Hit enter to search or ESC to close

Hockey and self-driving technology intersect when Pittsburgh Penguins data analysts Kat Wu and Nick Citrone sit down with Argo AI’s VP of Autonomy, Peter Carr, along with host Alex Roy. Learn how technology and predictive analytics help NHL teams gain a competitive advantage.

Listen On

Apple
Google
Spotify
iHeart Radio

Share

View Episode Transcript

Episode Transcript

Alex Roy (00:00:06):

Everyone welcome to No Parking, the podcast that cuts through the hype around self-driving and artificial intelligence. I’m Alex Roy. I’m not a big fan of the TV show Foundation. I love the expanse obviously, but this Foundation show, I’m not sure quite captured the magic of the original books by Isaac Asimov, because the idea that there might actually be science to gathering data and predicting outcomes is very, very cool. So if you make a TV show about that, it should be awesome.

 

Alex Roy (00:00:33):

Anyway, you see the idea of prediction and data aggregation all over sports in movies like Moneyball. And of course you see it in the development of autonomous vehicles, but how well does this work? How can data be used to predict sports or driving or the behavior of pedestrians and cyclists?

 

Alex Roy (00:00:51):

So I decided to find a little more about it. And I joined Peter Carr, the vice president of software engineering at Argo AI, and we got two people from the Pittsburgh Penguins, Nick Citron, a senior data scientist, and Katerina Wu, a data scientist. Their job is literally to do this very thing, and I’ll let them explain how data helps the Pittsburgh Penguins win. Let’s dive right in.

 

Alex Roy (00:01:22):

So Kat and Nick, well, let’s start with Kat. How did you get into sports AI?

 

Katerina Wu (00:01:29):

So I’ve been into hockey for a while, but then I never really looked at the mathematical side of it really until college when my team was on the brink of playoffs and a friend of mine made a bet that with me that they wouldn’t make it into the playoffs. And we were learning about Montecarlo simulations and my statistics class. And I made this really badly made spreadsheet that calculated the probability that my favorite hockey team would make the playoffs, and it turns out that they would.

 

Katerina Wu (00:02:08):

So I thought that this spreadsheet was pretty good and I thought, “This is pretty interesting,” and I was wondering if other people were doing anything like this, and that’s kind of how I delved into sports analytics in general. And from there on, I attended conferences and got to work at S&T Sports media technology, and it kind of evolved from there. And that’s kind of how I got to sports AI.

 

Alex Roy (00:02:37):

I know [Car 00:02:37] has got a crazy story. So Nick, what’s yours first because Car’s story is just going to go off the charts.

 

Nick Citron (00:02:43):

Yeah. Thanks, Alex. So I’ve always been a big sports fan, especially hockey. I’d say football and auto racing have been kind of my favorite sports to watch. And I was always interested in the stats that went with that, like who scored the most goals, who’s got the most passing yards, that kind of thing. But I thought that was just kind of a passing interest and a hobby more than anything. And then when I was in college, I was actually a business major originally. And my sophomore year, it was coming time to pick a track, be it finance or accounting or whatever, and there wasn’t one that really grabbed me and I had to take a regression analysis class. It’s just a core business requirement, and I fell in love with it right away. I was like, “Oh, the stats here, this is what I want be doing. This is what really interests me.”

 

Nick Citron (00:03:22):

And once I kind of discovered that, I realized I wanted to add stats as a major. Deciding I wanted to get into sports took about a week. You know what I mean? It took me a long time to realize I wanted do statistics. It didn’t take me that long to realize I wanted to tie into sports, because I’ve always been such a big sports fan.

 

Alex Roy (00:03:35):

Car, you were already getting your PhD, correct? And then you started playing hockey. Tell us that the backstory there.

 

Peter Car (00:03:45):

Let’s see. Yeah. So I went to the Australian National University to do my PhD, which being Canadian was like, do you really need to go to the other side of the world to understand things? And I had been doing computer vision, and specifically projection geometry, so it’s basically looking at images and trying to figure out from what you see in the image, can you figure out the 3D geometry of the world?

 

Peter Car (00:04:07):

One of the world experts just happens to be at that university. And I figured, hey, this will be fun. Life without winter. You’ve got to experience that at least once or twice. So that’s how I ended up in Australia.

 

Peter Car (00:04:18):

And I guess in some ways I got into the hockey circles in two different ways. The first of which is that Australians take sport incredibly seriously and the Australian National University is in Canberra, which is the capital. They also have the Australian Institute of Sport, which they built specifically because they… I don’t think they want any medals in the Montreal Olympics, and they said, “That’s unacceptable,” and they built a state-of-the-art facility. So when I’m here doing computer vision for my PhD work and trying to find funding, it’s like, well, I’m sure we can do something with the Institute of Sport. They probably have camera-based things.

 

Peter Car (00:04:53):

So we toured it and they did, and they had their swimming facility and they showed how they track the arm movements of the swimmers to look at the rates that they’re swimming at and saying like, “How do you see through the splashes and everything else if you want to get the arm movements and the stroke rates?”

 

Peter Car (00:05:08):

And then they said, “Oh, we have a hockey program.” I was like, “Oh, I would love to work on that.” It ended up being the field hockey program, which I discovered that in Australia, you have to say ice hockey. And so we spent a lot of time tracking their Olympic hockey players and the strategies that they use to try and prove out concepts the coaches had.

 

Peter Car (00:05:27):

And then it wasn’t until a Canada Day party at the Canadian embassy when the deputy high commissioner at the time, she was like, “Oh, there’s a new Canadian in the town.” She just sort of said, “The rink is that way. We play on Friday afternoons. I’ll see you there.” And so that’s how I got into the ice hockey, I guess, group that was in Canberra as well at the time, when I was doing my PhD.

 

Alex Roy (00:05:50):

I really admire all of you for choosing to go into a field by the most complex path, because it was… When I was at NYU, I was reading a lot of science fiction and I was really curious about system theory in general. And I read one line of one article and, and it said, “Well, if you don’t believe the future can’t be predicted, ask yourself why the water pressure in every major American city drops during half time.” And that was enough for me to know that, well, on one hand, I wanted to understand why. On the other hand, I would allow others to do the math.

 

Alex Roy (00:06:23):

So let’s talk about that math. So can you walk us through, Nick and Kat, walk us through how you use data and analytics to predict outcomes or improve the performance of the Pittsburgh Penguins.

 

Nick Citron (00:06:40):

Thanks, Alex. So that’s obviously a very broad question and there’s so many different answers, right? Because we look a lot of different types of data. So we look at individual player statistics, so things that individual players are doing. We look at team level data. We look at leaguewide trends and comparing leagues to each other. So there’s all different kinds of data.

 

Nick Citron (00:07:01):

Honestly, I think, and a big part of the philosophy that I have is just to try and answer questions, to try and help answer questions. It’s interesting because like I said, I’ve been a big hockey fan. I know a good amount about hockey, but the amount that I know about hockey pales in comparison to the amount that our coaches know, that our general manager know, that our players know.

Nick Citron (00:07:21):

So I think if we’re being used to the fullest, it’s almost if we can just help answer questions that they have, if there’s insights that they’re curious about or things they want to look into and we can help, they can communicate that to us and we can look in the data, get answers to that and give that back to them, I think that’s when we’re at our best.

 

Alex Roy (00:07:37):

Kat, explain what is the most valuable data on the players. What do you need?

 

Katerina Wu (00:07:42):

We use multiple different sources of data. So we have the raw NHL data that the NHL tracks, but then we also use third party data. But most recently we’ve also had the addition of the player tracking data from S& T that gives us more accurate in-depth data that we previously didn’t have. And we don’t really use the data as-is. We create our own data frames and then try to manipulate them to create data that we could actually use later on that match specifically the definitions that our coaches or GMs are looking for.

 

Alex Roy (00:08:29):

That’s a very fun, corporate answer. Peter, you know where I’m trying to go with this. Okay. Explain types of data that are of value to a coach, to anyone.

 

Katerina Wu (00:08:40):

I would say there’s certain metrics that the coaches are interested in, but in a more general sense, maybe a coach might be specifically interested in how often a player goes to the net, which maybe the average viewer wouldn’t want to keep track of just on their own.

 

Alex Roy (00:08:59):

Peter, what kind of data were you looking at back in the day when you were trying to model games in any field?

 

Peter Car (00:09:08):

We’ve had a huge technological revolution in being able to track players. And so when I was working on my PhD, there were not companies that had this available. So we would put cameras up. We would track all the players. And I think maybe coming back to in this case for field hockey, the Australian team had a tendency to play a really aggressive game and had to translate it. It’s like, oh, you’re forechecking.

 

Peter Car (00:09:33):

That’s really what they wanted to play. They played a high tempo, constant turnover of their players and rotating through. I’m like, this sounds a lot like ice hockey. If we just run faster and we have higher endurance, we put the opposing team, especially their defenders, under a tremendous amount of stress. It used to be sort of like you’d send the ball down to the other end of the field.

Peter Car (00:09:52):

All of a sudden now, the Australians would be running right up there and getting in your face and challenging you for the possession, whereas most teams would sit back and let you bring the ball back up the field.

 

Peter Car (00:10:01):

And so their core question was, we can train our players to have really high endurance and we’ll win every one-off game. And the problem was that when you get to a tournament, you’re always imposing a higher workload on your opponent, but then you beat them. It’s a knockout and you’re off to the next game. And it really was that our team had the stamina to play that level in tempo for an entire tournament versus… And I think the other thing too is your players have to run. Are you inducing a higher and even higher workload on the other team?

 

Peter Car (00:10:30):

So we spent a lot of time getting trajectories. And then from that trajectory data, looking at sort of distance covered, accelerations. And I think a lot of these sort of intangibles of… I think, how do you talk about like owning time and space? I think so many team sports come down to that.

 

Alex Roy (00:10:48):

So Nick, I mean, do you get questions from the coach? Like, “Gather up this data on X so I can make a decision, A versus B.” What kind of data is the most valuable to the coach?

 

Nick Citron (00:11:03):

Something similar to that. So one thing that we could get a request from a coach could be like, “Can you give me data on every time that our player is doing this thing? What’s the results from it?” So let’s say for example, if you were talking about football, it’d be every time that our receiver gets the ball on this side of the field, does he get the first down or not? The kind of rates or percentages of things happening given a situation, so you can say, if we put a player in situation X, how often is he succeeding? How often is he failing, and how does that compare to other players on our team? How does that compare to the league average?

 

Nick Citron (00:11:37):

Are there other confounding variables or things that are happening in the game situation that impact how successful you are? So based on if you’re facing pressure for something that’s happening, like if there’s someone else bearing down on you, or if you have more time and space, how does that change things?

 

Alex Roy (00:11:53):

So, Peter, when you said that there’s a bit of a technological revolution around this, from the point at which it became possible to gather data and make decisions from it, can a team have a competitive advantage because their data gathering and analytics is superior over a team that doesn’t take it seriously?

 

Peter Car (00:12:11):

I’m not sure I’m the best one to answer that. Let me tell you at least my perspective, I would love to hear what Nick and Kat have to say on it. So, I mean, at the time, I was at Disney at this point. We were doing a whole bunch of research work for ESPN. And a lot of the research questions we were kind of going after were similar things, what coaches would be asking about too. Would players and fans in particular, what are the storylines that come out of the data and what do computers do that’s different than sort of what an individual human can do?

 

Peter Car (00:12:43):

And that really was holding a season’s worth of tracking data in memory and to say, “Oh, this play right now looks exactly like this one from this game last season.” To some degree, we’ve got the color commentators and we’ve got the replay staff to go find that during breaks in play. But a computer can do that constantly all the time. We looked at a thing called ghosting and the Toronto Raptors sort of had used that. Now, did that lead to their success of winning their first NBA championship a few years later? I don’t know. I think it, and maybe this comes back to really a interesting thing with sports in particular, is that in the analytic view, we tend to be very probabilistic. We say, “Okay, well, nine times out of 10, this is probably what’s going to happen.”

 

Peter Car (00:13:22):

I’m curious. Does analytics make a team consistently better in that you can pretty much say, we will be higher in the rankings. Will we win the championship? I don’t know, because it comes down to that single instance, but you can say “I can play this way and more likely if I play 10 games, we’re going to win,” but in one game or a best of seven series, is there enough consistency of the situation to show that your strategy is going to play out in the long run?

 

Alex Roy (00:13:49):

Have any of you watched the television show Foundation or read the books by Isaac Asimov? I find that fascinating that three people whose lives are devoted to the study of data and prediction have not read these books because… And I have to set this up for you, and I’d love to hear what Kat has to say.

 

Alex Roy (00:14:08):

In the books written by Isaac Asimov over 60 years ago, the galactic empire in is in decline. The emperor is not happy about this and he summons a mathematician, Harry Seldon, and Seldon says, “Well, I could predict that the go galactic empire will collapse. I can’t tell you how to stop that. I can’t prevent that, but mathematically, I can tell you that it will, and that I can predict the actions of large groups of people, but not individuals.” And the emperor doesn’t like this and insists there must be a way. He demands that mathematically an answer is provided and Seldon cannot provide it.

Alex Roy (00:14:47):

And so here’s my question. Maybe Kat, you can answer this. You have large, massive amounts of data. Surely there must be a way to turn it into some workable thing that you can actually… It becomes actionable. How do you do that? What is that?

 

Katerina Wu (00:15:12):

I would say the biggest, large pieces of data we have right now are with the tracking data. Each game is around seven to eight gigabytes alone. So that’s just a massive amount of data coming in every game. I would say the biggest issue with that is just trying to tie all this data together into a workable form, because with the tracking data, we know where the player locations are, but we don’t necessarily know what they’re doing at that exact moment in time. So we might try to tie in a third party data that says right now, a player is taking a shot at this location. We might try to find a way to tie in the time from the player taking the shot to the word…

 

Katerina Wu (00:16:03):

… from the player taking the shot to where every single person is on the ice at that specific time. But it’s really difficult because all these pieces of information are being measured at different levels of accuracy, where the tracking data is being measure at hundreds of a second, multiple times a second. Whereas with the NHLs scoreboard is a whole integer and with third parties it might just be multiple decimal points. So we want to make sure that we’re actually picking up useful information with this and not just…

 

Alex Roy (00:16:49):

I’ve heard Peter talk about this in the context of autonomous vehicles in terms of fidelity. Am I correct?

 

Peter Car (00:16:57):

There’s that. If I can jump in with one other question first though. And I’m really curious that obviously there’s a whole bunch of data analytics happening with sports now. For hockey in particular, how important is domain knowledge about the game? Because certainly for people that I talk to who haven’t seen hockey before, I think there’s, how do you see any structure? This is pure chaos. And I think you look at basketball, there’s so many back and forth in basketball, you got 50 opportunities a game [inaudible 00:17:27]. Baseball is very discreet, fate- based. Football is every time there’s another down, the players reset.

 

Peter Car (00:17:34):

Hockey’s pretty much a face off and then let it go. And that could be two to three minutes of evolution of this continuous game. And so I think the ability to find structure I suspect is way more challenging.

 

Alex Roy (00:17:47):

Nick?

 

Nick Citron (00:17:47):

I think it is important for sure. I would say that one of the things I think is just true in data analytics in general, and especially in sports analytics is, a lot of the times you’re limited by your creativity with a data set. Because you can look at one data set of player fees and stats across all leagues and you can try and do your best to find out which players that are good this year are going to be good in two years, or how players age or any of those things. Or you can look at it from a totally different angle and look at which league are better than other leagues by how players changing leagues perform, right?

 

Nick Citron (00:18:18):

So much of what you can get out of a data set, questions you can answer, is how creatively you can slice it up and look at it different ways. And I think that having more domain knowledge of hockey makes it easier to think of ways to do that. I think that depending on the data you have, you could have very little knowledge of hockey and if a coach or someone comes to you and says, “Hey, I want you to measure this very specific thing,” you could probably figure it out, see what you need to do and be able to do it, right? But I think if you have a better knowledge of hockey, you probably are able to find more things on your own, try more things on your own.

 

Nick Citron (00:18:50):

And also I think another thing that’s easy is if you have better hockey knowledge, I think you can easier see if you have problems in your code, right? Because if your data or your report is telling you something that if you know about hockey doesn’t make any sense to you, like let’s say for example, you’re seeing that you’re more likely to score in the penalty kill in the power play in the long run. That’s probably something you have a mistake in your code and you have things switched up because that’s not likely, right?

 

Nick Citron (00:19:13):

So I think domain knowledge is definitely helpful. I don’t know. Maybe it’s not critical. I’ve worked with people before on a hockey project with someone who didn’t have much knowledge and was able to pick it up pretty fast and still contribute a lot. But I definitely think it helps.

 

Alex Roy (00:19:27):

Hold on a minute. In the movie Hunt for Red October, there is a scene where they’re trying to debate whether or not captain Ramius, Sean Connery, for those who haven’t seen the film, or can’t remember who Ramius was, and I think the thing is, oh, he always goes starboard at the bottom half of the hour. Now, are you trying to tell me that you don’t have some better than average, pretty strong metric for whether a goal leak can deflect a puck on Thursdays? There’s nothing?

 

Nick Citron (00:20:03):

I think when it comes to goalies, I think one of the things that’s a prevailing thought in hockey is that goalies, they’re really difficult to predict in general or difficult to figure out.

Alex Roy (00:20:13):

This is the voodoo mythology of goalies?

 

Nick Citron (00:20:16):

Yes. There is a phrase that is no disrespect goalies, but that goalies are voodoo, which I think is just saying that generally in most sports, in most positions, if you’re a star player one year, you’re going to perform really well the next year. A star forward in the NHL, if you’re putting up a point per game, unless you’re going to get hurt or unless you get traded to a bad team or whatever, you’re pretty likely going to score at a similar rate next season. It’s pretty likely that you point totals are going to be in the same ballpark.

 

Nick Citron (00:20:42):

Whereas for goal tending year to year at any level, the NHL, the AHL, in Europe, it’s very common to have a goalie be one of the best in the league next year and then one of the worst in the league the year after that. They’re so up and down. I’m not an expert on the goal tending. I can’t tell you exactly why that is. But in general, predicting future goal tending performance is very, very difficult. Even last year, Carey Price of the Montreal Canadians, he, I think had a… He was very much an average goalie in the regular season and a phenomenal goalie in the playoffs. And I can’t tell you why that was or how that’ll go in the future.

 

Alex Roy (00:21:16):

Cat, can you tell us why? Because Peter’s making a face. He’s making a face.

 

Katerina Wu (00:21:24):

Not with any more accuracy than Nick can.

 

Alex Roy (00:21:28):

Because everyone could tell you in car racing, I’m a car guy, and in car racing, you know that if it’s raining, some drivers are going to do better than others. You can also say, well, if they put on rain tires and they’re really good in the rain, well you can extrapolate from there and then chips go on the table.

 

Peter Car (00:21:57):

I think, Alex, I don’t know. I’d love to hear from Nick and Cat on this. I think part of it is in making a prediction, I can tell you what I’m going to predict and maybe you that’s good enough. I trust you. You’ve been right before. I’m going to bet again, right? But you might also ask the question, why? Why should I trust your prediction? And I’m sure from the coaching staff and everything else, as you said, the coach can have a hypothesis. You can go away. You can explore that, find the data to support it.

 

Peter Car (00:22:20):

But it the alternative. And that’s what we were doing at Disney with the work for ESPN was, look at the data and try to figure out trends and commentary storylines to it and just say, “Hey, here’s what we’ve come up with.” So we wouldn’t necessarily be predicting what’s going to be happening, the outcome of the game. That would be something that we were doing, but we were much more focused on why are you predicting that? And I think that’s a very important distinction in terms of how that information’s going to be used.

 

Alex Roy (00:22:47):

Hockey, the puck moves much like a ball does on a pool table just with more variables. So how does it work when you have so many shots that are expected against the goal? So can you walk me back through how that prediction is made? Whether the puck will pass the goalie? If a goalie is fixed… Do you see where I’m going with this?The expected goals against the goalie is obviously one of several calculations. Can you walk me back how that’s calculated?

 

Nick Citron (00:23:25):

Yeah. I can walk you through that. So expected goals is one of, I think, the more simple concepts in sports analytics. And the simplest way to think about it is just, if you imagine just every time that you take a shot on goal in the game, you probably have a good sense of if that shot is more likely to score than some other shots, right? So in hockey, if you think about a shot from the blue line, so that’s right when you enter the offensive zone where it’s further away, if the goalie’s not screened, which means no one’s in the way of you and the goalie blocking his view, you’re very unlikely to score in that goal. In the NHL, it’s definitely less than 2%. Versus if you’re right in front of the crease, right in the goalie’s face and you’re shooting from right in front of him, he’s going to stop that shot less often. You have a better chance of scoring on him, right?

 

Nick Citron (00:24:13):

So you can imagine right there that the shot from far away from the blue line is going to have a lower rate of scoring than the shot up close. And in the simplest version of expected goals model, if we just take a new account the distance to the goalie and the angle of the shot to the goalie, you can imagine just watching a couple games even just on yourself, just marking down where the distance and the angle are, did you score or not? And then you could just fit a simple regression model, which will take into account the distancing angle, the interaction and give you a percent chance to score.

 

Alex Roy (00:24:44):

I have to interrupt you right now, Peter. The more I hear from these analytics experts from the Penguins, the less faith I have in gambling on sports and the more faith I have in riding an autonomous vehicle, which seems to where the accuracy all seems to fit more predictable and rational in space than a puck and a hockey player. Am I correct?

 

Peter Car (00:25:11):

Yeah. I think that comes back to this question of structure, right? As we have the rules of the road, much like you have rules of hockey. You have general strategies of hockey that you can recognize. I think there’s general strategies of driving. To Nicks point about the likelihood of scoring from a particular shooting location. I think the next question then is, great, how do we create those opportunities? Because obviously every team looking at that was like, “All right, well then protect that area. Don’t let anybody get the puck. Get shots from high quality scoring chance locations.”

 

Peter Car (00:25:46):

And so I think the next question really is that competition over time and space. And I think to some degree, how do you start to quantify that intangible aspect of a player who can create opportunities for the team? Who can create time and space for a scoring opportunity? Or I think from the autonomous driving perspective, it’s similar in that we talk a lot about what constitutes good defensive driving, where you’re in a place there to be able to react to what’s going on in the world. I think they’re similar in terms of understanding that time and space, negotiation that happens.

 

Alex Roy (00:26:19):

The absence of a puck on our city streets seems to be the primary difference and a very happy one because there are high penalties for touching other players. Now that I’m thinking about this, what a relief that that’s the case. Because effectively, if you were to take a hockey game and move it out to city streets, it would be a demolition Derby. No wonder prediction is easier for autonomous vehicles. Is it harder to predict hockey than other sports? Football? Put cars aside. Football, baseball? Is hockey the hardest?

 

Nick Citron (00:26:52):

If you’re talking about like winning a championship or winning a game?

 

Alex Roy (00:26:55):

Well, let’s start with a game.

 

Nick Citron (00:26:56):

I think it’s more random than other games. I don’t know. It depends how you define hard, right?

 

Alex Roy (00:27:06):

You guys are all law school graduates basically. Discuss the semantics of analytics or analytics itself?

 

Nick Citron (00:27:13):

I can, I can say that I think in terms of how often better team wins the game, I think hockey is more random, so therefore harder to win as a better team compared to a sport like basketball and football, and I’m not sure about soccer or baseball. But I think basketball is something it’s interesting to compare it to. Because one of the reasons that hockey is more random and it’s harder for the better team to win is that it’s a low event game, right? And that in any game, there’s almost always less than 10 goals or sometimes only a couple goals. Whereas in basketball, I don’t know, each team has at least 50 possessions in a game, right? So you have a lot more instances of me dribbling down the court against you and my players being better and more possessions, more opportunities for my players to make that difference in skills shown, right? Whereas in a hockey game, it’s much more back and forth. You have fewer sustained offensive possessions. You have only a couple power plays, so there’s less events, less opportunity for better players to really make an impact and score a goal and win the game. If that makes sense.

 

Alex Roy (00:28:19):

I’ve never heard the term low event sport before, but I feel like I’m a low event sport guy, as in my personal life. Cat?

 

Katerina Wu (00:28:28):

I would say for hockey at least, it’s more difficult to isolate what each player contributes towards a game. Whereas with games like baseball, where every single event’s independent. The pitcher does one job and they pitch the ball and that has a certain outcome that goes with the game. Whereas with hockey, a pass to another player might not actually result in any specific change in that game outcome, but it might result in that certain shift might have a different outcome. And it’s really hard to isolate just saying that one player affected the entire shift or entire game.

 

Alex Roy (00:29:11):

Peter, your head has been shaking like you have comments on high versus low event sports. I find that fascinating.

 

Peter Car (00:29:22):

Well, I would say I think it comes down to the complexity of the sport, right? And if you look at the history of where we’ve seen it’s work analytics making impact. I think it started with baseball, again, baseball doesn’t involve a lot of X, Y continuous trajectory data, right? You can say there is a batter at the plate. There is a pitcher. You can get the information about those two individuals. You can look at the state of the game in terms of how many outs are there? How many runners on are? So it’s reduced down to a set of certain integers and natural numbers to talk through those things. And you see lots and lots of examples of them.

 

Peter Car (00:30:05):

Then I think there’s others, like player tracking and basketball got into it. And I think hockey, it comes back a little bit to what you talk about, that fidelity, realizing there’s so much going on in the game and the subtlety. And one is, can you even measure that subtlety to begin with of the posture of the players? Do they have their stick outstretched trying to challenge for the top possession and so forth?

 

Peter Car (00:30:25):

And then I think then when you do talk about the context, was somebody in front of the net? Is it a screenshot to begin with? And how tall are they? How big are they? What is their body posture? Was the pile of bodies in front of you on the shots coming through? You need to get all that information in, but they need to make sense of the sense of it as well.

 

Peter Car (00:30:41):

And so at Disney, we were doing deep networks and working through this stuff. And I’m curious, if you go down that path, does the analysis become so detached from the actual game that it’s not actually useful anymore? We were really looking at the models used for AlphaGo and so forth and saying, “Let’s go apply that to player tracking data.” I can tell you right now, all the stuff that would come out of our papers that we wrote, I don’t think any coach would look at it and be like, “Ah-ha, there’s the key to winning next week.” These graphs, I’m curious how do you go from the fidelity of the data, getting into really doing this analysis to then making it actionable and putting in, this is how we’re going to win.

 

Alex Roy (00:31:21):

I’d like to know.

 

Nick Citron (00:31:23):

I think one of the best things you can do there, or what we have to do is… Because you’re right. It’s very difficult. For example, even if a coach says something as simple as, “I want to measure every time we do this one thing,” that’s a lot easier said than done, right? To automatically have code that identifies, okay, this is what the thing is happening. This is the result. It’s a success or a failure, et cetera. And that’s why I think a lot of the numbers we give, what the value comes in is if we’re able to apply it to every game, every season, every team, and you can get a relative ranking, right? Because there’s some things that our coaches will watch and they’ll record our own number on things and we’ll automatically record them and they might trust their own numbers a little bit more, maybe it’s only a 95% match, but…

 

Nick Citron (00:32:03):

… Trust their own numbers a little bit more. Maybe it’s only 95% match, but if they trust that the way we’re doing things is sound, and the numbers are “close enough,” if we have those numbers for every team in the league, and we can know through code that we’re applying the same standard to every team in the league, then our relative standing in that metric should be accurate, right? Are we good at it compared to other teams? Are we bad at it compared to other teams? And with all these things, if I tell you that we do something 60% of the time, that’s not very helpful to you if you don’t know what the rest of the league is doing it. Maybe that’s really good. Maybe the rest of the league is doing it 30% of the time, and that’s phenomenal. Maybe it’s 80% of the time and we’re behind. You know what I mean? So I think-

 

Alex Roy (00:32:39):

Wait, are you telling me there’s no… Kat, I want to hear more from you. There’s no Moneyball equivalent for hockey?

 

Katerina Wu (00:32:45):

I think we’re getting closer to that now with player tracking data.

 

Alex Roy (00:32:50):

All right. Let’s talk about this player tracking data. Is it true that the NHL players only have one RFID on them? And other sports have two or more. Why only one in NHL?

 

Katerina Wu (00:33:01):

I’m not sure why they only chose one. At least the difference for me between the NFL players with the two RFID chips is that you’re able to tell the player’s orientation, if they’re facing forwards or backwards. And with only one chip for the NHL, you can’t automatically detect that. You have to infer things like that.

 

Alex Roy (00:33:32):

Using camera with experts like car. Wait. So do you have a chip in the hockey stick?

 

Katerina Wu (00:33:38):

We don’t. And that actually makes it a lot harder to tell-

 

Alex Roy (00:33:44):

There’s a chip in the puck though, right?

 

Katerina Wu (00:33:46):

Yeah.

 

Alex Roy (00:33:47):

Does that chip survive an entire game?

 

Katerina Wu (00:33:52):

Not always. And there’s actually been a lot of issues regarding the puck quality with the chip inside.

 

Alex Roy (00:34:01):

Define puck quality.

 

Katerina Wu (00:34:03):

At one point, I believe it was 16 games into the past season, the players were kind of, not exactly complaining, but they were-

Alex Roy (00:34:17):

Complaining.

 

Katerina Wu (00:34:18):

They were pointing out that the puck with the chip was affecting their scoring rates. They found that it wasn’t actually anything to do with the tracking data. It was just the coating of the puck wasn’t up to up the speed, I guess. It was the finishing of it.

 

Alex Roy (00:34:41):

In a blind test, you’re telling me the players knew if the puck hadn’t chip in it or not?

 

Katerina Wu (00:34:48):

There are some players who, just from years of playing, that probably could tell from maybe the feel of the puck or something that they were able to differentiate between a puck that didn’t have the chip or not. But for this particular event, it was just that they were noticing that the finishing of the puck was different, and they weren’t able to score the same opportunities because the puck was bouncing differently or something.

 

Alex Roy (00:35:14):

Was there a scientific test, a double blind test, and they statistically could tell? There was?

 

Katerina Wu (00:35:29):

Maybe not statistically, but there are definitely players who are very sensitive.

 

Alex Roy (00:35:33):

Something smells fishy about this. It seems like the… oh, Peter, you want to add something here?

 

Peter Car (00:35:42):

I’m going off on a tangent. You tell me whether this is interesting or not. But I’m really curious. From the tracking data that you got, how much data engineering do you have to do, and I suppose [inaudible 00:35:53] have to do? As you said, if one of the tracking devices stops working, how do you distinguish that from a power play or a penalty kill, where all of a sudden like, “Oh, I only have nine skaters on the ice”? So I assume you have to hook up to the play by play data, but the clocks aren’t synchronized, right? And so I assume there’s a whole bunch of data engineering work that needs to go on before this actually becomes sort of useful.

 

Katerina Wu (00:36:15):

Yeah, I remember there’s this one particular point where one of the players, their acceleration suddenly went from like five feet per second to 123 feet per second. And I was like, “What is going on?” It turns out, I think it just stopped tracking for a second, and he just suddenly appeared on the other side of the ice. So there are definitely instances like that where we have to be careful to not be using inaccurate data in our analysis.

 

Alex Roy (00:36:49):

Peter, you wanted to ask a question? Because I’ve got a question.

 

Peter Car (00:36:52):

Yeah. I have a question. So, I guess with the economics of what it is, right? The league imposed the salary cap as a way of equalizing player talent and so forth. Is there a salary cap on analytics in terms of, the data’s available to all the teams, Alex, if you were owning a team, wants to win and I’m sure he would then be saying like, “Okay, well how much can I spend on my analytics department?” Right? If there’s no limit that, I want the most insights possible.

 

Nick Citron (00:37:18):

There is not a salary cap on analytics. It’s the same thing for coaching, right? Some of the best coaches in the league are paid probably several times what some of the lower paid coaches are. So same with coaching with front office staff, or size of scouting department is another thing where some of the teams that have more resources will have more scouts and be able to see more players and other teams. So I think when it comes to analytics, that whole kind of front office or coaching, that whole side of the front office is not regulated by a cap like the players are.

 

Alex Roy (00:37:47):

Well, if I were Mark Cuban, or any billionaire owner of a sports team, I would hire a bunch of people like Peter Carr to do computer vision, machine learning analysis, on all the rival teams. Then I would buy up all the data aggregators in the sector just give the opposing teams a mountain of work, and then hire as many analytics folks as I could to just create a bear of a problem for everyone else. And then I would demand predictions. Demand. And then bonuses would be given based on the accuracy of predictions. I think that incentivization structure might radically change this entire sport. Tell me my friends, is there any sport outside of hockey you feel has really nailed analytics in such a way they have delivered actionable insights to coaches and team owners or recruit?

 

Nick Citron (00:38:37):

I mean, first of all, I do think that we have delivered actionable insights to coaches and GM. I do think that our work has had a lot of value and there are a lot of things that we can measure and inform, even if overall there’s a lot of complexity and a lot of context and things like that. I think that baseball certainly lends itself, like Katarina said, to analytics the best.

 

Alex Roy (00:38:59):

Because it’s a low event sport.

Nick Citron (00:39:01):

Yeah. Because most plays have only the pitcher, the catcher and the batter have any say. And like Peter said, before every play, you can pretty much define this situation almost exactly. Nowadays teams are shifting more, but before that you even knew where basically everyone was standing. And so it’s a lot easier to measure, I think. There’s a lot more to be measured. Also, one of the things that I love about baseball is that we have pretty much accurate data back like a hundred years, which is really cool, at least box score data. Right? So I feel like stats have played a bigger role in baseball for a longer period of time, even if they weren’t attached to analytics. I think batting average and home run records and all that stuff has just been such a core part of baseball for so long that it maybe it was quicker to be adapted in a lot of ways.

 

Alex Roy (00:39:52):

Well, that’s interesting. So can anyone tell me, what is the history of data gathering across different sports? I mean, when did that begin in football, or hockey?

 

Nick Citron (00:40:04):

I don’t know when they started. I do know that originally second assists weren’t tracked in hockey, which is interesting. Originally it was only the goal scorer and the primary assist. So I don’t know exactly when, but at some point secondary assists were added, and some point totals definitely increased after that. Also, I know that scoring rates in hockey have changed a lot over time. Wayne Gretzky has just about every hockey record, including the goals record, and he’s currently being chased by Alex Ovechkin on the Capitals. But one of the issues that makes it harder for Ovechkin to catch Gretzky is there’s just less goals per game now, overall.

 

Nick Citron (00:40:36):

And a lot of that has to do with the better goaltending play. Goaltending style really developed in the eighties and nineties with the butterfly style over the old standup style. So goaltending in general has gotten a lot better, which means that there’s less goals in a game, right? Which means that the rate of all these things happening, the total goals scored in a game has dropped, which means players today, they have a harder time kind of chasing records of old players. So that’s kind of one way that hockey data has kind of shifted just with the way the game has changed.

 

Alex Roy (00:41:04):

Peter, I have a question for you. So you came from sports analytics and prediction and went into autonomous vehicles. So what similarities did you carry over towards prediction of outcomes in city streets for cars? And what differences existed?

 

Peter Car (00:41:28):

So I think the biggest similarity was probably thinking about the world in terms of time and space. I think on the driving side, it comes down to this negotiation of: Who goes first? Right? So when two cars come to an intersection with stop signs, the concept of simultaneous arrival, sort of everybody’s interpretation of what that is. And I think it changes a lot in the time of day too, that at rush hour, all of a sudden, half a second is probably different and I got there first and I’m going, versus a Sunday afternoon. So there’s a lot of understanding true probabilities, understanding the spacial relationships. I think what we talked about here too, the context in terms of: Was there a cyclist on the road at the same time? That’s somewhat equivalent to: Is there somebody in front of the net screening the shot? And it makes a really big difference in terms of how you think about the situation.

 

Peter Car (00:42:21):

I think a lot of that carried over. The biggest distinction has probably been the fact that we have roads and we have lines painted on the roads, and there’s a lot of structure, there’s traffic lights to regulate things and so forth. And I think that comes back to, hockey’s kind of a free for all, right? The puck drops, and then after that who knows what’s going to happen?

 

Alex Roy (00:42:38):

It’s a high-event sport.

 

Peter Car (00:42:41):

It’s a high event sport. We should talk about that some more too, but even, as you said, like compared to basketball, which generally sort of resets after every possession for the most part, to baseball that sort of constantly resets, football, right? Again, you’ve got the time to get your team back and formation start again. So I think that’s where I think hockey’s at the frontier of really dealing with that complexity.

 

Peter Car (00:43:05):

And I think Alex, to your other point about: When did this first start? I think in every sport, people were annotating things by hand. I know anecdotes I’ve heard, at least from the NHL, which I think is hilarious is that the home team was responsible for providing the statistics to the game. And so you’d look and you count the number of hits, and the home team always had more hits in the game than the away team. And so just, even understanding the bias of where your data’s coming from to begin with, I think we had some early stuff available and you could do a few things with understanding the game, but the equalizer, I think, has been computer based sort of analysis or observations of the game.

 

Alex Roy (00:43:46):

Kat, you have a look on your face like you have something to say about this.

 

Katerina Wu (00:43:52):

Yeah, there’s definitely, within the NHL, just using their NHL hit system from arena to arena, every arena has their own tracker, I guess, of someone who’s in charge of tracking where hits are, who was doing the hits. And there’s definitely been studies that show that certain arenas kind of pad their own players’ numbers a bit more than others. So maybe some of the older data isn’t entirely… They’re human gathered, rather than computer processed.

 

Alex Roy (00:44:39):

Peter, on city streets, cities gather data about different intersection safety, and some do it better than others. Am I correct?

 

Peter Car (00:44:51):

I think that’s true, but maybe where you’re going with this, there are different conventions in terms of, between state to state, between country and country, bike lanes for instance, and bike lanes in what they do within the different cities, there are different urban designs. And I think that’s, again, one of these really interesting ones where you look and realize there are… Or the cities know, right? There are intersections that just statistically have higher collision rates, and then you sort of start to go look and be like, “What is it about the urban planning and the way that those streets have been designed in intersections?” Certainly in the areas of Pittsburgh that are being rebuilt, you can see that they’re trying to improve traffic flow, or make it safer to get through intersections as well.

 

Alex Roy (00:45:32):

This is all making me feel like I made a huge mistake when I did not pursue more math study, because sports analytics sounds like the most fun subset of all math. But I’m a little insulted by this high versus low event hierarchy, because I’m thinking about when I was driving cross country years ago, and we don’t have to get into that, thinking about how few events occurred between New York and LA, and how easy it was, relative to people who have to drive around a track that is complex. And that, in a way, I guess I kind of cheated and I should admit that after all these years. What I did, and what I’m most famous for, was a fairly low event feat. And I’m sorry that this had to be the form in which I admit that. If a young sports fan who hates math and thinks they love playing games hears this episode and wants to pursue a career, what should they study? Is it just math? Or is there a specialization within math which would lead them to a seat like one of yours, Nick and Kat? Or Peter.

 

Nick Citron (00:46:44):

I think statistics. I mean, obviously I think statistics is a good fit for sports because it leads naturally to prediction. You’re trying to observe an event in a game and make a prediction on how likely it’s something to happen. I think it just lends itself well to what you’re trying to do in sports, which is watch a game be played, learn about how the game’s being played, try and make predictions on how you can play it better or decisions you can make to be better at it. There’s certainly other types of math which leads to different types of modeling and things like that, which are very useful also. So I wouldn’t say it’s just stats by any means.

 

Nick Citron (00:47:14):

I think if you’re interested in sports statistics, one of the struggles with it is that if you’re looking specifically to work for a team, there’s just not that many jobs, just because there’s not that many pro teams. Even if every one of the big four sports teams in the US doubled their analytics department tomorrow, that’s maybe a thousand jobs at most. You know what I mean? It’s just not that many compared to other industry. That being said, one of I think the nicest things about sports analytics is there is so much that you can learn about and so much you can do entirely on your own if you just have a basic computer and internet connection.

 

Alex Roy (00:47:47):

That’s funny because there is a shortage of really talented people in the autonomous vehicle sector, which is always hiring. And I would imagine that if someone was a fan of sports and could not find a role at a team, I think we have a non solicit with the Penguins, maybe not official-

 

Alex Roy (00:48:03):

[inaudible 00:48:00] roll it a team. I think we have a non-solicit with the Penguins, maybe not officially but unofficially because Argo AI happens to love the Penguins. It would seem like that would be a great place to look for talent, right, Peter?

 

Peter Car (00:48:14):

Yeah, I think mathematics. I was about to say my background’s in physics and to some degree it pulls in a lot of mathematics. What it does also do is really push and say, how are you bottling the world? And so there’s a lot of that intangible critical thinking, which was my question earlier about the domain expertise, to say, ‘I played hockey before. I know what it’s like on the ice.’ The direction you’re facing in is pretty important, but I can also see players who just put the stick behind with one hand and are still able to grab the puck without turning around.

 

Peter Car (00:48:44):

And so I think the training that you get from a physics background really helps because you’ve got the mathematics and a lot of critical thinking of just, how do you model the world? And I think the other one they get a lot is those thought experiments. The fact that what Einstein did, five papers in a few years, never having to leave his apartment and just said, ‘What if the following…?’ He came up with trains and all these experiments to work through and basically came up with relativity without ever actually having to do any experiments or observe that something was bizarre. He could just think his way through a problem.

 

Alex Roy (00:49:19):

Kat, if you had to come up with a wishlist, even one thing, a tool or data set, that would help you do your job, what would it be?

 

Katerina Wu (00:49:28):

I don’t think we’re legally allowed to measure this, but I would love to know how players are thinking and feeling, because that’s one part that we just would probably never have data for.

 

Alex Roy (00:49:44):

What do you mean by that, how they’re thinking, feeling?

 

Katerina Wu (00:49:50):

A player probably would never just come into a game saying, ‘Oh, I feel awful. I’m not going to play very well today.’ But to know that ahead of time would definitely affect their performance and would affect the actions that they make. And I think that would be valuable to analytics, but just not anything that we would ever have access to.

 

Alex Roy (00:50:11):

That’s interesting because the hours of sleep… Peter, we know that human drivers, their attentiveness varies greatly. Do we know that people are more or less alert at 9:00 AM than at 11:00 PM? That’s probably not something that you would have data on, but in a world of human drivers, I know that if people are driving at 2:00 AM, some greater percentage of them are probably drunk or tired, and we know this. And I know that when I was driving cross country at 3:00, 4:00 in the morning, I was more tired. My circadian rhythms are off. And I think we could probably also say if a hockey game took place at 4:45 in the morning, it would be a very different game than at 7:00 PM. Is there any way to infer player… If it’s an away game?

 

Nick Citron (00:51:04):

That is something that is considered. For example, one thing I can tell you where you can find it is, if you subscribe to The Athletic, they have a page there where they basically cover all the games for each day and they have a statistician who works for them and gives the odds of each team to win. And one of the variables he includes is rest, like how many days rest is this team playing on? Is this team playing on a back-to-back, which means they played yesterday? Do they have two days rest? Are they playing three games in four days? So that’s definitely a quick approximation of how rested are you, how tired are you, when did you last play a game, for sure.

 

Alex Roy (00:51:37):

Again, I’m feeling a lot more trusting of autonomous vehicles after this conversation than before. Peter, have I missed anything?

 

Peter Car (00:51:44):

Well, I’m really curious. Hockey in particular, you talk about the momentum. A big hit can be a shift in momentum. To some degree, is there a way to directly measure that? How do you account for that in the data when you’re looking at it, or is that just something that it’s great for the TV commentators to talk about, and it has no impact on the game itself?

Alex Roy (00:52:10):

It’s a high event sport, guys. You never know what event it might be.

 

Nick Citron (00:52:14):

Yeah, I can say that personally I haven’t ever looked or done any kind of work on specifically momentum. Yeah, it’s definitely something that is talked about a lot in hockey and it definitely seems, from watching it, to be something that plays a big role, but I haven’t attempted to capture that specifically yet, at least.

 

Katerina Wu (00:52:37):

I did have a question on the similarity between field hockey and ice hockey. Is there a one-to-one comparison where it’s just played on a different surface, or are there actual tactical differences between the games?

 

Peter Car (00:52:52):

So one of the ones that threw me off… I knew virtually nothing about this sport, man. I was like, ‘Oh, it’s just hockey played on grass. It could be a little bit slower.’ And then you realize that, the first thing they do is hand you a stick, ‘It’s one-sided.’ ‘What do you mean, it’s one sided?’ ‘There’s another side of the stick,’ and you realize, no, it’s got a hook, one side is flat and the other side is round.

 

Peter Car (00:53:12):

And so as a result, it’s very much like ice hockey where you can use your body to protect, in this case, the ball, and so you’re able to stick handle only on the right hand side. As a result, because the sticks are all one-sided, it’s much easier to attack up the right hand side of the field, because you’ve got your body in the way of the ball, and so you see this bias.

 

Peter Car (00:53:32):

Whereas I think for all that ice hockey analysis, you look at the player tracking data and you expect it to be all over the place. Whereas field hockey, everybody knows it’s easier advance the ball down the right hand side of the field, and so as a result, you expect that to happen.

 

Peter Car (00:53:47):

And now the question is, do you double down and shift all your players that way to begin with, to say, ‘We know where you’re going,’ cause statistically you’ve got a better chance of holding onto possession there, versus the teams who then say, ‘Okay, well, we can attack up the left hand side of the field. The ball’s going to be exposed,’ but you can move the ball faster. You can be in better passing positions and so forth, so that was the one that threw me off. I haven’t really seen too many sports where you don’t have symmetry between the left and right side of the field.

 

Alex Roy (00:54:18):

I’m going to ask one final question of Kat and Nick each. I ask you to answer this question. Who will win the Stanley Cup this year? And if you can’t answer that, who won’t win the Stanley Cup this year?

 

Nick Citron (00:54:31):

Well, I can definitely tell you that I’m a huge, huge fan of the New York Jets, and I don’t think they’re going to win the Super Bowl or the Stanley Cup this season, so that’ll be my answer there.

 

Alex Roy (00:54:43):

Kat, same question. Who will or won’t win the Stanley Cup this season?

 

Katerina Wu (00:54:49):

I have a very hard time saying it either way, because we’ve only been a few games into the season and already there’s been huge levels of disparity between the teams. It could go a complete opposite direction at any given point within the season. The St. Louis Blues, when they won the cup, they were last in the league up until January. So I truly cannot say.

 

Alex Roy (00:55:18):

May I guess on what Peter Carr’s answer would be?

 

Peter Car (00:55:24):

I would love to hear what you-

 

Alex Roy (00:55:27):

Peter Carr’s answer is, one cannot say who will win it, the Stanley Cup, only that someone will, and many will not. And many will not. One will and many will not, right? Is that the correct phrasing?

 

Peter Car (00:55:47):

I suppose you could say that. I think hockey for me, though, is always one of passion. And so, coming from Toronto, to some degree the only thing I care about is, it can’t be Montreal cause I just cannot take all the taunting. So, last year was a very nervous year as we got through the playoff [crosstalk 00:56:07]

 

Alex Roy (00:56:07):

I really predicted wrong.

 

Nick Citron (00:56:09):

I have one question for Peter. I don’t know how much he’ll be able to answer this, but I recently watched the Indy Autonomous Challenge at the Indianapolis Motor Speedway where they had self-driving cars racing laps around the track and I’m a big IndyCar fan, so it was pretty cool to see. I was surprised because, not knowing a lot about autonomous cars, I thought, okay, this is a track where they know the track ahead of time. The track hasn’t changed in a hundred years. So I expected most of the cars to be able to complete the track, complete the laps, and it would just be who’s fastest, but surprisingly several cars crashed and hit the walls.

 

Nick Citron (00:56:42):

So I guess, what are some things you think that the average person wouldn’t consider difficulties? Because not knowing a lot about autonomous cars, I would think if you’re giving me a set track where I know exactly which four turns I have to take and what they look like and what the banking is and everything, that would be almost the easiest that my job could get. So, I thought that was really interesting that it proved so challenging.

 

Peter Car (00:57:04):

I think when you look at it like, ‘Okay, where is the car right now? We’re on the track.’ You can pre-compute the ideal path to go through and then like, ‘Okay, are we tracking that really, really well?’ And I think, as soon as you deviate from that, the core challenge in a lot of the AI then is, how do you take that driving knowledge and translate that into, ‘Okay, well, if I do the corrections right now,’ you may even cause the car to flip over, right?

 

Peter Car (00:57:33):

It’s that skill of driving to realize, I know the optimal path. What we can’t do is pre-program, well, if you’re at this position and you’re deviating ever so slightly from the path or you’re really far from the path, here’s the correction to get you back on, right? The optimal correction to get you back on. I think that’s what makes it such a huge challenge to this problem, is that you have to recreate a lot of that, even intelligence and the experience that comes from driving multiple courses and knowing how to respond in any circumstance.

 

Nick Citron (00:58:04):

Yeah, that makes sense cause I guess when you talk about driving, so much of it is feel for the car and the best drivers are the ones that are able to drive it on the edge and not lose it. So you’re right, I can’t imagine how hard it would be for a computer to try and learn that skill.

 

Alex Roy (00:58:17):

There was a fun article in the 90s. I think it was in CAR magazine about… Who was the designer of the McLaren, the original F1? I don’t remember his name, Gordon?

 

Nick Citron (00:58:29):

Esteban?

 

Alex Roy (00:58:30):

No, the McLaren F1 came out in the 90s… Gordon Murray. So, the magazine asked Gordon Murray and several other famous car designers/engineers, to speculate on what a race car would be in 50 years. And they all discussed aerodynamics and X, Y, and Z. But I think Gordon Murray’s answer was different. He said, “Well, it would resemble like a high-speed caterpillar or a train. And the human on board would not have a traditional set of controls. You’d only have like a rocker switch. You’ve got left, right and maybe a button. And so it would have a pre-determined, optimal path around the track and the only thing the driver would have to decide would be, if it encountered another vehicle, whether to pass on the left or the right. The car would just execute the optimal pass, and then an emergency stop button.”

 

Alex Roy (00:59:36):

And I thought that that was really fascinating, and then I went to see an event called Self Racing Cars, which happens every year at Thunderhill. I went the week before the Indy Autonomous Challenge, and I watched some of the world’s leading minds attempt to get cars around this racetrack. And exactly as Peter Carr described, it was a lot harder than one would think. A lot harder. And so whenever people say, ‘Well, self-driving is achieved, it’s going to be here tomorrow,’ all great things can always be improved and will be forever. It’s the nature of invention.

 

Alex Roy (01:00:16):

Interesting, interesting. I have a lot more faith that data analytics is of value to sports and autonomous vehicles, but I still don’t feel like it’s really been implemented properly in Hollywood because you did need to be a data scientist to know that the Foundation TV show, just when it got to the script phase, it wasn’t going to be awesome.

 

Alex Roy (01:00:37):

Anyway, if you enjoyed today’s episode, please connect with us on social. We’re on Twitter @NoParkingPod. I’m everywhere on social @AlexRoy144. And please share No Parking with a friend, like us, subscribe, give us five star reviews wherever you listen to podcasts, and this show is managed by the Civic Entertainment Group. Until next time, I’m Alex Roy, and this is the No Parking Podcast.