When I was a boy, I had a friend who loved inventing new sports. And as I found myself, say, rollerblading while trying to strike a football with a hockey stick into a basketball hoop, I always remembered a bit too late how terrible these ideas were. For the most part, people have figured out the good sports already: Baseball. Basketball. Football. Soccer. But where humans may have peaked, computers might just be getting started.
Speedgate is the first sport designed from the ground up by artificial intelligence. A pet project of the creative agency AKQA for Design Week Portland, the team trained a neural network on short written descriptions and rules from roughly 400 different sports. And the words that the AI spit out in response were slowly refined–with a bit of interpretation by the people–into the game known as “Speedgate.”
“Our creative team spent a lot of time looking at a large spreadsheet (which is not the glamorized AI collaboration that most people may think of),” writes Kathryn Webb, AI Practice Lead at AKQA. “It took a number of iterations to get to a place where the outputs where both comprehensible and creatively interesting.”
For instance, if the system suggested descriptions like “a sport and volleyball, rink grappling,” the team would interpret that as “Volleyball game in a small room where you are attached to the floor.” In another case, the AI suggested “a sport event major participants of on leaps ball saw also a pommel , types ) two parallel , quickest played supine heights,” which the team translated to the hilarious sport of pommel horse sawing, in which “two people sit on pommel horses on opposite sides of a giant log, they rock back and forth with a saw.”
These new sports would certainly be spectacles to behold, but they weren’t the ideal new sports AKQA imagined them. In fact, AKQA set a high bar for itself: It wanted to create a game that was easy to learn, fun, accessible to many, and would make a great workout for those who wanted to move.