was this new splash of energy with the I mean, maybe no, we start off well, you can imagine everything is suddenly up in the air, it's not clear what is right and what is wrong. And that heavily favors the better player because, well, you know, things get out of patterns. And suddenly, macros had much more room to operate, and that he or he used very aggressively and now somehow AI has breached a sort of given some fresh air into chess, where I mean, people were talking about everything was ending up in drawers and suddenly became much more exciting. Well, now of course, people are starting to understand it better and well, it's a bit on the downside again, but it's clear that it has sort of renewed chess in many ways and any kind of new burst of energy will favor the stronger player Magnus is a stronger player, so he's been extremely good at grasping the ideas and sort of employing them quickly towards his opponents. Okay,
I see do believe what you say. But if AI was so good, why do you still have a job?
Yeah, that's a good point. I mean, well, I still go for walks with him and I play basketball with him. But also, I mean, you don't need someone to interpretate the AI. I mean, Well, imagine we are. It's a good question. We are playing a tournament, right? I mean, Magnus has to play an important game tomorrow. He still needs to get the eight hours of sleep he needs to be completely fresh. You need someone who is well, whose night's sleep is more dispensable to actually work with the AI at night and such maybe you can create AI that can do that as well and that will have to do something else. But I mean, you still need someone to interpretate it and well imagine we go for a walk and macros has some chess ideas. Well, it's my job to insert it into the AI to debate with AI have been macros have a point or not can be used as tomorrow and such or can we use this and a half, half a year so I mean, that's still needs to be some kind of assistance working like that. So I mean, at the top level, though, is still useful, guys like me, but I think it's a good point to have that. Well, in 50 years, we might have a problem, but that's good. It's not the youngest species will go extinct, I'm afraid.
Okay. Well, I think we're going to come back to this because I want to probe you more about the issue of the human versus the machine. But let's turn to Sebastian, let me ask you, how is AI changing what you're able to do on your platforms to help young learners at chess become better players and turn better players into great players?
Yeah, so um, it's on test platforms. Nowadays. It's, it's about a range of things. So on the one hand side, you have the post game analysis and recommendations that you received after your game. So I'm just 24. When you play a game of chess, it will analyze which which openings you have played and how you've behaved during the game and record Do the right kind of video courses that we have on our website.
the play Magnus apps where we have these different playing styles like that you looted to earlier that aren't AI in the, in the strict sense but they but they can they can be expanded into this direction and kind of have an algorithmic basis that makes sense to to look into for for this for the future. And then it's it's about a lot about the the error finding of like, how do you how do you tell someone what what they could have done better and from there, and the question is, do you tell them directly to improve this, this or that? Or do you derive from that the recommendation for example, which human coach they should be working with. That's then the mixer. That is kind of our strategy. So we are we're launching shortly a life coaching platform under coaches.com. where people can after having their games analyzed find the right coach, for example. And that is basically what we believe in at this point that the mix of like what what Peter highness doing from Magnus, the mix of humans using AI to derive implications for other humans is the winning way to go. And and until we reach a different AI a paradigm for what is beyond anything.
So we have several questions from the chat that's online right now. And I'm going to ask the first one, and it comes from a person in Nelson. And he asks, Are grandmasters trade trading themselves against AI opponents? I suppose not just using AI for training but learning actually how to play against AI eyes
Yeah. I think we are not really playing training games anymore. I think also that well, in 1997 computers beat the world champion at the time, somehow playing games against computers has stopped becoming interesting. I mean, we lose to our phone, I have lost to my flight seat, for instance. I mean, there is simply, I mean, they are too strong. It doesn't make make sense anymore. But we can debate chess with it, and we can analyze our games and well, it basically, you know, you have some kind of creative human idea, you show it to them and you get their word. And you can sometimes you can convince them they say, okay, maybe it's bad. And you say, maybe what about this moment? Well, you have some kind of dialogue with it. And that's actually the interesting thing with the the neural networks that well, we were used to this that sort of would often refuel our chests in a sort of brutal way because he could calculate way far. But now we can also debate as I say, I mean, by analyzing but the the neural networks who will show some kind of intuitive feeling and that sir, I think a lot of people are learning from that at the moment reading books, analyzing with them and so on. And I mean, there is a lot of knowledge there, but also Well, it's a knowledge that you will unlock by watching them or by basically asking them questions. I mean, you know, it's a passive unit, right? You have to sort of push it to be able to understand its, you know, possibilities and the knowledge to extract from it.
There's another question that we have and let me It builds on what you've just answered in hiding. So let me let me state it which is from Vladimir Putin. And he asks, Is the game of chess where we can evaluate the AI decision an AI can evaluate the humans one, does it help to make AI and humans become closer in future terms?
Again, this is too abstract
question for me to solve I'm, I'm a chess analyst. The sounds more philosophical in some way. I would say it's, it's hard to say. But I think that well, the neural networks definitely has become much closer to the way human things than how it was earlier. So I think they also helpful to, to, to understand how humans understand the chess position on such but I think, as far as I understand they are still a bit like the human brain that you can see the output but which process goes on exactly? It's more difficult to to define.
Let me try to maybe take a stab at an answer. And to say that there's an imperfection in AI. And it has to do with values, because AI doesn't know what the values are doesn't have a sense of values, and why would it it would need to be encoded in tribal values, and you'd need a training set of data to teach it so you wouldn't so let me give you an example of those values, by way of an analogy, so in the Khan Academy, which is a online, free educational service where kids can take learn, look at videos and learn things, as well as take tests. What they found was that if you have students, if you hit people up with really difficult questions all at once the student becomes demotivated, becomes psyched out, becomes disappointed and therefore doesn't return. But if you start them off and have them get a few questions to build their self confidence, they'll do really well. And other online platforms have noticed that as well. And within the AI, you'd be able to find out that you'd be able to increase engagement, and speed up the learning of someone who's learning chess to become good to become great. By allowing them to win. You don't just always crush them all the time as it's interesting. But here's the deal. You wouldn't know to build that feature in to support humans self confidence, unless you had the idea to do that, because AI on its own wouldn't have done that. So the lesson here is that unless we have human beings sort of in charge, To the AI and directing its values to AI just simply becomes a really good pattern recognition engine, but not really at these higher levels of performance in the way that we want to help humans in their lives.
But I mean, there's also the whole track of explained ability in AI, which, which is trying to solve this. Of course, it isn't solved yet. But that will that will have a big impact on chess as well. Well, so how so? Well, I mean, when. So, when, when you when you reach the level where you, you can explain some of the thoughts that I has, and in by discontinuing some of the some of the strings of thought but focusing on some others and visualizing those, that is also an interesting way of of learning and chess, and there are approaches have that going on and of transferring that to general decision making. And this is something that it's interesting for all people interest.
So I as you were speaking, Sebastian, another question popped up, and I think it seems to be relevant to this, which is, and it came fleetingly. So I don't know who asked it. It's in the chat. Is that Is he playing any games today? And it poses to me the question does does Magnus play any games not that are not chess for the purpose of improving his chess game a sort of, if you will, human transfer learning in which Katon or Tic Tac Toe makes him a better chess player.
Well, I can just quickly show so here you have our Magnus Carlsen chess tour with the landorus epi rapid challenges the tournament that is going on right now. And here we have today like the game of all games, so to speak, Magnus Carlsen against the corona Camorra starting at 330. And you can watch it directly on July 24. So that's the that's kind of the big innovation we have made during the COVID-19 or Coronavirus crisis to, to host the first big online tour and chess. And that's coming going into the semifinals today, the second tournament, but to go back to Magnus, I think he he likes to do things that that are basically not putting his bed that have them to relax. So he's very much on and off from Peter and he can give the better, better.
I mean, I mean, he's playing a lot of other games. I mean, I mean, we're playing football and basketball all the time. We also play various Cup games and such and I think, Well, I think that I saw some research person that also begs that should I quit, I mean, who is the most successful the kids who have done A lot of various sports are someone who specialized in one thing. And they often tend to think that those who is doing a lot of sports actually managed to get an abstract understanding of how to solve complex problems. And even though that they're not directly the same, you learn to solve a new kind of situation. And I think in that sense, it's quite helpful that you are capable to play all kinds of games. But of course, there are many games that are similar to chess. And probably, for instance, playing some game like shogi would not be so helpful for Magnus because the overlap is huge. And that means that some intuition you have in chess could actually get a bit altered by the inputs are probably you need to when you are the absolute top, you also need to keep somewhat distance for things that are similar but could somehow hurt your incredibly strong intuition. But I think also Magnus loves competing and that's a great way of doing and I think also competing is a great way of relaxing, and I mean, well. Not that it relaxes you, but in man if you're playing a football match, you actually stopped Thinking about you have this important time and you maybe need to play a world championship match. If you are just going for walks, you will think about it all the time. But if you engage in something else, you actually managed to rescue your brain a bit. And so he's doing a lot of competition, I think in order to, you know, train himself a lot, I think he's found this interesting, and it somehow relaxes.
So let me let me pick up on this and, and ask the question, is he learning? Or is he optimizing? And the difference is, does he need to become good and stay good? Does he need to learn new information? Or does he just need to optimize what he already knows?
Well, I would almost think that he only needs to optimize but that's maybe well, okay becomes arrogant, but I think Magnus is clearly the strongest chess player in the world. So it also becomes a bit of a matter of what is your objective? Do you want to stay the best or do you want to maximize your potential. Well, luckily, I think people don't really talk about these things. They do it also. I mean, we are not training in a, let's say organized way. It's not like let's say a football team where, okay, we're training from eight o'clock to one o'clock and we're doing this repetitions. I mean, chess is a somewhat creative sport. So you do things that are interesting. You read books, you follow, see games, Magnus will play games online and such. So I mean, Magnus has never been sort of extremely structured his mobbin, as he would call it, a product of passion. He's doing things that interests him. And basically, curiosity is that incredibly strong force. And I think a very organized person will get to a certain stage, but to really, you know, be the best. You have to love what you're doing. You have to be incredibly curious about what is chess what to discover the secrets and you also want to win. So I think it's not done on a conscious level. It's more simply that what Don't think you can stop himself. I mean, you can see that you might play a tournament during the day, but don't even play fun games at night and such. So that's how I think the best ones are basically living from it, I think, well, there was recently this TV series called The Last Dance about Michael Jordan, you can see that, well, he will be the first guy in the gym. And it's not like that is organized when you just can't stop himself because he likes basketball or chess, in this case too much. Okay.
There's a question.
That is, that is there builds on this, and it's quite cute, and it's by Eric van Marine, and he asks, What is more exciting to watch stockfish against alpha zero or two human beings?
Well, as the numbers speak very clearly to two human beings. Why do you think that is? Because humans like humans, humans don't love to watch something that is better than them and have no flaws. Like, just like, will we ever want to watch the FIFA version of two of two football teams against each other? No, like it's completely meaningless. Well, okay, I mean, maybe one day, maybe one day we say, Okay, we have abstracted all of our knowledge into these, and they make the perfect moves, but it's still, it's just one one level removed from us. And that removes empathy and therefore interest from us. But that's about it.
I mean, it takes a lot. I mean, even the Blitz game that's being played by Carlson and Nakamura takes an hour, but I could see, I could watch stockfish and alpha zero in four seconds.
Yeah, you can watch anything in four seconds. If you speeded up the recording. I mean,
I although we could speed up the recording, I chose four seconds as some of the people online might know, because that was the amount of time that alpha zero played each of its games. When he was creating its own training data, it was programmed to actually play each match of four seconds. And of course, it might have come up with a different strategy. If it was five seconds, or if it was limited to two seconds.
I both agree and disagree with Sebastian, I would say that, of course, watching human plays is the most interesting. I mean, well, you know, you wrote for one of the players, you have emotions involved and such. I actually think computers watching computers play against each other. It's very interesting. But I understand I'm a huge minority. I mean, we are 10 people doing it and discussing it and such. And it's, it's interesting on academical sense, because, you know, I mean, well, you can see, they try and push the boundaries of how you can play chess. And it's, it's very interesting in the scientific sense, but in a sports sense, in an emotional sense. Of course, Sebastian is correct that, well, it becomes quite insane to sit in a room for stockfish or for the whole whole day, right? I mean, well, you want to hope for the weekend or the American or the Russian and this guy or that guy, so As a spectator sport, it makes little sense but in an academical sense, and if you look at chess as an academical discipline, where you want to see optimized strategies, you can actually be interesting. I'll even say beautiful. But I mean, understand that Sebastian is working on making chess for the masses. And there that's probably a wrong turn about
that sense of but of course, like or even if you find that the from the beauty perspective, it's nice to watch it.
So we have a, we have a question from Abbe Marais. And what are they asked and I'm going to develop it a bit more is do you think AI will develop emotional intelligence in the near future? Now, let me develop this this idea of emotional intelligence. Ai doesn't no mercy and AI isn't cruel, but a great Grandmaster playing someone playing his or her lesser can take a game and where the person is going to lose and draw it out by an extra 1515 moves just to twist the knife. Fun and see the person squirm. Right. So I posed the question, will AI develop this emotional intelligence as a asks? And either? And if so, will it be merciful? Or will it be cruel?
I mean, I'm from the chess world, right? I mean, the reason chess is so good for, you know, AI research is that there is a result that is either one or half a point or zero. So it's extremely definable. As far as I understand this is, let's say, I mean, you also using neural networks and self driving cars, and they are these things actually come into play. I mean, imagine some very bad situation where the car actually has to choose, am I gonna hit the cyclist or the opposing opposing driver? And, well, what matters in this situation, Asians there, you actually suddenly have an ethical choice that has to be taken by a neural network and such. I have no clue how they do this. I'm a I'm a chess grandmaster. I operate in black and white and such. I mean, these are much bigger challenges, and you can maybe call them soft or more difficult challenges in many ways. I have no idea how to be honest.
Let me say that if there's anyone who is an AI expert, and not just a chess expert who's on the chat and and on our call right now, and wants to answer this in the chat, and also send a reference, if there's a if there's a paper that's relevant, please do because it's an interesting question. Now we have others as well. And one is an anonymous person is picking up on a question I've already asked. But I clearly we didn't i didn't press hard enough, and we didn't get a good enough answer. So let's build on it and look and say, the question is, what's the most surprising discovery that Magnus has derived from alpha zeros way to play?
Clearly clearly for you, Peter, yeah, just say on the on the previous point, like, probably also depends on who you Values who evaluates if something is cruel, or, or not? Because they I might might not find it a cruel.
No. I mean, well, it's a good question. It's also a bit of a personal question. I mean, Magnus is a competitive person. He's trying to win games later today. So to you know, give the exact secret of what is it the magnet is most impressed with by the AI would probably be a mistake and such but well, I mentioned some things that we've talked about the flank attack was very surprising. I think also, I written some article debating that, where we used to think that King safety was hiding the king away, then they can also be king safety in the open space and such. There is some definitely some some new things. But I think also, well, if people want to understand what Magnus has learned from AI, they should look at his games. I mean, he's trying to use it against the opponents. I think chess is a competitive sport. So we should Get out to have our secret. So Magnus answers these questions should be in the upcoming games not in a in a seminar. Sorry.
That's a very fair answer. Even if it's an unsatisfying answer, it's a very fair answer.
It's also a beautiful answer. Deliver the answer on the court.
Yeah, yeah, that's right. So, Sebastian, how is AI and strategy being taught? Can you actually when you have a coach, teach young learners, how to actually advance through the game, or your instructors learning from AI and actually improving their technique of teaching and training? And if so, what are they doing? What are they learning and how are they doing that?
So we have within our group, we have a number of products that take basically different roles to what's the sewn in. In inside of our app suite, we have an N called Magnus trainer, vre. Yet it's more of this creative approach of building your intuition. That is basically a guided path kind of the way the way Magnus also like to play around with with with his own just learning when he was younger. Then there's trestle, which is the largest digital marketplace for for digitized chess books, and applies things such as spatial repetition to, to learning to learning chess, and then yeah, that's basically the structured part maybe that you have alluded to. We're a human also tries to bring the best structure into each topic. And then all those structures are basically competing with each other for delivering the best answers to users in this marketplace and users can rate it and so on. And then coaches which is launching soon as then there are different categories of there will be different categories of instructors. And some the the highest rated or categorized ones will be going through a rigorous training program to deliver a certain quality standard in all their teaching that they that they do. And in that we are around that we are developing all the tools to make them the best coaches. And those really can be technology supported. And yeah, like that's, that's a thing, how much I can disclose at this point.
Now, we've discussed earlier this idea that how we design the platform and the values that we put into it affects the outcomes that we get. And one of the things I think a lot of chess players sort of are surprised about kind of some of pull their hair over is the fact that it's the last bastion of society where there's very little gender diversity. Where are all the women in chess why Is this a domain that is really still very X Y chromosome?
Yes. So it's a minefield to answer and but I'll just try my best not to get killed. And so then the it's open for everyone. But then we it's, of course be. We talked to Peter and I talked about this the last time that it has a bit to do with this self fulfilling prophecy that when there are a lot of men doing something, it might already be less inclusive for women that try to get into the sport. at the club level, for example, online might be more more open, because the gender is not so much in the focus when you're playing online. And then there's the other aspect of opportunity costs that when When,
a lot of a lot of women when they, when they have these skills, they also have a lot of other opportunities and and offers that they maybe prefer and that are maybe seen as less less risky. But yeah, it's in principle there's no reason why women should not be represented equally at the top of the chess world and V, VR, VR having a lot of women coaches for example. And so on the on the trainer level on coaches we are for example, aiming for gender, for 5050 gender, gender balance, and in the chest streaming world do you have actually I don't know if it's 5050 but definitely a lot more women amongst the in the entertainment around chess and the teaching around chess. Then in the in the world elite. And that's that there can be many reasons for this, just like in many other sports, but are many other disciplines, not not just sports. But Peter, you you had also studied some papers on this topic, I think. Yeah,
mean, you're asking a very big question there. And there was an interesting conference in London last December and the leading research on the subject per se, it's actually a sort of umbrella of a lot of different reasons. I mean, a strong factor is just numbers. I mean, there is many, many more males and females who plays chess, and that also will make it more competitive between them. So for instance, let's say in Denmark, we have hundred female players only, and maybe 30 or 40 of them has actually reached the national team in one way or the other. So, I mean, it becomes rather easy to fulfill your goals and then the next step would be to do well on the international level, which suddenly becomes much more difficult. And we will see huge rates of dropout. And Sebastian mentioned, I think also there is some kind of effect, like you can maybe see and let's say get to areas that simply because the distribution is incredibly big like 95 to 5%, it means that a lot of females will be discouraged to go there. If they go there, they will be treated differently and so on. And it seems to have some effects. There is of course also the debate if there is obvious physical differences that will explain it and there could be some according to the science, I'm not a brain scientist, I explained it
to you fund the
former Women's World Champion in chess, this question as well and she she said there's still at the at the top level that there is also a physiological difference in how how fit how fitness effects The concentration when you're in a, let's say, like a world championship match that is like 12 days, six and six, six hours every day of hyper concentrated activity that burns thousands and thousands of calories. So, but I'm, I'm not the ultimate expert on this in
that case, it's not so often decided by Perfect. Okay, sorry. Yeah.
Yeah. So it sounds to me that you want a female leader, if your country is undergoing the COVID-19 pandemic, but you want a male Grandmaster, if you've got a 16 day tournament, is that how it's going to come down?
Yeah, and this also puzzles me because it's clear that in the other, let's say, intellectual fields, I mean, we would feel very insulted if we would separate the male and female ability, while in chess that's considered somewhat normal. So I mean, I like that you're suggesting that the female leaders is actually doing better. Why don't just is the worst. But I think also Sebastian has a point that if people were in countries where females actually can have on their careers, they tend to do so because they consider it more prestigious than chess.
Well, what we've seen in mathematics is that when we took steps to increase the diversity of faculties and in advanced mathematics, they started producing extraordinary accomplishments that just weren't being done by their male peers. Now, it's just because you had new brains going into it, but it seemed like there was seemed like they could just simply they just became great mathematicians, right as amended. And it seemed that seemed to be an area that was really for men and not for women. Of course that just changed. So you wonder what can what can be done particularly with AI and I particularly like the idea of it happening online. We know that during the the work from home ethos, because of the pandemic and zoom. A lot of introverts don't perform well in a public Setting are doing really great work in their companies through zoom, because it's something somehow it gives them that confidence and protection. And so you wonder if you're not going to have online with AI, more women, if you encourage them coming into chess and advancing into the higher ranks of it through AI and online chess where they might get discouraged if they went to their local chess club meeting.
Yeah, I think actually, years ago that was done some scientific experiments where they let you know, well, females play online against someone who didn't know that they were female, or vice versa, the female wouldn't know if she played a male or female. And then you could see their female improvement actually getting better sort of the idea was to prove that there is certain structures that was keeping them down, but there was also other experiments of going in a different conclusion But no, this is female and chest is actually very popular among scientists because it's a very interesting field of offer study. But the ability to actually get a lot of females to play has not really succeeded yet, but it's it's a work in progress, I would say.
Now, we're coming up against time, we're supposed to end in four minutes. So we have just simply time for a final question. I'm going to see if there's what we have from our people online. And let me see, there's I there's two that I think are really interesting. One is about whether chess will become an esport. And then the next one, which everyone thinks is, Will AI, after winning chess go into all other areas of human strategy and, and mastery. So who wants to take the question of Esports maybe that could be used Sebastian, and who wants to take the question of broader strategic mastery in all domains of life, maybe you Peter hanneke, or you Talk about both, but I invite you both to close your close the session. final comments.
Okay. I would say yes to both.
In esport it depends also how you define eSports. But we are definitely have made it our mission to make chess something that many more people can follow. It's much more aligned with the societal values today than that, that we are trying to promote in many areas, also in the UN, and then the many other sports. That's Let's celebrate human achievement and in in the, with the mind and not just the bodies. And I think chest is the perfect sport to carry this and to carry all the brands that want to be associated with that and turn that into a beautiful viewing experience. A lot of sports have done it and I mean not to look down at any other sports but if if snooker can do it or not can do it, then chess can do it. And now, we don't call those other two eSports. But I think we can also say the same thing that if I myself, I enjoy Counter Strike, for example, a lot, but is it whatever you'd like my son to watch? Or would I prefer that he, he grows up where at least like chess, it's kind of having a similar market share within eSports. And I think that should be possible and that's what we're working on. And then of course, there should be a lot of scientific benefits to be derived from all of this and that leads me to Peters to Peter Hannah.
Yeah, you're sort of asked about the the scope of the possibilities and will deep mind that the company that make alpha zero, they have enlarged it, I mean, chess is a simple game in one sense that there is perfect information you make a move that is the opponent's turn, but their their next step was making something called alpha star, which is playing Starcraft two and It is actually become stronger than the best humans in the world. But you'll understand that Starcraft is a game where you move at the same time, sometimes you cannot see what the opponent is doing and such. And basically based on visual input and self learning is becoming better than the best humans. And also, it's not something that is able to push the buttons quicker, because that they compensate for it simply becomes better spreadsheets, ugly by just learning, you know, watching and making mistakes and learning and in games that I sort of played your computer game at the you know, well, at the same time. And so that sounds insane to me. So what this is one of the boundaries is very hard to grasp, and it's incredibly interesting area. And I think, yes, we're just happy that they kind of passed by us on the way and they gave us a lot of knowledge. But of course, Their aim is not to build good chess machines. The aim is to do something interesting for society. And that's very much work in happening right? Now,
right now, before I let you go, there is one final question for you because we've already brought it up. But one of the people on our chat felt like we didn't give it justice. And the question was, was Matthew O'Shea? Who asks, has anti cheating been discussed? Now, we raised the issue earlier at the very beginning about AI being used for anti cheating, but we didn't develop it anymore. So why don't we just take a minute to explain what's happening?
Sure. So we're basically using AI and machine learning to to profile players and how they how they behave to set a baseline and then to use that and enter cheating analysis or as a as a signal for further human review. And that is kind of what we are, that has been developed together with with with universities and we are going to take that to the next level. In and further and further anti cheating AI research but there's only so much you can use this for on on top level top player level because of course, somebody like Magnus Carlsen would often reflect as a cheater we have such a system and whenever one of the top 10 in the world makes a novelty it might be triggering the same signal. And really that's then VR simply cutting edge and we're AI yeah will will not be able to prevent or be the ruling thing there. you rather have to really simply assume that people have a reputation to lose that is much more valuable for them then then then the temptation to cheat. That's why for example, in our Magnus Carlsen chess tour, the even though it is for for $1 million prize money we we are absolutely convinced that none of the participants have an incentive to cheat. That being said, we have also entered mechanisms that are non AI based like cameras, surveying them and that evaluate what is going on in the room. And, and things like that.
With this, I think we've hit up against time unless I'm told otherwise. So I believe it's the moment for me to pass the baton back to Kenya to close the session. Unless the ICU lets me know that we have extra time to go forward. But I'm not certain we do. So I'm waiting to hear a signal either from them or from Sydney to take it over for me.
Yes, thank you very much candidates.
Thank you very much for your participation. I think it was a very engaging discussion. So the recording will be available soon on our website AI for good.itu.int and we invite you to follow us on social media at AI for Good and join us next week for the AI for Good innovation factory session to be held on Thursday. The 14th At 4pm Geneva time, Thank you and see you next week. And thanks again. Follow
Bye Bye everybody.
Yeah. Hey, thank you very much. That was great. I really appreciate it really nice.
Take time. Okay, take Okay. Okay.
Yeah, thank you really good. Like
All right. Am I the only Ken did you leave? Yeah, I guess
I ran this question so much do we still have 60 people okay.
No remaining participants this is Bastiaan from IQ and we will now close the