EA Sports says video game A.I. is poised for a big leap

The seventh generation, which saw the release of the Xbox 360, PlayStation 3, and Nintendo Wii, saw the rise of online gaming (and the widening of the player audience). The eighth, with the PlayStation 4 and Xbox One, brought a level of graphical quality that bordered on photorealistic.

It’s too early, of course, to determine the legacy of the current generation. But it’s entirely possible that 10 years from now, these will be the game systems where leaps in artificial intelligence is what everyone remembers the most.

The computational power of the PlayStation 5 and Xbox Series X is exponentially higher. The PS5 boasts 10.28 teraflops of computing power, compared to 1.8 teraflops in the PlayStation 4. And the Xbox Series X boasts 12 teraflops. While those figures are jargon to most people, they are an invitation to create smarter, more lifelike characters on screen for developers.

A.I. in games varies by genre. In action games, for example, you want an opponent who can act as a bullet sponge or who can push your skill levels to their limit. Role playing games often feature companions who fight alongside you. But in sports games, you’ve got up to 21 A.I. characters on screen, each of whom is often based on a real-world person. And that presents a special challenge. But developers at EA Sports say they’ve never been more excited to recreate those athletes in games.

“A.I. is very well set up to be the thing that defines this generation of console,” says Sean Ramjagsingh, vice president and general manager for EA Sports. “The extra computing power of the new consoles unlocks extra abilities for us from a development perspective and [between] what’s happening with the capturing of real-world data and the way data’s being leveraged now to drive different technologies, there are a lot of opportunities for us moving forward.”

A.I. training is handled differently among EA’s various sports properties, but the flagship Madden NFL series uses perhaps the most advanced methods.

The NFL has been keeping statistics about games and players since 1920. But in 2014, it outfitted RFID tags onto the shoulder pads of every player in a program called Next Gen Stats, later adding them to referee uniforms, first down markers and the pylons at the edge of the end zone. Using those, the league was able to capture a treasure trove of more precise data, including player speed, field location, and movement patterns.

Madden NFL

Madden NFL 22, which debuted last year, deeply integrated that data, allowing it to make onscreen players more realistic than ever.

“We took player route data from actual players and used it in the game,” says Connor Dougan, senior game design director for the franchise. “There’s so much cool data that we can use to make the most realistic representations. For example, the Baltimore Ravens use a pistol formation [compared to] the San Francisco 49ers, who get rid of the ball quick and have lots of motion in the pre-play. We use that data to model our play calling.”

Good A.I. is especially important in Madden, as 70% of the games are played against the console or computer, Dougan says.

FIFA

EA’s FIFA franchise doesn’t have quite that level of data available to it, but it, too, has adapted the ways it gathers data to make the player behavior in the game as close to real life as possible. While the developers continue to rely on league experts, who follow every team and gather as much data as possible, the studio also changed the way it did motion capture for FIFA 22, which has also informed the game’s A.I.

Rather than having players recreate the motion certain plays in a studio, they equipped whole teams with Xsens motion capture suits and had two play a full 90-minute match against each other. Rather than having data from one or two players at a time, this allowed developers to gather data in a full 11-versus-11-player match, resulting in a game that was much more lifelike.

Longer term, says Kantcho Doskov, gameplay design director on EA’s FIFA franchise, the team imagines a world where they can get data from real-world tournaments in a similar fashion to incorporate into the game.

“We’re at a place where graphics are really, really good,” says Doskov. “Over the next few years, A.I. is what’s going to improve and really, really take a step higher. When I look back at the first version of the game we put on PS4 vs. last year’s, there’s a big difference. If you look at them side by side, things look so much better, players behave much more intelligently.  Over that life cycle of (the next) five to eight years, I expect improvements to be made every year.”

NHL

The NHL franchise at EA Sports isn’t quite as big as the Madden or FIFA games, but it is also taking advantage of the capabilities of new consoles to improve the A.I. of its titles.

Madden NFL 22 introduced a feature called Superstar X-Factor, which showcased the league’s top players, recreating some of their unique offensive or defensive abilities in the game. And negotiations are underway to collect more data from the league.

“The exciting part is once we have that real world data, we can add more nuance,” says Ramjagsingh. “The complexity of capturing a team sport is a complex, complex process. We’ve been limited in the past by the data available to us, but with the computational power of (today’s) consoles, those barriers are removed. There’s no shortage of where we can take our A.I.”

When it comes to video games, artificial intelligence is both an ally and an opponent. Its purpose, though, isn’t necessarily to challenge players, but rather to enhance the gaming experience. And all of the areas where it intersects must work seamlessly. For example, in a Madden game, A.I. players must block, run play patterns, avoid other characters onscreen, and be aware of the clock.

It’s something that has to be welcoming to new players, who want a casual or even arcade-like gaming experience, but smart enough that it can challenge the most dedicated players, who follow the real world leagues religiously. And it has to do it in an unobtrusive way, almost invisible to the player holding the controller.

“People don’t really comment about the A.I. until it doesn’t do what they want it to do,” says Ramjagsingh.