Following an impressive but not altogether unexpected rally in the fourth quarter, quarterback Tom Brady led the New England Patriots to victory in Super Bowl LI on Wednesday, beating the Atlanta Falcons 27-24.
Well, OK. Not the real Super Bowl. The simulated one. The Madden one. Every year since 2004, EA Sports has used their NFL-sponsored videogame franchise to predict which team will take the Vince Lombardi Trophy home. Equal parts marketing ploy and artificial intelligence experiment, the digital bowls showcase an intriguing side of sports videogames—and the Madden franchise in particular. They’re not just entertaining games to play with friends over beers in the off season. They’re rigorous, exacting recreations of real-life athleticism. Simulations that might, in fact, run better without gamers than they do with them.
Let’s take Super Bowl XLIX, for instance. In that game, the New England Patriots faced off against the Seattle Seahawks. In real life, the Patriots won, 28-24. In Madden ’15, the most recent title in the series at the time, the Patriots also won—28-24. While that may be the most striking success of these annual simulations, it’s not the only one. Not counting this year, which is yet to be determined, Madden has correctly predicted the Super Bowl’s winner an impressive nine out of 13 times.
Getting It Right
How do they do it? “It’s pretty straightforward,” says Josh Rabenovets, the senior brand director at Electronic Arts for Madden. The most recent version of the game—Madden NFL 17 in this case—is played on an Xbox One. They run one game, extensively recording the results and tracking all manner of statistics, which they then publish online alongside footage of the in-game events. “We don’t do much to it,” he says. “We just show it like it is.”
EA is able to keep it simple because showing it like it is means showing off what has become an exacting and increasingly realistic simulation of bodies and space. One that can operate without clear human oversight. Whole games can be played by simply letting the AI make every decision for the players on the field, without any input from the players on the couch. Left to its own devices, Madden will play itself.
Play By Numbers
Games like Madden weren’t always so accurate. A vast system of numbers determines how each player on the field will act, and early on they were all working from essentially the same set of digits. Every in-game player performed identically. To solve this, sports games implemented, using data provided thanks to licenses from professional sports franchises. After that, the digital guys on the field stopped playing like bots and started playing like ballers. The numbers were able to shape both the way they reacted to being controlled by gamers, and also the way they reacted when being controlled by the game’s artificial intelligence.
The NFL treats us like a 33rd NFL franchise. We get all their statistics and all their injury reports and things of that nature in a real-time fashion when the teams do. Josh Rabenovets, senior brand director at Electronic Arts for Madden
Over the years, Madden has developed probably the most complex ranking system out there, with dozens of individual metrics from strength to speed to situational awareness. Football itself is a difficult sport to simulate, and an even harder one to track statistically. In the NFL, only a handful of quantified individual statistics for each player are kept, and even those that are kept are less informative than they might appear. After all, every action a player takes on the football field is the result of an impossible to quantify alchemy between 22 individuals on offense and defense. In basketball, for instance, you can easily and neatly quantify how accurately someone shoots free throws. That’s a simple interaction between an athlete and the hoop. In football, every throw, every kick, every rushing yard gained or lost relies on at least two people, each of whom are responding to just under two dozen others. Football players are all tangled up with each other, making individual statistical profiles almost impossible to come up with reliably.
The NFL’s 33rd Franchise
To manage it, the developers at Madden have a process that’s both science and art. Going off of the statistics provided by the NFL, poring over game footage and every bit of publicly available data on every single player, the Madden developers assign numbers to individual ratings by hand, tweaking them as new information comes to light. They then use a weighted average of these numbers to provide an overall ranking that determines each player’s overall quality. On top of that, the data is constantly being revised and made available throughout the year to Madden players via downloadable updates.
“The NFL treats us like a 33rd NFL franchise,” Rabenovets says. “We get all their statistics and all their injury reports and things of that nature in a real-time fashion when the teams do. We’re able to use that data much like a general manager of a team would as we update our rosters.”
What’s even more striking about this labor-intensive process is that, up until two years ago, it was the work of one man, Donny Moore, who was affectionately called the “ratings czar” by developers and fans. Since his departure to the world of fantasy sports in 2015, EA has been a bit secretive about who has taken over his role. Rabenovets demurred when asked, but said the process remains the same.
And that process should remain the same—it’s proven very successful at predicting Super Bowls. Given the role human judgment plays in all this, it’s even more impressive that Madden is as accurate as it is. Projection is always an inexact endeavor, subject to chance and error. But 9-3 is a heckuva record. This Sunday, we’ll find out if EA is able to improve its numbers.
source : https://www.wired.com/2017/02/super-bowl-madden-predictions/