How AKQA Used A.I. to Invent a Completely New Sport

A.I. can write books, compose music, create commercials. But can it dream up an entirely new sport—one that’s actually fun to play? 

That’s the question some creatives AKQA posed when brainstorming ideas for an annual event called Design Week Portland. With a short time frame of just a few months, they decided to try. They fed data from about 400 sports around the world into an A.I. and challenged it to come up with something completely new. And hopefully, worth playing. 

The result is Speedgate.

Teams of six players each play on a field with one large gate in the middle and two smaller gates on each end. It’s played with a rubgy ball—and is described by the AKQA creators as a kind of mix of rugby, Ultimate frisbee and croquet. 

Muse met up with AKQA’s creative and tech leads on the project, Whitney Jenkins and Kathryn Webb, to hear more about the curious endeavor. As it turns out, much like in sport, it took some remarkable teamwork—in this case, between A.I. and humans—to create something that might just take off in leagues around the world. 

Muse: How did the idea for trying this come about in the first place? 

Whitney Jenkins: It started with us taking a brief that we do every year, which is an event in Portland called Design Week Portland. Like a local TED type thing. Typically, the agencies will do a tech crawl, or an agency tour, or something like that, and we thought it would be really cool if we utilized some of our new technology. We just thought, what if we used A.I. to create something? I think our only goal was to just have fun and maybe explore some new crazy ideas. 

That’s a nice starting point.

Whitney Jenkins: A lot of us work on athletic things, Nike and stuff. We love sports. We love basketball, football, soccer. And we thought, what could happen if we tried to make A.I. help us generate a new sport? That was really the genesis of it.

What were the technical challenges to start with?

Kathryn Webb: A big decision upfront was just what role the A.I. was going to play in the project. I don’t think we were ever under any illusion that we would make this kind of superbrain in the space of two weeks. So it was really about: Where can we leverage it in the process of concepting the sport, flushing things out? Choosing the right algorithms to help us generate a number of concepts. We generated over 1,000 different sports concepts by training a recurrent neural network on 400 existing sports around the world. 

And then we used a deep convolutional generative adversarial network—the shorthand name is DCGAN—we trained that on 10,000 sports logos, with the aim to not just generate something resembling a logo but to contribute to the brand design in some way. And then we were doing it in a very short space of time and operating with very limited data. I know 400 sports sounds like a lot, but when you’re talking about training an A.I., it’s not actually that many. And I think the whole project was maybe a month and a half total. 

Whitney Jenkins: Knowing limited time, limited budget, limited team—only about eight people—we thought we need to create some objectives for ourselves. So we set three goals as to what we would judge as a good sport. 

No. 1: Easy to learn, fun to play. We wanted something you could watch on the sidelines, and then was fun once you’re playing it. 

No. 2: Something that would make for a really great workout. Something you would play and be like, that’s way better than going to the gym. Something with a lot of cardio, running, kicking, catching, throwing. 

No. 3: Accessibility. Something like soccer, football, that’s very accessible. Anybody can get a ball. You can play on the beach. Very little equipment. We wanted something where anybody could watch and go, “Oh, we can get some sticks or just get a ball. We could play that. That seems really easy.” But then also accessible from the standpoint of, “If I’m in a wheelchair, can I play Speedgate?” “If I have limited mobility, can I play Speedgate?” Those were our three criteria.

Can you describe the game?

Whitney Jenkins: Sure. First, it’s important to remember that the algorithm is not just taking sports and putting it into like a slot machine. It’s not just spitting out combinations, it’s spitting out sport-like ideas. But us, as humans, we need reference points to understand. So we tell people it’s kind of a combination of Ultimate frisbee, croquet and rugby. The A.I. never said that—that’s how we understand it as humans.

You have a big field, like a soccer field. It can be played indoors on a basketball court as well. You have gates. A gate is basically two poles—we use six-foot training poles. There’s one gate in the middle and then smaller gates on each end. The basic objective is to kick the ball through the center gate, and that unlocks your end gate. So once you’ve kicked it through the center gate, you can then score on the end gate. So, in croquet, you’re hitting the ball through a series of hoops. We’ve got it down to two, the center and then the end where you score. And you can score from either direction. Kind of like hockey, you can go around the back [of the gate]. In Speedgate, you can kick the ball through from the front or you go around the back and kick the ball through the other way.

You’re playing against another team.

Whitney Jenkins: We’re playing against another team, and they’re also trying to kick it through the center gate to unlock their end gate. 

And after you score, it resets?

Kathryn Webb: After you score, the center gate resets. We use kind of a softer rugby ball because it’s all quite close proximity once you’re trying to score a goal, and there’s someone in the goal so we want to consider that. And you can’t move while you have the ball. 

Whitney Jenkins: Your center gate has this big X in it, and that means players are never allowed in there. It’s an out-of-bounds zone in the middle of the field. What requires you to do as a team is you’ve got to kick it to a teammate. And so by doing that, you can unlock this end and then go and work together to score. So one of the really cool things about it is, it 100 percent depends on teamwork. You can’t be a LeBron type character and steal the ball and take it all the way down by yourself. You have to work with your team because all gates have to be cleared with a kick. 

What did you learn from this as far as how good an A.I. is at creating a sport?

Kathryn Webb: I think what we learned and appreciated was the fact that it was always a collaboration. We had a lot of back and forth between the A.I. coming up with ideas and us saying, “That sounds good” or “That’s humanly impossible.” You do need a degree of human influence. We couldn’t have trained an A.I. at this point to identify if the game was fun. That was a real human investigation.

That’s a tough metric to understand.

Kathryn Webb: Yeah. We could think of ways to quantify if we had loads of money and loads of time! But I think that we really had a good working process. I think also we got a amount of insight into sport itself. There were things that came through when we were generating rules by looking at existing sports. There were a lot of kind of aspects of sportsmanship that we didn’t necessarily expect that came through in unusual ways. For example, there was a suggestion that at the beginning of each game you swap poles with the other team. As kind of a sportsmanly gesture. 

Whitney Jenkins: Like in soccer, the players will swap jerseys as a mutual respect. But the A.I. came up with that on its own, that you would swap poles. Or it said, “Remove the top of the pole.” So I guess at the end of the season you wouldn’t have any pole left. You’d have all these pole tops sitting around!

How about the name and the logo? Where did they come from? 

Kathryn Webb: We did actually train the model to generate names. And that was one of the areas where it made me realize how uncreative so many sports’ names are. The model was really keen, when we gave it some priming, to come out with whatever the prefix word was plus “ball.” We took a number of different words from the sport itself so—game, throw, catch, energy. And it was really keen to put “ball” on the end and we kind of wrestled with that for a while.

Whitney Jenkins: Football. Basketball. Baseball.

Kathryn Webb: That was one of the areas where it was a struggle to get a good creative input from the A.I. So yeah. 

Whitney Jenkins: Truth be told, “gateball” was one of the first results that we thought, “Oh that’s great. We love that.” Turns out gateball is a very small sport in Japan that is a form of croquet. But we didn’t want this to be a solo A.I. experience. We wanted to create a human experience. I think our goal was, Speedgate should be a sport that you could play, your kids could play, people could play and never know a computer touched it. It’s a great sport in its own right, and we don’t need to say, “It’s an A.I. sport, so that’s why it’s weird.” You could play this and go, “That’s so cool. How’d you guys come up with that?” It stands on its own. Once we played it a few times, we started talking about, “Well, let’s add words like speed or gates or zones or halves or kick.” And so Speedgate naturally came from the evolution of working through the A.I. names. 

Kathryn Webb: In terms of the logo, we built a logo generating model and generated around 6,400 different logo outputs. There was some real value in that. A lot of the things we took from that were the structure and the crest shape. The color scheme, we took a lot of that. Even just the positioning of the lettering. However, again, the way the visual A.I. is generated, you rarely get this kind of crips lines. Like someone would sketch something out.

Whitney Jenkins: Or a kindergartener was just having fun.

Kathryn Webb: In the same way that if you have an idea, as a creative, you’re just sketching it out first. It was at the beginning of the process. So the team helped take that forward into what you see now. But I don’t think we would have the same logo if we hadn’t used that.

Whitney Jenkins: What I love about the logo part is, the story we want to tell is A.I. as a creative collaborator. The reality is, and Kathryn says this a lot, with an infinite amount of time, data and money, A.I. could’ve come up with this whole thing, but we did this in a handful of weeks. So, use A.I. for the first step, which is exploration, then go to a designer and say, “Don’t start from scratch, start with all these ideas the computer came up with.” And the designer goes, “Cool. I’m seeing a lot of cool football crest type ideas. How about that? We like that.” It was like having A.I. be like a junior designer on the team, and to give credit to it in that fashion, I think, is really inspiring. 

Kathryn Webb: We work with a lot of sporting clients—obviously, Nike—and we have experienced branding the sport, but we’ve never invented a sport before. We didn’t really know where to start. So the intention was that we could boil down these 400 sports from all around the world and there would be some kind of understanding formed—that even from our own cultural perspective we wouldn’t necessarily have gotten to. So I think that that really helped to kick start the concept of the whole thing.

There are even Speedgate leagues forming, right?

Whitney Jenkins: We have 50 countries that have registered to create a Speedgate league. We’ve got 10 universities we’re working with to start Speedgate as an intramural sport this fall in a few months. When people play it, they’re like, “It’s fun. It’s easy. It’s kind of unlike anything I’ve ever played before.” So in that way, we’re really excited about the goals we set for ourselves.

Kathryn Webb: Something we’re really proud of as well is we’ve been working with the Oregon Sports Authority. They’ve recognized Speedgate as an official sport in the state of Oregon.

How long before it’s in the Olympics?

Whitney Jenkins: Oh man. Well, let’s see. When’s the next one?!

It certainly feels like an A.I./human success story. 

Whitney Jenkins: One last little thing. This is Gunner, the gate bot [see above]. One of the interesting outputs that the A.I. come up with on its own is, whoever scores the most points is called the Gunner. It just named that position. We don’t know what that means. But we thought, “Well, let’s make a robot that personifies the A.I. and the contribution.” And since it made the greatest contribution to Speedgate, we named it Gunner. So Gunner is our gift back to the A.I. to say thank you for helping us get there.