AI's Crystal Ball: Decoding Football's Next Play

This interview features Professor D.J. Lee from Brigham Young University, discussing his football analytics project that employs artificial intelligence and computer vision. He delves into how his team uses neural networks and deep learning to detect players on the field, determine their positions, and recognize formations using video footage of football games. The ultimate objective is to fully annotate entire football games, extracting predictive data that can provide coaches with a competitive advantage. Listen in to learn more about this intriguing application of AI to college and high school football.

D.J. LEE
https://ece.byu.edu/directory/d-j-lee
https://news.byu.edu/intellect/new-byu-developed-ai-tecg-could-benefit-future-super-bowl-opponents
https://www.fry-ai.com/p/ai-football-coach-dj-lees-play-prediction-algorithm

FRY-AI.COM
https://www.fry-ai.com/subscribe
https://twitter.com/lazukars
https://twitter.com/thefryai

Creators & Guests

Host

Ryan Lazuka

The lighthearted Artificial intelligence Journalist. Building the easiest to read AI Email Newsletter Daily Twitter Threads about AI

What is Behind The Bots?

Join us as we delve into the fascinating world of Artificial Intelligence (AI) by interviewing the brightest minds and exploring cutting-edge projects. From innovative ideas to groundbreaking individuals, we're here to uncover the latest developments and thought-provoking discussions in the AI space.

D.J. Lee: I am a professor at Brigham Young University, BYU. I've been here 22 years. Before that, I worked in the industry for over 11 years. My research has always been in computer vision, you know, visual information processing. So, in about 10 years, well, 12 years ago, I kind of started using machine learning for my visual information processing.

So, been doing that since. And of course, I want to mention one thing, you know, when people think right now, okay, nowadays, when people talk about AI, they would talk about chat GPT, right? Okay. And what I want to say is that AI is not just chat GPT. It's got other things also, okay. And one of those many things that I do, it's using that AI for visual information processing. So, that's kind of my expertise.

Hunter Kallay: How did you get interested in that sort of thing? Is it something you always wanted to do, or was there somebody inspired you?

D.J. Lee: Well, I've always been doing visual information processing, so images, videos, that input from camera, okay, about like 35 years ago. Since the smartphone became so popular everywhere, the cost of using camera dropped also. And I think that push, that push the use of vision or visual information processing quite a bit in the last probably 15 years. So, it's a lot easier to process, I mean, to get videos and images for processing. I've always been interested in processing videos and images. And early days, I used that technology for processing images for food, inspection, agriculture, robots, that require visual information. So, that's kind of my focus for a while. And about 12 years ago, I kind of switched from traditional computer vision to more of the AI-based vision.

So, I have successfully applied that to, again, food and agricultural industries. I taught a few classes using AI vision for self-driving cars. And so, that kind of thing, that's my interest and my passion. So, besides football, I also use it for many things like facial motion for authentication. So, that's actually the biggest thing that I'm doing right now, is to authenticate, to verify your identity. You don't just show your face, you also use your facial motion to make a specific motion as your password.

So, you could just wink or you could do hello or you could do open sesame. So, that's what we, the biggest thing I'm doing now besides football. I know you're interested in football, but I want to mention that that's also very cool so that I increase the security. So, instead of people using a picture of your face to get into your system, they also need to know the secret facial password.

Ryan Lazuka: Yeah. So, do you think Apple will do sort of implement something like that for their phones eventually?

D.J. Lee: No, no, no. I think Apple has a face ID that works very well, right? Uh-huh. It's scanning the three-dimensional of your face and it's got infrared so you can do it in the dark. So, our technology is small for, you know, other authentication like on your laptop, check into a hotel, go to your hotel room, get in your car, you know, so just add a security to the traditional facial recognition. So, that's also AI vision, right? So, it's just using artificial intelligence to analyze your facial motion.

Ryan Lazuka: So, you have a lot, it sounds like there's two parts of it. There's the image part, you need a good camera obviously and those are sort of prevalent everywhere now, but you also need the computer science behind it to analyze what the camera is looking at. Can you answer a little bit about how that works with football or any other projects you're working on as well?

D.J. Lee: So, other projects, you know, so like I said about 12, 15 years ago, mostly you do in a more traditional way. So, human has to look at the image. So, give you an example. If I want to use this technology to inspect, say, apple or some fruit, I need to design an algorithm to look into what information I need. So, human design it, okay? I'm looking at the rot spot on the apple surface or I'm looking at a, you know, broken part or something.

So, I need to design algorithm to look at what I should look for. So, you have human involved. So, that's how it works. And then you, of course, you write your code to implement it. But now, a lot of these kind of applications, you don't get human involved. You just provide the images, say, this is a good apple, this is a bad apple, good apple, bad apple, good apple, bad apple. And you get enough of that and the algorithm learns. So, then you learn what's good, what's bad.

Ryan Lazuka: Do you feed those into some kind of program then, like computer program?

D.J. Lee: Well, now, you know, the most popular ones we use, like a Google tensor flow. You've heard of that, yep. We've learning and meta has the pie torch, you know, so these are the libraries or the tools that we use. And a lot of people design this neural network using these tools. And different networks perform different tasks.

So, that's kind of the trend right now going. So, you don't get too much human involvement. You just provide samples and it learns on its own. So, for like, that's kind of how it works for inspection applications. So, for football project, we have a similar network. We will go detect the players and then you have to differentiate players from anything in the background, you know.

Also, maybe logo or some painting on the grass. So, first task is detect the players and then you can track them if you have video, how they move, right. So, what we have done was to first we detect the players and we use neural network, you know, we call deep learning to do that. And then the second step is once we detect them, we know their location. Then we apply another neural network to determine what positions they play in. It's this player, a quarterback or this player, a receiver, offense line, or tight end or running back.

Okay. So, we call that position. So, we detect the location and we determine their positions. Then the third step is also using a neural network to then recognize the formation.

Okay. So, there are different formations in football, you know, they have formation families and I don't know if you play Madden video game. So, you can choose from different formation family and then pick a formation and then they just, you know, show you the players, you know, and then you hit play and then they start moving.

Yeah. So, our third step is actually looking at those, the locations and the positions they play in and then we can recognize the formation. Yeah, that's what we have so far and it's kind of limited to the data that we have because we started three years ago, we thought, AI, vision, it can definitely be applied to football. So, without any data, we just purchased a copy of, I think it was Madden 20. Okay. And I hire students to play video games.

Ryan Lazuka: Yeah, that's a great class to be in there.

D.J. Lee: And they collect a lot of images and we use those images to train our algorithm for all three steps, finding the players, determining their positions they play in and then the formation. Okay. So, that's what we have accomplished, right, so far. And it was limited to things that we, well, one thing I didn't realize until recently, actually in the Madden video game, you can actually record video like instant replay. So you play, you finish one play, you can go to instant replay to replay that play. And then you can position the cameras anywhere you want. And then you can record and we can record a video.

If that's the input, then we can do a lot of more than what we have so far. So we could take a video recording of the entire game and we can have the cut up for each play. And then we can recognize the formation of each play. We can even detect the down, you know, first and 10, second and three. Where you start and then, you know, so pretty much you can annotate the entire game and then you can cut out the part that's, you know, substitution or timeout or, you know, end of the quarter, end of the half.

So you can get rid of all of that. You only keep a video clip for each play from the beginning of the game to the end. So now with that information, I also know what formation they start for each play. And I can also track the movement of the player. So then you know how the quarterback move, how the receiver run their routes, you know, quarter running back, how they move. So you not only have the video that you can review, then you also can tell what formation and how players react in that situation. And with all this data, then it comes into AI, right? So you can then train your AI to even like detect the tendency or even predict or at least for training purposes, right? Yeah, that's incredible. You have the players prepare for the game and you have the coaches to gameplay.

Ryan Lazuka: So it's almost like you could say to, if you guys put all the data, say, BYU is playing, I don't know who the rival is, but they're...

D.J. Lee: Utah, okay, University of Utah.

Ryan Lazuka: Utah, okay. So say they're playing the University of Utah, you could analyze Utah's last 20 games or 10 games or whatever, however many, and you could find out the tendencies of players in particular situations. So the coach could say, okay, if this player on Utah is doing this, he's probably gonna react this way, like 80% of the time. Is that sort of the advantage of doing these kind of things?

D.J. Lee: Yeah, if we have video.

Ryan Lazuka: If you have video, oh, which is harder, which is hard for college, I'm probably, I'd imagine.

D.J. Lee: And the way they record it, it's quite different.

Ryan Lazuka: I was reading an article, you need overhead shots, right? Like that helps you guys.

D.J. Lee: The higher the better, but it doesn't need a bird's eye view. Just higher, like BYU, if we could mount the camera by the press box and look down the whole field and you get the data automatically. That's just as good as bird's eye?

Yeah, I mean, we can live with that. Of course, higher is better. Bird's eye view is ideal, but I know it's probably not possible, but the higher the better. So you don't need to hire people to manually collect that data. You can just run it through and then you get the data.

Hunter Kallay: Why is it the case that higher is better? I was thinking maybe being closer to certain players you'd be able to see certain movements better, why would you wanna be higher? So you'd see the whole field.

D.J. Lee: Okay, so I mean, you don't necessarily have to see the whole field, but you do want to be able to see all the players for one play. You can turn the camera, okay? But you want to be able to be higher up so that you can see all 22 players. Right now the data we have, the camera is looking at from the back of the offense at about 30 degrees above ground. So now you see in that case, the quarterback will be most likely blocking the center and the offense line quite often blocked the camera view to see the defense people, defense line. So that's kind of the biggest challenge. And it's something that can be a void, the challenge can be a void if we can have video from higher up. The higher the better because you don't have the occlusion problem.

Ryan Lazuka: Is it possible to get a drone to record the games? I don't know.

D.J. Lee: I mean, would that be cool? Yeah. Or even like the blip, you know?

Ryan Lazuka: Yeah, yeah, the good year of blip, yeah.

D.J. Lee: Yeah, just stay up there, look down. Yeah, some NFL, they have the camera that controlled by wire that you can hover over, but they move a lot. Sometimes they zoom in, you know?

Right. You see a specific player or play. So you want to be able to really get the data. You want to be able to see the whole, the entire play plays out.

So you see all 22 players. Right now it's a little bit challenging. I mean, we're doing okay. We can detect the players about, you know, 95, 96% of the time of the players we have in the data. So we occasionally miss a few because we know you got to have 22 on the field. So if we detect 21, so we know one, it's missing. And we based on where they are, we could kind of guess, you know, it's the center that's missing. Or, you know, just there are some rules that can be applied to mitigate that kind of problem. But it would be ideal to see it from high up and entire play. Of course, the best would be to see the entire field.

Hunter Kallay: So how did you get this idea to start applying it to football? Was it one of the students that came up with it? Was it something that you came up with?

D.J. Lee: Well, probably collectively, you know. So I like to watch football games, you know, college or professional. And three years ago we were thinking, what can we do? Maybe we could help BYU to play better, you know.

Ryan Lazuka: After a big loss, you're like, we can help here. Yeah.

D.J. Lee: So then we talked to them and then they gave us some videos to work on. But then the videos we got, I'm not quite ideal because they zoom in or zoom back out. So that's why we went to use Madden game to collect data. Just for proof of concept, right? So, yeah, because I'm working on research in this area. So I just watching football game, I was like, no, no, wait, if we could do that, that'd be nice. You know, so.

Hunter Kallay: Yeah, yeah. When you're in a certain area, you view everything through that certain lens of research, I understand.

D.J. Lee: Exactly. I see an apple, it's a supermarket and I say, oh, is that good or bad? You know. Oh, that's so funny. Color is not good. Oh, shape is weird, you know. This should be used to make applesauce. It shouldn't be in the, you know, right? There's something. Yeah, it's kind of a, I guess, occupational hazard.

Hunter Kallay: Yeah, from a philosophy background, I definitely understand that you're viewing everything. You don't look at the world the same way some other people do. But so how many students are working on this with you? Is it primarily you or?

D.J. Lee: First, I had two students work on it. And then one left. The other one finished to where we are. And then two more students took over just for a little bit because we didn't have any resources. So I kind of redirected focus to the facial motion authentication because feel like that's more promising. We have two PhD students could go into this, you know, if we could get some, the collaboration or some funding. But BYU, University Communications Department send out a press release in February.

So right before Super Bowl. And they wrote an article about our research and they released that. And since then I have received multiple requests for interview, phone interview or Zoom, you know. And there are several articles came out some from universities, you know.

And so because of that publicity, this project looks like it's going to come back in life. So I get called from football coaches. Also talk to a company that does the sports analytics, you know, for football, other sports. So I'm getting some, you know, like interest in this. So I hire another student just last week to talk to them about the project. pick it up and work on it more.

Hunter Kallay: Is the BYU coach in your ear wanting you to get this done or what?

D.J. Lee: Well, I talked to one person on the team and he's very interested. He told me that, oh, this could go further. You could make it a very good training tool. But then the football season starts and then everybody gets so busy. Yeah, so my plan is to go back to Madden and do the instant replay to collect more images and videos and really prove that, hey, this thing could work if we have video and then hopefully, we can take it somewhere.

I'm sure there are other people, teams maybe trying to do that, but I think we have a pretty good focus, right? So annotate video, annotate the entire football game, extract the data. And once you get that data, then you have many different ways to use that data.

Hunter Kallay: Yeah, and it seems like football nowadays is all about analytics.

D.J. Lee: I think, yeah, it's kind of going that direction.

Hunter Kallay: A lot of coaches are making their decisions based on analytics. So I think that would actually adopt this sort of thing openly.

D.J. Lee: Yeah, I would think so. So right now, maybe people are doing it on their own or they have some subscription for some service, but you got to automate that. If you can automate that process, then the data you get will be a lot more than what you can do manually. You mentioned, so because of this press release, we're also helping annotating tennis match. Yeah.

Yeah, it's easier. You only have two players or four players, okay? And they don't, I mean, they don't have weird formations and trick plays, you know? So, but we're doing...

Ryan Lazuka: Your angles are probably really good for tennis, I'd imagine they're all over.

D.J. Lee: You can mount it at fixed location for the entire match that you don't have to move it. Yeah, things like that. And there's a lot of things you can do, you know, using AI, you know, so match AI with vision. It's quite interesting what you can do, you know?

Ryan Lazuka: Yeah, so you guys are, sounds like you're one of your prime focuses on collecting good data right now. Do you have a program to analyze the data for football that is like open source or some kind of software program?

D.J. Lee: No, we have software can do the formation recognition now, but it's only for the still image. Okay. That we can do already. Gotcha. And it worked very well. So what I'm thinking is to have a higher review of the whole play, then we can track how the players move and then you can expand from there to annotate the entire football game.

Ryan Lazuka: Gotcha. So that's the goal is to create a software to analyze an entire football game one day. And use AI to predict the player movement and things like that.

D.J. Lee: Yeah, but you want to use AI sure to analyze the data, but you need to have data to be analyzed, right? Okay. So you need to have a, you know, say for example, this team, you may want to have the last 10, 20 games and you can extract the data automatically. Then you can apply the data to your algorithm to analyze.

But other than if not, you just, it's just a long process manually doing that. And also it's not accurate, right? So using video, you can actually locate the players very accurately how they move. You think if you do it manually, it won't be very accurate how they move.

Ryan Lazuka: Like when you say how they move, what do you mean by that?

D.J. Lee: Like, you know, say once the ball is snapped, all 22 players will be moving, right? Okay. Where they move. So how from this formation, how they move, you know? So for example, the quarterback, how the quarterback move once the ball is snapped while the defense back, how they react to it.

Ryan Lazuka: So you're saying like how they move as in analyzing how they move and accurately recording that data in some kind of.

D.J. Lee: Yeah, from the video, you know, they move, right? Yeah. But we can detect the player and track their movement so we know how, for example, how the receiver is run. You know, I'm not a coach. So I'm still learning what kind of data would be useful. Okay. But I'd imagine if we can do what I just described, then there would be a lot of data that you can extract from there.

Ryan Lazuka: Yeah, I'm sure like one of the things you could probably do is if you analyze a bunch of data, you put it into some kind of software eventually one day, have AI analyze it. And you can say if the running back lines up behind the quarterback, this far off on the right side, 90% chance he's gonna run it off the middle or 90% chance he's gonna take a sweep to the right. So then that gives the defense a huge advantage. And also the situation. Yeah, third down, second down.

D.J. Lee: Yeah, yeah, the down. First and 10 or it's a third and nine. And also the score. Are you leading or are you trailing in the time? The clock management, right? So are we in the last 30 seconds?

I think you react differently. Yeah, so it's information to be used, right? But then you need to have the information. And this you cannot generate, right? This you need to get this data from the actual game.

Hunter Kallay: How important is that you track the ball? So we're talking about player movements, like tracking the ball. I mean, I'm sure when you look at player movements, you'd be able to tell this is a pass play, this is a run play, just by the way that they're running, or the way they're blocking. But is it important that you track the ball and if so, how?

D.J. Lee: Tracking the ball would be challenging, okay? It's not impossible because you could get information from how they move. I mean, kind of predict the running bag has the ball or you don't, kind of other information you can learn. But you do need to have good quality of the video if you wanna do that.

Hunter Kallay: I'm thinking it's possible if you look at the way, because we know the ball is gonna start with the quarterback. So if you look at the way the ball, who the quarterback maybe gets close to, if he gets close enough to the running bag, we can maybe look at the defenders, who the defenders are kind of going towards, could have the ball or something like that, I'm thinking.

D.J. Lee: Yeah, yeah, yeah. I mean, defenders could be full. That's true too. If they take it well.

Hunter Kallay: Right, you'd see it later in the play when the play ended or something like that. So this is what we have

D.J. Lee: dealt with, because we are able to work only on the image right now. So if I have the recording of the entire play, I can process it based on all this you just described, right? In the end, where all the defenders move toward, and where it's stopped, and then you can kind of predict or kind of analyze and say, yeah, the ball, it's the running bag at the ball, or the quarterback, tuck it in and then run for it, or the pass, you know.

Hunter Kallay: Especially in BYU and other college teams that run the option, and they fake the handoff so the defense gets tricked. I wonder if there would be a way to see, at the end of the play, whoever gets tackled, there would be a way to backtrack it. It's a good point, but now that we backtrack the play, we know he had the ball the whole time.

D.J. Lee: Yeah, so if you know the movement of the players, then you can see why they all run toward that, in the end, right? So even if you don't see the ball, you could still extract some use for information for it.

Ryan Lazuka: Have you had any NFL teams reach out to you?

D.J. Lee: No, no, no. They're probably doing something they just won't tell you, right?

Ryan Lazuka: Okay, yeah, yeah. They have their own program.

D.J. Lee: Yeah, but I'm thinking, you know, if I could help at the... Well, you got to start somewhere, right? But right now, I think I'm going to try to start from high school. Yeah, that's a great idea.

I was told that high school plays are, you know, simpler, and it's easier to recognize and predict. And then we'll go from there, and then, you know, once you get something, yeah, maybe, you know. One day they will come and say, hey, we like this, you know. I don't know. We started this for fun. Like I said, you know, we were like, let's work on something else. But then with the publicity that was generated by BYU press release, No, I'm having a second thought, you know.

Yeah, so hire a student. I'm going to go back and actually get videos from the Madden game. And I also have access to some high school videos. And we're going to try to use those and see if we can do something.

Ryan Lazuka: So when you're analyzing the Madden game, is the goal, it sounds like it's two different worlds. You got the Madden world, the data for that and the data for the college football or the real world. For the Madden world, is it you put in so much data after a while, you can play a game and have the data predict what's going to happen? Is that the goal of it?

D.J. Lee: I'm hoping. Right now, the way we use it is that we, because when you play the game, you can pick the formation, how you want to play the offense. And then you record, you can do the instant replay to record it. So now we can use that data to do, to improve what we have got. So we could, we could find a player, we can look at the players, find what positions they play in, what formation, and then we can track them, how they move. So this will be good data, very good data for training our network.

Ryan Lazuka: I mean, I, it's like, I think the options are endless here. Like one of the things that could happen is, like eSports is a big thing. I'm not a big video game player, but I'm sure if some eSports guy wants to leg up on his competition, this could help him because he can have, he has all this Madden data and he'd be like, okay, I know what this guy's next play is going to be.

I'm going to call this play because he thinks I'm going to call this play. You know, like, there's big money in that too. I mean, I don't know. It's just, it's interesting.

D.J. Lee: Yeah, it's interesting. But right now we're just hoping, but we could do what you just said too, you know, but it's easier to collect data that this way. And once we collect the data, we train our algorithm, and we can go back, get more data to test it. And then we can train it more and then we can go get the real data to train it.

Hunter Kallay: Yeah, I think starting with high schools are really good idea with a simple place. I think once you get to college in the NFL, especially like the higher level college teams, a lot of the routes that they're running, the receivers are running, a lot of them are improvisations. So they're not even planned a certain way. It's just based on where the, how they're going to react.

Yeah, which I also think if this developed enough, you could be able to predict that sort of thing too. Not only the set play, but maybe if they're going to run this play, if we lay off the receiver a lot, he's going to tend to do this sort of thing, like individual player scouting too. Or if we press up on this receiver and play him close, this is the sort of thing that he likes to do. So you can get kind of individual scouting tendencies.

D.J. Lee: Yeah, like you look at it and say, well, that's not very smart move that way, you know,

Hunter Kallay: the coaches could say, because when coaches are putting in these game plans, like if a cornerback's guarding a receiver, covering a receiver, a lot of times they'll, they'll focus all week, not just on the team entirely, but that one player. Right, like you're going to be covering number 18 this week. So let's look at his data. You know, when it's third down, this is what he likes to do.

When you play, you know, 10 feet away from him, this is what he tends to do 80% of the time on third down. So it could really help individual player analysis in that way too, which would be really cool.

D.J. Lee: And you could also be individual training of individual player.

Ryan Lazuka: That's true too. Because you can point out their bad tendencies. Is that what you're saying?

D.J. Lee: Or like, If you have this video or I mean the data, and then you can see, okay, if we're going to play this formation, you know, for defense, can I recognize the offense formation and how they do it, right? So, and for the offense, you can see, okay, with this formation, I recognize, and I'm the, I'm the receiver, and the defense move that way. What would be the best way for me to move in this situation?

Hunter Kallay: Just in recognizing the formations is such a huge step to be able to do.

D.J. Lee: Yeah, as far as I know, for at least high school level, you don't have that yet. And I just, I Google it the other day, there are like 16,000 high school football teams in the U.S.

Ryan Lazuka: And just like high school football is huge. I live in Cleveland, Ohio, Hunters in Tennessee now because it's going to school there. But yeah, high school football around here is huge and some teams have gigantic huge, like there's one team by St. Ed's is a football team by us. And they're a football team, their offensive line is like as big as NFL teams, you know, high school high school football players are like 340 pounds or something on the offensive line. So this would be a huge help to a smaller team that needs that is trying to compete with them because there's no way they can deal with them physically.

They need to leg up some mentally, you know, yeah, that's true. But then there's another thing that I saw something recently where AI is being used to help college football teams recruit high school kids. And this might be a good application for that as well, because they can see the tendencies of the of the kids they're trying to recruit instead of going to see them in person.

D.J. Lee: But you got to start with data.

Hunter Kallay: Is there anything else about the project you wanted to share with us that you thought maybe we didn't ask enough or that you thought was interesting about the project?

D.J. Lee: No, I think you you have asked all the good questions. The fun ones.

Hunter Kallay: It's a very interesting project and I think it definitely would be adopted by coaches. I mean, if you think about it, ultimately the coach is probably not going to go completely off of the data, but I don't think that's the point of what you're doing. Anyway, you know, even if you got it to that point, I think it would be an aid to the coach that this is what you can do. And then the coach can take that information and help make a more help him make a more informed decision. Is that correct?

D.J. Lee: Yeah, there's a one of the students is show me today that Amazon Prime Video is adding features in football game when you watch as a viewer when you watch the football game. It will run in real time to tell you what would be the best play.

Wow. So get get get viewers involved in the game. You know, it would be very interesting to watch a game like that. You will hear a lot of say, oh, that's stupid.

Ryan Lazuka: Everyone knows the best play.

D.J. Lee: Yeah, and that's I think that's applying AI for it.

Hunter Kallay: So is there a way that we're able to see the latest updates on the project if there's ever any updates will, you know, will be why you will your department release anything or is there any way to follow kind of how this is going?

D.J. Lee: Well, yeah, sure. You know, we, I mean, it depends, you know, usually when we accomplish something, we will publish a paper like we did early this year. But then, you know, if we get to work on something like I, I describe, and is sponsored by someone who doesn't want people to know, you know what I mean? Yeah, then maybe not. It depends, you

Hunter Kallay: know, but he wants to steal this idea or something.

Ryan Lazuka: I mean, if you want you football ends up in a national football in the national championship game, we know your project's doing very well.

D.J. Lee: I don't know. Oh, any, or maybe a high school in Cleveland, right? Yeah, there we go. No, we're just, we're just doing it mostly because it's fun. And we know, you know, we have the knowledge to try out a few things. And we're just hoping that it could be, it could become useful. But I don't think it will ever replace the coaches. Okay, so for sure. Imagine you will have two supercomputers play it against each other. Yeah, that's crazy to think about. Then you don't even need players anymore. Right.

Ryan Lazuka: That's one rabbit hole after the next. No, you just use

D.J. Lee: two supercomputers are playing a Madden game. You know, I don't think it's, it's going to happen, right? There's the limit. Okay. Because then you take, you take away the fun. Then why, why are you doing it? You know, why are you watching it?

Hunter Kallay: There's fun in being able to predict what the other teams going to do and try to outsmart them and now perform.

D.J. Lee: That's fun. Yes. Strategies outsmart your opponents, right? That's fun. But, but you can use all kind of tools you can find, but you can have, you can have the actual plays, you know, designed by a computer in real time. Right.

Like, you know, in baseball, they have the already have technology to show the strike zone and tell you if it's strike or not, right? It's outside, outside, but why they don't replace human. Yeah, that's a great point. Because it's fun. You need to have human involved.

Hunter Kallay: And you'd have nobody to yell at it when your team.

D.J. Lee: That's true. That takes away a lot of fun. Somebody's got to scream at the TV. Oh, yeah, you gotta have that right. Yeah, as much as it advanced, it just have to stop somewhere.

Hunter Kallay: If anybody watches and wants to learn more about the project, be sure to check it out. Check out the department, look for updates on papers if anything comes out. And then be sure to subscribe to Ryan and I's newsletter. It's a weekday newsletter providing cool updates and the latest tools in artificial intelligence called Fry AI. The fries are just for fun because Ryan's daughter likes French fries. So you can find that at fry-ai .com.