• 6 months ago
Shan, RJ, and Bobby broke down Sean Lee’s involvement with Adrenaline and how the company uses AI to predict NFL plays. They analyzed the "True Play AI" model, its role in MNFbroadcasts, coaches that confuse the computer, and more.

Category

🥇
Sports
Transcript
00:00With some Cowboy audio that kind of went viral.
00:03Yeah, we, uh, I was, uh, over the weekend or on Friday, I was actually trying to gather
00:08audio for something completely different.
00:09I was trying to get some audio from a MIT Sloan sports analytics conference that happened
00:16a couple of weeks ago where the Cowboys analytics director, John Park was at it.
00:20And while I was watching this thing and just kind of zoning out whenever it wasn't John
00:25Park talking, there was somebody from the ESPN broadcast who was talking about that
00:28broadcast. They debuted late last season on Monday night football.
00:31The one that was like Dan Orlovsky was on the call and it was like this analytics broadcast
00:35with all these different things. And the guy's talking about it and he just says, yeah, and
00:39we needed it to be really good. Like, like we needed the predictive AI to be really good
00:44about this and that. He goes, so we actually ended up getting a call from the league office
00:47and they were asking us if we wanted to partner up with Sean Lee, the former Cowboys linebacker.
00:53And I was like, oh, I didn't even know Sean Lee was doing anything like this. We had heard
00:57he was walking around Super Bowl doing media for something, but we didn't know what.
01:01Well, apparently Sean Lee is now the like one of the main investors and creators of this
01:06company called Adrenaline, which attempts to create an AI engine that can, for broadcasts
01:14and other thing, be a predictive tool in play calling. So he has really done a deep dive
01:21now into this analytics world and this AI world and understood how all of this kind of
01:25plays together. So we got some clips here. I found this. This is from Super Bowl week
01:29and it had kind of been buried on a LinkedIn page, but there was some interesting stuff
01:33in here. This was Sean Lee at this conference, this little panel that was hosted by Cynthia
01:38Freeland talking about a number of different things. The very first thing, how his analytics
01:42company Adrenaline is using AI.
01:45Yeah. You know, so Adrenaline, we are an AI company that predicts outcomes in games and
01:50we've built an artificial intelligence called TruePlay AI that integrates next-gen stats with
01:55our historical database to make dynamic play-by-play predictions. You know, as a linebacker, all
02:02I did was try to anticipate what type of plays were coming my way from the offense. And the
02:06only way to know that is to understand how teams use their personnel groupings, how teams
02:10use their formations, and how they use their best players. So that's what we do. In real
02:14time, we use the next-gen stats sensor data to tap into the formation. We identify that
02:19formation. We take it to our database, understand the historical uses, and then we make predictions
02:24on run pass, on targeted receiver, and on blitz probability. And really, everything that
02:29players and coaches use to game plan, that's what's in our AI. We say we want our AI to be
02:33like Peyton Manning's brain, like Luke Kuechly's brain. So all the data points that players use
02:39in real time, we're putting that into our AI.
02:41And I got some feedback. Very, very little, but there was a small subsection. Because obviously
02:46there's growing controversy about the role of AI in society with anything.
02:51And people not want, there are some people who are just like, absolutely not. If it is
02:54AI, it cannot be used as a tool for anything. And there's a bigger resistance to it.
02:58And I saw there were a couple people who were like, oh, come on. Like the classic throwback
03:03I study guy, the guy who just had the brain to do it, is now just turning this over to
03:08AI? Like, this seems so against what you would expect.
03:12Correct. And I think as you listen to this, and I've already reached out to the company,
03:16I was like, hey, can you get Sean on? That'd be nice if we could get Sean on.
03:20But when you hear him kind of describe how he thinks it can be used, I think you understand
03:24a little bit better. This is cut. We'll go 17, Peyton. Sean Lee, on why he chose to partner
03:30his tool with the Monday Night Broadcast.
03:32From a standpoint of, you know, being application-wise, working with Monday Night Football this year
03:39and building Monday Night Football Playbook, you know, really what you said, when we bring
03:43in the meeting room, meeting room to the living room, we're basically trying to teach football
03:49through the same lens that players and coaches use. I mean, really, NFL audiences are as informed
03:56as they've ever been, and we want to cater to those audiences. And we say that our probabilities
04:00are jumping off points for the broadcasters to be able to teach the why, like you were
04:04saying, the why teams are operating in the way they are. And we have two clips. Yeah,
04:08so this first clip, you're going to see, anytime the formation moves, our prediction's going
04:12to move. So, Walker...
04:14Who's going to get the ball?
04:15Who's going to get the ball?
04:16Okay.
04:16You've got the run-pass probability. You've got the bullets probability.
04:19Love it.
04:19In motion out, we got this, we identify the formation called Tripp's gone far. But not only are
04:24we identifying Tripp's gone far, we're identifying Tripp's gone far with JSN in the
04:28backfield, which is going to make pass probability go, you know, up to nearly 100%. These are
04:34the details that we're tracking. Now, why is he in the backfield? It's because, you know,
04:38San Francisco plays a ton of zone. This isn't a way to get their best receiver isolated on
04:43a linebacker. I should not be covering this guy. Troy should be covering guys like this.
04:48And I didn't watch this broadcast last year. I don't know if you guys did, the analytics
04:52one.
04:52Because Orlovsky was on it?
04:54Yeah, mainly. I mean, I could have muted it, I guess. But so they have three Monday
04:57night broadcasts. They got that, the Manning and then...
05:00The Manning cast isn't on every week. So they wanted a supplemental broadcast when Manning
05:04cast was off.
05:05Okay.
05:05And so it's fascinating that we always marveled at Sean Lee. Like whenever we would see a
05:12clip of him talking about something, it's like, how does he know like all these different
05:16things? Like how can he just look at...
05:17People want on the coach.
05:17He's correcting coaches and he's doing this and that. And when you see him talking about like
05:22just the fact of he says, look, when you look at the formation, got this. He goes, this
05:25is near 100% pass probability. And I think a lot of people sitting there, they may have
05:29an inkling there's a pass, but most people watching on TV would never think this is 100%
05:33going to be a pass. But that's the way Sean Lee's head works. So that's the technology
05:38as applied to TV. They got into a Q&A session, which is where I think the most interesting
05:43stuff comes from. So he was asked, this is clip 19, Peyton. He was asked specifically,
05:49all right, this AI tool that you have for broadcast. Well, how is this going to apply
05:52to teams? How are teams potentially going to take tools like yours and then apply it
05:57to in-game stuff?
05:59That's where I see AI going a lot, intra-week. Like as a player, when I first went, oh, we
06:04were just watching film constantly. There's not a lot of data. The coaches would talk to
06:07you, but if you had AI to help direct you, what film I should be watching, what plays are
06:12coming my way? You know, I don't need to work on this play, but I need to work on this
06:16play. To me, like we said, there's only so many hours in the week. If you can start to
06:20build some of that efficiency into intra-week, understanding how a team's going to attack
06:25you, practice one play 50 times that you know is coming. Don't practice the plays that
06:30you know aren't. That's going to make you a better football player over time. That's
06:33where I see it, intra-week being used for football teams.
06:36Sean, it seems like your data already can do predictive play calling, essentially. Is that
06:41something that you would draw a line against? You don't want to see coaches using AI live
06:46in games to make sense?
06:47Yeah, no, I would agree with that. But even so, so we make a prediction, 80% run, 20%
06:54pass.
06:54Well, guess what? That 20% happens. And as a linebacker, sometimes that 20% is your responsibility.
06:59So you still have to dive into the X's and O's. You still have to prepare and understand.
07:04You've got to prepare for the 80%. You've got to prepare for the 20%. So to me, using it in
07:10the
07:10game, I don't think it's actually going to help you that much.
07:13I think that's a really interesting angle because there's so many people that are concerned
07:17or worried about. There's an article from The Athletic we read last year that was like,
07:22could we one day see a play caller be a machine? Like essentially, could an AI unit become so
07:29advanced that they're able to look at probabilities, they're able to read formations or it is able
07:34to read formations and then it's able to just call a play. And if you need to check it at
07:38the
07:38line of scrimmage, fine. But the initial play call comes in here from the AI unit.
07:43And that's where Sean Lee's saying, no, you still can't lean on it because we're saying
07:4880% probability. That does not mean... There's definitely 20% that could happen.
07:53And when it happens, you cannot have somebody who's unprepared for that.
07:56So it's more... I think when we've talked about this before in life in general, the AI tools
08:03that are available, are you using it to make things easier or using it to do your work for you?
08:09And that's the thing that Sean Lee's saying can't happen. You cannot use a tool like this to do your
08:14work for you. And that's where you're going to end up getting in trouble. Sean Lee here, this is 19A.
08:19I thought this was an interesting question. He was asked about, given what he's seen with his model
08:24and studying in these games, which play caller in the NFL most frequently confuses his AI model?
08:30Sean McVay is, you know, really tough to track because he generally this year, he used a lot
08:35more 13 personnel, but generally he doesn't use a lot of just 11 personnel, similar formations.
08:41And so what we're doing is we're tracking the formation, the splits, where every player is
08:46specifically in line. And with Coach McVay, a lot of his stuff looked a lot of the same.
08:51The percentages seemed the same. Now this year they came out in 13 personnel, but then they would
08:56make it look like it was a spread 11 personnel type of formation.
08:5913 with three tight ends.
09:01Yeah, I should, you know, with three tight ends. But because he's so versatile as a coach,
09:06he can build his, you know, he can run his whole playbook when he has three tight ends
09:10on the field and only has one tight end on the field. So that's a guy I've admired, you
09:14know, when he was in Washington, I played against him a lot. It was always tough. And with our
09:18models, he is one of the better guys at being able to not be able to predict what's coming.
09:23Do trees of coaches tend to follow that as well?
09:25Absolutely. Especially, you know, Coach Shanahan and the way, you know, the versatility of all
09:29their formations. They do the same thing with the fullback, you know, Yerchek and then, you
09:34know, Kittle. They're so versatile that they can line up in what would be called pro-I, which
09:38is a big run formation. Then all of a sudden they're in an empty formation and you have
09:42a fullback who's running the slant and the ball's getting thrown to him. So that versatility
09:47of coaches based on personnel, based on formation is really what makes it tough for defenses to be
09:52able to combat. Sean McVay, the genius. Beating the computer. He's beating the computer at chess
09:58is basically what they're saying, which him to that point, it reminded me of something
10:02that I think I've referenced before, but I remember I was talking to somebody a few years
10:06ago and they had said that I was asking about a question about defensive philosophy and he
10:12kind of like bristled and he was like, well, he's like, there's really, he's like, this is
10:16kind of a controversial take. He's like, there's no such thing as defensive philosophy. I
10:20was like, but how do you figure? And he's like, because offenses are always advancing
10:24and offenses are always changing designs and they're always looking for ways to attack.
10:28If you are somebody who is married to, this is my scheme. This is what I do. You're going
10:32to be out of the league eventually because people are going to pick it apart. Defense
10:35is naturally reactive. You are reacting to the trends across the league and you're making
10:38your scheme that, and he's like, so there's not the only philosophy with defenses. You need
10:43to stop the offense and whatever way that gets done is the way that it gets done. So interesting
10:48stuff there from Sean Lee. And then one more note that had nothing to do with AI, but I
10:51thought was very interesting given his injury history. Because one of the first comments
10:54I got in reply, they said, Sean Lee tore his ACL giving this panel. Sean Lee, did you prefer
11:00turf or grass as a player playing on turf? You can wear the same cleats week in and week
11:04out. You get on grass and it's muddy and now you've got to put a different cleat on and
11:08you put studs on. I mean, there's just so much that goes into preparing for games. And
11:12I know each player has their own way of what they want to play on. But yeah, I prefer
11:16to play on turf too. Wow. I've never heard, especially somebody I feel like of like Sean
11:23Lee, who's like super cerebral and looks at things all these different ways. He's the
11:27first one I've ever heard to say, oh, 100% turf every day of the week. That's the one
11:31that I want to play on. I've thrown. My brains just stopped working. You know what I was
11:37expecting that when I played that clip, I was loaded up ready for him to go like chop to
11:42go. Well, I'm going to take a field or turf argument from the guy who can't stay
11:46healthy. The guy who's going to get hurt every time. He's the one who's going to
11:48lecture me about, about how a great turf is. I don't think so. But either way, I
11:52think it's fascinating stuff from, from Sean Lee. And the fact that I guess the
11:56application being right now broadcast, but what he really wants it to be, as you
11:59can hear is he wants teams to use it in prep work. He does not want them to use it
12:03in games, but he wants people to like start using tools like this to identify
12:07trends and then apply it to what kind of practice work are you doing?
12:09So,

Recommended