StarTalk Radio - What Are the Odds?

Episode Date: December 4, 2020

What are the odds your crazy play at the end of the game works? Neil deGrasse Tyson and co-hosts Gary O’Reilly and Chuck Nice investigate the use of AI and machine learning to predict and understand sports with mathematician and author Matt Ginsberg. NOTE: StarTalk+ Patrons can watch or listen to this entire episode commercial-free. Photo Credit: Storyblocks. Subscribe to SiriusXM Podcasts+ on Apple Podcasts to listen to new episodes ad-free and a whole week early.

Welcome to StarTalk, your place in the universe where science and pop culture collide. StarTalk begins right now. This is StarTalk Sports Edition. I'm your host, Neil deGrasse Tyson, your personal astrophysicist. And today's sports edition we've titled, What are the Odds? And if you're going to start a program with that question, somewhere in there, there's going to have to be a mathematician. So we'll get to that in just a moment.

But first, let me introduce my co-host, Chuck Nice. Chuck, you baby. Hey, Neil. What's happening? All right. Now, you pay your rent telling jokes,, so why do we have you on this? Actually, if I were good enough

to pay my rent telling jokes, I wouldn't be here now. Oh! Alright. So, Chuck, we love you because you bring some levity where it's sometimes needed, but also you're a sports enthusiast on a level that I think is uncommon among people who are not otherwise paid to have that level of knowledge.

So good to have you. But where we get our street cred, of course, is from Gary O'Reilly. Gary. Hey, Neil. What's up? All right. You're a stateside, but you were raised in the UK, played professional soccer there, and love talking to you about your injuries and how you recovered from them. We have whole shows on that.

There's a lot. But you come from the professional world, and there's some insight there that you could bring to this conversation. Because what we have found is that superstition can manifest itself in sports. And superstition tends to show up when people are not entirely in control of the outcome. And so how much in control are they not? Is this even quantifiable? Should managers study this? Should players study this?

And so generally when those kinds of issues arise, you need a mathematician. And so we found one. We found the perfect mathematician for this. And who do we have here? Dr. Matthew Ginsberg. Matt, I think it is, if I can call you. Welcome to StarTalk. Thank you. And yes, it's Matt. It is Matt. Excellent. Excellent. So you have a PhD in mathematics. I do. That's dangerous, you see, because people don't know how important math is to what they do. And then a mathematician starts walking up and comes in the back door and you say, uh-oh, they're ready to sort of shake this up and to tell us where we're misthinking.

They're going to straighten us out. They're going to ruin our superstitions they're going to and so do you ruin people's days with your ex what i found is that is that when you have two people who are having an argument and you come in and you just completely settle it with mathematics you just say they always both love you i mean it's just it's totally the way to make friends. Okay. Yeah. Okay. Actually, I think you're wrong there, Matt. Yeah, I think he's lying.

Yeah. I think what happens is they both kind of go, that guy's a dick. They're both mad at you because they both look like idiots, as everybody does next to mathematics. Well, they really, they just want to have the argument. They actually don't really care about being right or wrong. They just care that they're disagreeing. No, no, the difference is, here's how it manifests. If you settle it with a solution that neither of them had previously embraced, then they're both wrong.

And then, then, then that, that magnifies your dickitude. So we should actually, we should get my wife in here. Cause she will tell you that she will be delighted at anything that you guys can do to,

to minimize my dickitude. Okay. What's surprising about this conversation is it took so long to get here. We're already... No, but guess what? We're already making plans to bring him back. Based on that single episode.

And who knew the word dickitude actually existed? So, let me ask. And by the way, we have a third segment where we just sort of chew the fat and shoot the shit. Can you stay for that segment? Because there are places I want to get into your

professional profile that would not be part of what we know we must get through. Because we want to talk about what kind of statistics do you... Where did you get your phd where were i got my phd in astrophysics oh it's mathematical astrophysics from roger penrose at oxford oh okay let me tell my co-authors he just totally name dropped there yeah

who hardly anyone heard of before a month ago when he got the nobel prize right that was like total total name drop okay that's a double name drop because you got the oxford on the back of it you got the oxford you got penrose who's famous among all astrophysicists and the and the the throwdown of the of the no But so anyhow, so then you started, decided to slum it and do pure math. Before I went to Oxford, I was at Caltech for a year, where my supervisor was Kip Thorne, who also went on to win the Nobel Prize.

And I actually think that I get credit. This is me. I'm the common link between these two scientists. So the reason they both won the Nobel Prize is because they got to see the advice from someone. You know, it's interesting because Tom Cruise Who probably will never win An Academy Award for acting

Has been in movies where You know five other Academy Awards Have been given to the movie To the other actors So he could say that he enabled The acting of There you go This is probably a dickitude That's true He could say that he enabled the acting of, yeah. There you go. There it is. Exactly.

Yeah. This is probably a dickitude. That's true. Yeah, I was going to say, I'm kind of siding with your wife here. So what is the knowledge you bring to assist sports people? What have you studied in your math? So the math really was twister theory and quantum mechanics and stuff like that. But since then, I spent a lot of time doing AI.

And I made, I think, most of my career doing artificial intelligence and statistics and machine learning. And that's what I bring, is a knowledge of statistics and what you can predict and what you can't and how that kind of information can help people make better decisions as they play their games. When we talk about predictive analytics, are we really just talking about a different forms of statistics and the use of statistics? The, the, we're often talking about machine learning and machine learning is really just sort of statistics on steroids.

Right. Yeah. Wait, are steroids allowed in computing? I'm sorry, I'm going to have to invalidate this program. It's on steroids. It's a power surge. I participated a long time ago in the World Computer Bridge Championship. I wrote a computer program that played bridge.

And I was at the World Computer Bridge Championship, and they actually drug tested the programs. Took a sample of blood. I have no idea what this means as far as steroids and so forth, are there are people worrying about this issue so wait so let me ask if you said there's some things that you can't predict is because there's insufficient data to make a lot a reliable prediction or would you are you did you mean to imply that there are some things that can never be predicted no matter what so you can't at some level you can't actually predict anything with certainty, right? So we all think the sun's going to come up tomorrow, but there are a variety of things that could cause it not to.

No, no, don't go there. No. You'd say that like, yeah, there's a 50% chance the sun is not going to rise. If I'm a betting person, we're good. So let's not worry about the philosophical limits of what could or could not happen. Let's just talk about what a certain betting person would bet on. How about that? So what I'm trying to say is all you can ever do about the outside world is quote odds.

Now, if you can make a bet, if somebody is willing to bet me that the sun is not going to come tomorrow, I'm going to take that bet. I'll make that bet because the sun is not going to come tomorrow i'm going to take that even odd i'll make that bet because the sun's coming up tomorrow you know 99.999 and a lot of nines similarly if i want to predict that you know babe ruth pointing to a point in the stands i'm going to hit a home run and it's going to go there it's not a hundred percent there's some chance that he gets to honor that prediction. And the question is, how accurately can you gauge the probabilities and how close can you get them to one or zero? Correct. And so a zero is certainty it won't happen. And a one is certainty it will happen.

Right. Okay. And so, of course, sports, there's a lot of betting in sports. And if you could put a shading of insight into an outcome, you could become a very wealthy man. Yeah, but I wouldn't have any fun. Okay. Okay. Yeah.

Uh, I, let me just, uh, put a quick addendum to that, Matt. Um, you would with all the money.

So yes, you're right. You would not have any fun in the process. However, the results. The result, you have fun afterwards. Afterwards. Afterwards. All right, so what about curses where people have long losing streaks and they think they're cursed? And this is where superstition comes in.

And it seems to me mathematicians are highly needed in those moments. So all we can tell you that you know you're probably not cursed okay so there's a chance you're actually cursed that's what you're saying there's a chance of anything right now it's imagine i give you a coin right and you start tossing it and comes up heads heads heads heads heads at some point you will conclude that the coin is biased the coin's not fair and how long it takes you will conclude that the coin is biased. The coin's not fair. And how long it takes you to admit that the coin is unfair is really a function of how confident you were initially that the coin was fair. The more confident you are the coin is fair, the longer

it takes you to believe you're actually dealing with an unfair coin. Curses are the same, right? So a person, I think, should, but we we have no there's no known mechanism by which you can be cursed so my belief that i am cursed is it's incredibly unlikely but eventually the evidence would could overwhelm it now you know the red socks not winning the world series for a thousand years they didn't win the world series yeah yeah but it's just because they stunk right it's not because they were cursed yes well they sold the bandino and then they sold a lot of other players to the new york yankees and surprising enough the yankees went ahead and won a load of world series and the red sox had been winning before yeah exactly they're quite dominant in fact so i think that you know so a curse is such an extraordinarily unlikely event.

But what you asked me at the beginning is the probability is never zero or one. The probability that I'm cursed is, you know, microscopic. But eventually, I might eventually be convinced that I was cursed. I don't think, you know, for a baseball team to be cursed, I believe you would need more than a lifetime's worth of evidence. The Red Sox didn't even come close. Right. Do you not get a herd mentality where all of a sudden that so many people buy into it that they cannot remove that thought from their mind and therefore it kind of self-prophesizes?

I don't think so. So I've actually, I've looked at this. Wait, wait, just wait, Gary. You're talking about a case where that's how voodoo might work, right? Where you're so convinced you're going to fail that it throws off your game and then you fail and then you credit the voodoo, blame the voodoo for that. Because you have to find a fool guy. Therefore, it's the curse or the superstition. Performing at very high level athletics, you know, that 90% of the game is half mental.

That's what you're saying there. Absolutely. A lot of it is. That's what I tried to look at. So I played tennis in high school. I was awful. So you took up astrophysics. I'm glad.

Okay. Yeah, I know. Part of the reason I became convinced that I was terrible and that made me worse. But I think when you're talking about professional athletes, if they do that to themselves, they're not professional athletes. They just can't, they're competing against people who don't. If they're susceptible. One of the things I did when I was working with the Oregon Ducks volleyball team

is I had a bunch of statistics and just out of curiosity. You live in Oregon. You're in Eugene, Oregon. I do. I live in Eugene, Oregon. I've worked with the Ducks a bunch. I was working with the volleyball team. And they gave me statistics on every play. And I said, okay, let's look at individual players who play poorly for the first half of a match.

Is there any correlation between they're playing poorly the first half of a match and they're playing poorly the second half of a match? In other words, have they gotten in their own heads and decided they're having a bad night? And the answer was no, there was no such correlation. The kids are as near to machines as the coach can make them. They play badly on one point, they shake it off, they play well the next. And let me just add, and just to, to, just add, and just to blow some smoke in your direction, you have to know that that was the right question to ask. That's a brilliant question and a simple question that I bet most people don't think of.

Because what you're saying is if you're starting to have a bad game, and then does it affect you mentally? Because you already know what they are physically. You know what they're capable of. And you ask the right question. And not enough people in the world think about the value of the right question asked at the right time. So thank you.

I think that one of the things that I have learned, so a lot of what I've done has involved large amounts of data. And what you do is you want to take the data that you get from whatever source, and you're looking for patterns and the patterns you're looking for surprises because when you find a pattern it's a surprise might be a curse might be lack of curse but you're looking you learn to look for patterns and data that's what machine learning is about that's what statistics is about it's what data science is about

and when i got all this volleyball data i just started doing everything I could think of. And I asked a bunch of questions that were less interesting. But, you know, this one, thank you for saying it was a good question, but it's something that the data was able to advise me on. And, you know, so I've done, I did the same thing with basketball. basketball. So I got every basketball play for a year. And I started looking to see, is there anything that predicts whether a player is going to perform well or poorly in any particular game? And the answer turned out to be yes. And it was basically the number of minutes he had played previously. And basketball has a very long memory. It turns out that playing a lot early in the season, you carry that burden with you for the whole rest of the season.

But does it help you or hurt you if you played a lot? It hurts. It hurts you, yeah. It's season fatigue. It is. As a matter of fact, last year, before the whole COVID, so not this past season that they played, but the season before, there was a big argument about certain players.

Wait, wait, so Chuck, if you want to go before, it's just BC before COVID. That's all. Oh, yeah, BC. There you go. But there was a big controversy about allowing players to kind of take a break during the season, take a break during the season, sit out games that weren't as crucial to allow their bodies to refresh and re-energize in some way. And it just became a huge, huge thing. And the idea is –

Damn it, I'm paying $80 for a ticket. I want to see Shaq play. That's it. And that's basically where the controversy is. And the other controversy was – Because I'm paying his salary. Right. The other controversy is, hey, we never did it like this before. You know, the guys before who came before us are tougher.

Those guys grounded out for the whole season. What makes you so soft that you got to take a break? Do you want me to go through my injury list? Because this is another hour now the guys in my era did exactly that and we have injuries that last for days if we start with one end and go to the other we didn't have guys like matt who would sit down and now you find an elite sport we call asset management i gotta pull you out for two games so as i can have you for the next two months because if I play you in those next two games you're out for those two months because you're burnt your muscles are

in a good condition and there's a higher probability that you're one of the things that I found and it was interesting so it's not just you know sit Steph Curry it's at what point so you know there's three minutes left in the game. You're up six. Is that enough that you should sit him down and give him six more minutes of rest today so that he plays better tomorrow or the next day? And the interesting thing is, if you have all the numbers, this becomes a purely quantitative question.

You know, if I sit him down for the rest of this game, my chances of winning this game go down a tiny bit. I've probably won it because we're up six with three minutes up. But my chances of winning the day after tomorrow go up, and the game after that, they go up. So you end up with this incredibly

quantitative question, and I actually crunched it all. Okay, so let's come back to that. We've got to take a break now, but when we come back, we're going to talk about number crunching in exactly as you're describing, Matt, and I'm going to ask you the question, if it's all just a matter of numbers, who needs coaches when StarTalk Sports Edition continues? We're back. StarTalk Sports Edition.

We're doing this by the numbers. What are the odds? What does that even mean? We got a mathematician, Matt. Matt, love having you here. Just because I think everyone should have their own personal mathematician in our speech. You just reach for them when you need one. And that's just, I love, love having the guys in

the bar who are having a fight. They don't want their own person. They don't want their own mathematician. I want our own mathematician, Matt Ginsburg. So we, we left off with you describing the decision to sit a high value player if you're leading, because you don't need them to win the game. And that could help you win tomorrow need them to win the game and that could help you win tomorrow's game or the next game and see these are decisions not according to you not left to the judgment of the coach but these are with fast computing and ai thinking about it you can make these decisions in an instant all right so presumably that's not just for basketball, for other sports as well. So, yes, you can. I mean, you can help all over. You help in different ways. You know, so I've I worked with the Ducks football team on whether they should punt or go for it on fourth down.

I worked with the volleyball team on when they should call timeouts. I also did some, you know, so I worked with the volleyball team. This was not so successful. So I worked with the volleyball team. This was not so successful. We set up a system where there were two high-definition cameras, and they would watch the volleyball, and the girls on the team had smartwatches on their arms under a shooting sleeve. And if the serve was going to be out, we would make the smartwatches buzz,

and we could tell before the serve got there. That was not successful. The NCAA found out about it. They told us we were bad people and made us stop. But is it illegal? They actually said it was illegal. Well, they didn't know it should be illegal until they saw that you were doing it, right? Isn't it?

Well, they said it was illegal, and we said, why? It doesn't break any of the rules. And they said, we don't actually know which rule it breaks, but we are telling you that it does break the rules and you have to stop. So what they should have said is not that it breaks rules, but that it's cheating. And so did it calculate based on a ballistic trajectory and not an aerodynamic trajectory? So you would get it wrong if the ball has a spin on it. One of the things I did, obviously obviously is look to see whether the ballistic trajectory got the right obviously of course well you have to you've got to test what you're doing

you can't just make stuff up and then expect so and the answer was i was able to accurately predict whether a volleyball serve was going to be in or out so that was that was just it worked but but cheating you they can't just say you're cheating right yeah i was gonna say how does that change how would that change the game is it no you decide whether you're going to return or not serves can be deadly and if they're out you'd rather not have to try to return it this is very simple yeah but you would have oh i see what you're saying so it happened in that in that such short period of time that if you're

wearing the watch and it buzzes yes do nothing that's brilliant i would be wow i would be buzzing their watches pretty much as the ball crossed over the net okay because that that that's what i was trying to figure out is how much in real time and how what so your reaction time is part of this too well i i didn't do anything right the computer was it was all a computer he programs ai he doesn't he doesn't have to do a damn thing he's not even here right now this is avatar you know there you go i got all right so so what is the future of coaches if you can make key decisions such as that by AI? What do you need a coach for?

So I think this is not just about coaches, right? So we have AI entities making more and more decisions for us. And I think the thing that you have to realize is we solve problems totally differently than machines do. So, you know, when a person plays chess, they look at a handful of chess positions. When Deep Blue plays chess, they look at, it looks at billions. When we solve problems by pattern recognition,

they solve problems by searching. The net result is that we are good at one kind of thing and computers are good at a different kind of thing. So the question about balancing somebody's energy, how many minutes should they sit out this game and next game? Machines are good at that, we're lousy. Machines should do that. Questions about motivating the players to practice harder,

to not get drunk before games. Whatever it is the coaches are doing that is very much a personal interaction, adding value, adding experience, adding, you know, pattern recognition kinds of things. We're going to keep doing that. That's what, no, no, no, no, no, no, no, no, no, no, no, no, no, because what happens is I was out all night and I was drinking and I come in and AI tests my blood and it says your alcohol is 2% blood. I got that joke from a movie some years ago.

I know a couple of those guys. And so then AI says, my data shows that your performance will be diminished by this amount. And so they will know exactly how much your performance would be diminished. But that's not the problem. I don't need a coach for that when you can know the chemistry of that phenomenon.

You don't need a coach to tell you that your drunk player shouldn't be playing. You need a coach to convince the player not to get drunk the night before a game. But didn't Joe Namath get drunk the night before the Super Bowl and still win the Super Bowl? I mean, did he? How do you know and still win the Super Bowl? I mean – Did he? How do you know that? What, did you – I was with him.

Yeah, you were with him. I was wearing the pantyhose. So let me ask. For those who are not 100 years old, he did a TV commercial advertising pantyhose. After I think he shaved his legs, did he? I don't know. But anyhow.

Anyway, let's move on so so so man um baseball is famous for how much statistics plays in the decisions of coaches on the level will this shift an entire infield of players to one side because of the statistics of how the batter hits the ball do other sports lend themselves to this level of analysis? Yes. Yeah. Short answer. Basketball is right on baseball's heels. So I know a bunch of people involved in the basketball world, and they are incredibly committed to exploiting statistical information the same way the baseball guys do. The other sports are behind. So I've been looking at football and talking to football people and they know that they're not taking advantage of data as much as some of these

other guys. And as you began just a moment ago, the biggest decision play is, do you kick or do you go for fourth down? I mean, fourth and short, do you kick or do you go for it? And that's, you know, everyone, that's everyone debates that the day after, all right, if it didn't go the way people wanted. How does data help that? You just do it. You just figure it out. And I did work for the Oregon Ducks, and I actually made a deal with them

that I would figure out what they should do on fourth down, and I would get four season tickets. I worked. Good deal. And I told them every week I would go to them with this giant book of what to do in every possible fourth down situation. And I would give it to the coach. And I said the same thing. I said, look, here's the deal. Just stop punting between the 35 yard lines. You'll be fine. And the Oregon Ducks stopped punting between the 35-yard lines. And they had an amazing year, in part because not punting between the 35-yard lines in college football is just right. I did then go to him the next year and ask him if he wanted to do it again. And he said no.

And I said, what? And he said, I know not to punt between the 35-yard lines. I'm going to keep my tickets. And he sent me on my way. five yard lines. I'm going to keep my tickets. And he's, and he sent me on my way. So they are starting to make the decision this way, but not terribly effectively yet. I mean, you can do so much more than fourth and however, fourth and whatever. So what you're saying is your, your, your, your advice to him was consistent with his life experience and therefore it had no value to him. No, no, no. He, had no value to him no no no he he

when i started working with him he never punted he always punted between third and five drive lines everybody punted so the first time he followed my advice he was playing utah was fourth and five from midfield and he decided to go for it because i had been telling him to do this forever utah was so surprised they had to waste a timeout because they had put their punt return team on the field and he put his normal offense on. So they had to blow a timeout and then he got it. And he thought, well, that's pretty cool.

And he did it again. And he learned. He just started doing better. And if you watch college football now, you will see that between the 35-yard lines, they go for it a lot. And this is starting to trickle into the nfl also and and and what is the fourth between the 35 yard lines is there a qualifier because fourth and 15 on a penalty between the 35 yard lines that's a different

it's a lot different than fourth and two between the 35 where's the threshold answer is there is essentially no qualifier i don't i didn't run it out to fourth and 47 but but go for it and 15 i think 15 may have been around the break point but you really just go for it and the reason is because some percentage of plays get you 15 yards right sometimes you make it but mostly punting does not help you because college punters aren't that good they're going to punt it in the zone so instead of having it at midfield they have it at their own 20 and it's not really that much better it's not that much and also too um college returns are a lot advance a lot further because in the pros, if you look at the pros, when they punt, those dudes are right there.

And most times that player is going down. Occasionally you'll see a breakaway. You might see somebody get 20 or 30 yards on a return, which is far more routine in college than it is in the pros. That's why they're called special teams, Chuck. The other thing I've told him is you should always go for it on fourth or one on your own nine go for it on fourth and one and the numbers just say it's better because every now and then you're going to get it and then you get a new life punting from your own nine you're screwed you

were screwed on your nine you're still screwed because the other team is going to have the ball on your own you know on your 40 and you're still screwed because the other team is going to have the ball on your own, you know, on your 40 and you're still there. And the coach actually told me, he said, stop telling me about this. I can't go for it on fourth and one from my own nine because if I'm wrong, they're going to fire me. I don't care if it's right. Stop talking to me about it. I'm just not interested. So I stopped talking about it.

But so coaches are acting. Coaches have a lot. So I stopped talking about it. But so, so coaches are acting, coaches have a lot more to balance than simple quantitative decisions, right? They have to balance the press and the fans and people's moods and stuff that I don't know anything about because I'm just a numbers guy. So you telling me the facts don't care, right? So when the coach, yeah, the, the, the coach is just riddled with emotions and he's going to get fired if dot dot dot what happens when you slide the dossier across the desk and say this is factual how much pushback and and do is it never going to happen or do you get like the embrace the coach eventually gives you or is it always it varies varies enormously, right? It's a, it's a personal thing. I mean,

so I've worked with people who have a huge non-invented here syndrome. You can tell them any idea and they will tell you, yeah, I already thought of that. It doesn't matter if they have or not. Yeah. Yeah. Okay. Dealing with football coaches, a football coach may say, if I go for it on my own nine and fourth and one and I'm wrong, I get fired and he's right. Yeah.

Because it's his job to know what gets him fired. Then you can't, you know, he said, stop trying to talk to me about this. Yeah. But in baseball, if you make a statistical decision, you're not going to get fired because everyone knows the power of those statistics. And then in the long run, you're going to win. And you have 162 freaking games because it it's a marathon not a sprint now maybe football each game has is a little more critical let me how about what about the hail mary pass is that is that a statistical thing i don't know i i actually um just looked at this and i can tell

you that it looks like you it works about two percent of the time but it's riskless right because yeah you are you have the ball at midfield. There's six seconds left in the game. You're down by four. If you don't throw a Hail Mary pass, you are guaranteed to lose. If you do throw a Hail Mary pass, you have a 2% chance of winning. I'll take 2% over 0% any day. So it's not the kind of decision.

What people get upset with is you go for it on fourth and one. And statistically, it's right, but it was wrong spectacularly today. the kind of decision what people get upset with is you go for it on fourth and one and statistically it's right but it was wrong spectacularly today okay okay so the the the alternative to a Hail Mary pass is a quarterback run with 12 consecutive laterals that's that's the alternative to that and have you thought much about that that's that a lot of ball handling, which always carries a risk. No, so you have to understand I can't think about that because I don't think about football. I think about numbers. And there is no data on people trying this multiple lateral play.

You could say what fraction of laterals are fumbled, right? Because it's another ball handle, right? And, and so there's gotta be some access into that. The problem is that every lateral has to be backwards. So eventually you may, you may run out of field, but the real problem is you're not moving the ball forward as you're, as you're laterally. Occasionally you do say somebody, you know, there's a kickoff return at the end of the game and people start lateraling around it never works i've never seen him work i've seen hail mary's work yeah right right and then there's also the trick play like which is a variation of a lateral where and it normally only happens on kickoffs where the returner will receive the ball.

And then because you can forward pass as long as the ball is not going forward. So he throws the ball across the field to another player, which looks like a forward pass. But because it's going in and back in back it's going backwards it's considered a lateral and then that player is free from obstruction he's wide open and he's got a big long field of him and if he's fast enough he can actually avoid being tackled so yeah now we gotta take a quick break when we come back uh matt I want to get into sort of what made you who you are if I can spend a couple of minutes doing that and still

connected to the theme what mad scientist puts you together Matt in what basement were you invented alright we'll take a quick break this is Star Talk Sports Edition by the numbers when you return We'll take a quick break. This is StarTalk Sports Edition by The Numbers. When we return. We're back.

Star Talk. Sports edition. What are the odds? And we've got mathematician Matthew Ginsberg. So, Matt, you briefly mentioned that, and you just said off the cuff, oh, Hail Mary passes work 2% of the time. And, you know know i don't i don't have to run numbers to know that it doesn't happen often but when you say two percent you probably mean exactly two percent so how did you how do you actually get that number so for that

number which um your team actually asked me about i looked at all the football data I have. So I have every NFL play since 1999. And the question is, which are the Hail Marys? So I said, okay, you got to be down more than a field goal, but a touchdown can tie it. So you got to be down four to eight points. You have to be at least 30 yards away from the end zone. So you have to have at least 30 yards to go.

It has to be the end of the game, which I defined as either less than 20 seconds left to play or less than a minute left to play and it's fourth down. So you're desperate. And then I looked at all the plays that were like that and how many of them were actually recorded as being a deep pass, which is part of the data I have. There were 187 plays in desperate situations like that, that were recorded as deep passes. And of those 187 plays, four were touchdowns. So I can then tell you.

That's your 2%. 2.1% of Hail Mary's work. Wow. Wow. Yeah. But as you say, 2.1% of all this data, I just did a bunch of queries. Yeah. It's over over zero percent right i mean if you don't do it you you know you're not

that's okay so in this third segment which we've we're still trying to find a good name for it but until we do we're just calling it shoot the shit so so what um i got a question for you, Matt. If I remember correctly, was it Deep Blue? Yes. Or whichever IBM computer beat the world's best Go champion. That was a Google computer called. Google. Sorry, Google. That's right.

Google. What was the name of that computer? I think it was AlphaGo is I think what they called it. OK, AlphaGo. And I remember chatting with some Google people, and apparently you can do one of two things if you have machine learning, deep machine learning AI. You can say, all right, let's study every single game ever played,

which you can do with chess because every important game is recorded, right, and have it learn moves that worked and didn't work, and it could have the sum of all human knowledge and thereby beat practically any human who plays it. That makes sense. But then in the follow-on Google program, it didn't do that. It said, what is the goal of the game? Oh, it's to demark territories.

Is that the goal of the game? Oh, it's to demark territories. Is that the goal? Well, let me play myself and do what I think I need to do to win. And let me play a billion games against myself. Doing that, that program beat the previous program that had beaten all previous Go players. And then it exploded. It exploded. Yeah. program that had beaten all previous go players and then it exploded yeah and then a tear in the space-time continuum opened up so my question is man at what point are you going to say my best data are all of the games ever played rather than my best data are the games that could be played that i can think up and have never been

played and thereby i'm going to make a giant leap for the sport so like the first person to bunt that must have been a crazy like what is that oh something had to invent that right uh it was fosbury it was a fosbury yeah fosbury floposbury flop. Yeah. 1968 Olympics, Mexico City. He does something brand new. Just to be clear, because I've noticed that people know that that's a better way to jump over the bar, but they don't always know why. But give me like 30 seconds to explain that. So if you watch old high jumpers, they would go forward over the bar, and you can only really bend at your waist. Your legs can't bend in the opposite direction than they do.

So your legs have to clear the bar straight. Whereas if you go backwards over the bar, your back can curl, you can curl backwards at your waist, and your legs can dangle. So you can make a hemicircle, a semicircle, and curl yourself around the bar, taking your center of mass below the bar, even though your body goes above the bar. So you can jump higher than ever before by actually jumping lower than ever before in a world record setting jump. Because when you're curling, your center of mass is no longer in your body. It's outside of your body, and you didn't actually have to jump as high. So you could say that was

cheating, but they left it in. And now everybody jumps that way. So I just have to slip that in, because I've noticed that not everyone knows that, even some athletes. So continue, Matt. Sorry, I interrupted. So, what I do, what people working in machine learning generally do,

is that, and I said this before, you look for patterns in data. The more data you have, the more likely it is that the patterns you detect are real, the more likely, the easier it becomes to find the patterns. So you want as much data as you possibly can have. And there obviously is a choice. You either can use naturally occurring data, data from chess games, or you can use synthetic data that you generated by playing yourself, for example. And there's a trade-off.

The naturally occurring data is at some level going to be better. It's hard to imagine computer simulations of football games being of much value, for example. But you do have the opportunity to generate more data by playing yourself, by using other computers. At this point, computers are way better at chess than we are. So the data you want, you want to look at the data from Stockfish, which is the best non-machine learning-based program.

From my perspective, you just want to do what works. I don't care where the data comes from, as long as what I learn from it is actually valid. Yeah, but it's more fun if your program invents a new football play that no one thought of. Yeah. Because it didn't want to base itself on games that were played before. Wow.

I want to see innovation because that's, you know, and I guess that's what trick plays are. Yes. And is that true creativity? And is that what we're doing as human beings? Looking around and seeing, looking for misdirection, looking for ways to make your opponent think something that's not happening. That's how we come up with new plays. And that is also a part of pattern recognition.

As a matter of fact, in football, they call it disguising a defense and all that kind of stuff. Yeah. And not only that, the very final play in the movie Major League was basically a trick play. And that's baseball. Right. What happens is the batter points to the outfield, okay? And so that's, you know, he's showing some gonads there, right? Points like he's going to hit.

And then instead he turns around and bunts. And so the third baseman has to come running in because they were playing in deep. There was a whole construct. Let me give you a – let me actually try and answer this okay no we'd rather just argue at you i know i'm gonna actually ask you the question back when i worked for the ducks i actually did invent a new play and i described the play as an optional self-safety so there are situations in football where you actually want to score a safety on yourself and then take the kickoff from the 20 and so forth and so on. And I created another kind of play, which was you go back for a pass,

throw a pass. If your pass is basically guaranteed because the guy is just wide open, throw the pass. Otherwise, step out the back of the end zone and do a safety on yourself. New play. Nobody had ever thought of before. So possibility one is that the Ducks win a crucial game because I gave them this new play. Possibility two is the Ducks win a crucial game because I got them to stop punting between the 35-yard lines. For me, Possibility 2 is just as much fun. That's the question you originally asked.

It's just as much fun as Possibility 1 because I did it. I know I did it. I don't have to go and pick up a girl in a bar by telling her I invented the passing self-safety. By the way, that wouldn't work. Okay, good. the passing self-safety. By the way, that wouldn't work. Okay, good. It would work in the nerd bar.

The nerd bar works every time. If you've said that, you've tried it. No, no, it works in the nerd bar. Otherwise, go on. I know. When I watch a football game, when I watch a college football game and I see somebody punt, not between the 35-yard lines,

I smile from ear to ear because I did that. It's just as good as if I invented this play that people run from time to time it's just as much fun what i did was just as impactful and whether it was something that that sort of looks really splashy like pointing to the outfield and then bunting or whether it's something that's just grinding it out by doing the numbers and figuring out what to do on fourth down, impact is impact. Winning is winning. All right, so then why don't batters who are statistically predicted to hit 80% of their line drives to the left side of the infield,

why don't they just punch one to the right side or save that for when they really need it? And was it an ego that they're fighting yeah i'll get through them even though they're putting their best stuff on me you have to ask a baseball player i have played a bit and your natural swing is going to go a particular way and messing with it is at your own peril i got you right okay all right so can i ask you let's look at football and basketball because they are not as statistically driven as baseball is there one thing in each sport barring the fourth and one situation that you know you hear announcers talking about is there one thing in each sport that you would say, this is what they're all not doing, that if they did, they would find greater success?

So basketball, basketball really is very statistical these days. And I don't understand basketball nearly as well as I understand the other games. So I'm going to leave basketball out of it. Football, yes. So there is one incredibly easy thing that people don't do in football and they're crazy. And that is you score a touchdown and you have to decide whether to kick the extra point or go for two. It turns out that your expected number of points scored is pretty much the same. You're about 50-50 to make it if you go for two. So all you're doing

when you decide to go for two instead of kicking the extra point is you're adding 50 50 to make it if you go for two so all you're doing when you decide to go for two instead of picking an extra point is you're adding sort of noise you're adding uncertainty and that means and this is incredibly simple and they should just i have to i have to butt in because you said something that i don't know is if it's completely clear to others so when he says your expected results are the same what he's saying is if there's a 50% chance you're going to score two points, then half the time you score it, the other half you don't. On average, you scored one extra point, which is the same as the guaranteed one extra point that was the traditional add-on. So I think I said that correctly, right, Matt? Okay.

Yes. So you can either have a very straightforward one point. You know, it's going to be one. It's not going to be zero. It's not going to be two. It's just going to be one. Or you can have like a noisy one point where it's half the time it's zero and half the time it's two. So the question you should be asking is, do I want noise in the football score? And the answer is, if you're playing a team that's better than you, you do. Because the way you're going to win this game is by getting a little bit lucky. And you want to give the gods of luck as much chance as possible to help you. You want to increase what's called the standard deviation in the score. You can't move the average, but you can make it noisier. And you want to climb out to the

wings of that distribution where you could have actually a chance of winning that exactly as the underdog yes so every underdog in football that pours a touchdown should go for two and it's obvious and it's simple and i have no idea why they don't do it they don't think about the problem this way apparently it's not obvious to people don't have a phd in mathematics with a degree that they earned from a Nobel Prize winning brilliant. We got to actually land this plane. But to take us out, Matt, just tell me, what was Factor Man? What is that book?

You wrote a book in 2018. I did. What is that? I have spent a lot of my professional career working on a particular technical problem. And if you solve it, you can solve all the other problems, literally. And it's a book about a guy who does. And it's a novel. Oh, cool. Cool. It's a thriller. It's a thriller, a mathematical thriller, two words you've never seen in the same sentence. I've often talked to my friends about, you know, if you solve this problem,

it is actually a race between whether you take over the world but the government kills you and and and this guy who solves this problem is in this race he knows he's in this race he doesn't really want i'm watching that when it comes out on netflix because they'll surely buy the story anyway that's what factor man's about oh excellent excellent i'll'll look for that. And greatest name ever for a computer program, Dr. Phil with a F-I-L-L. Tell me about that. Dr. Phil is, I don't remember who suggested the name. I actually, it's a program that solves crossword puzzles. And Neil, you and I have both created crossword puzzles.

I co-created the crossword puzzles for the New York Times. I am awful at crossword puzzles. I co-created the crossword puzzles for the New York Times. I am awful at crossword puzzles. When I have a puzzle in the New York Times, it takes about a year from the time you submit it till the time it comes out. I can't do my own puzzles when they're finally published. So I wrote this program that's really good at solving crossword puzzles. It's sort of my revenge on all the people who are so much better than I am. I needed a name and I asked the community of crossword puzzle constructors for a name. And one guy suggested deep clue.

Nice. Good one. That was good. And somebody else suggested Dr. Phil, which I thought was a tiny bit better. So that's how Dr. Very cool.

So, so in other words, never wrong a geek because whatever they're not good at, they'll create a robot. That is. Take over. Yeah, man'll create a robot that is. Yeah. Take over the world. Yeah. Matt, we've got to call it quits there.

But thank you for being on StarTalk. Sports industry, delight to hear how you've contributed to this. And this is surely not the end of that story. There is much more infusion that math will have, not only in sports, but the rest of our lives. I think, overall, the more the better. We look forward to this. Thanks for being on StarTalk. Chuck, Gary,

always a pleasure. We're here. I'm Neil deGrasse Tyson, your personal astrophysicist, as always, bidding you to keep looking up.