Humans have employed AI to identify what’s in photos, create artificial faces, and even play video games better than we can. For the first time, AI has created a sport. Nvidia and a design agency called AKQA fed neural networks data about sports that already exist, and they came up with a new one. It’s called Speedgate, and it’s a mix of rugby, field hockey, soccer, and more.
The team started with the rules for 400 of the world’s most popular sports — that’s more than 7,300 individual rules. That data passed through a recurrent neural network and a deep convolutional generative adversarial network. A recurrent neural network uses connections between the individual nodes to process sequences of inputs in time. A generative adversarial network uses two networks in competition with each other. One of the two generates an output, and the other decides if the output seems “real.”
AKQA and Nvidia ran the networks on clusters of Tesla GPUs to generate more than 1,000 different sports complete with rules and logos. Of course, machines have no regard for human safety, so many of the suggested games were too dangerous or impractical. There was one that involved underwater parkour and another based on exploding frisbees. And who wouldn’t want to throw balls back and forth while walking a tightrope between two hot air balloons? The humans on the team evaluated the outputs and pared them down to a list of 10 possibilities and then field tested three of them. Speedgate was the winner.
To play Speedgate, you need two teams of six players and a few flexible poles to create the gates. There’s a center gate and one at each end of the field. Three players on each team can go wherever they like on the field, but the other three can only cross into the other team’s half when their team has control of the gate. You get control by kicking the ball through the center gate, but you can’t step in the center gate area.
You can toss the ball between players, but you can’t run with it. If you catch the ball, it has to be passed or kicked within three seconds. If you don’t, the other team can try to snatch it away. You score points by kicking the ball through the other team’s gate in either direction. The winner is the team with the most points at the end of three 7-minute periods.
There’s a full website with rules, field measurements, and even a recommendation for what ball to use (a size 4 rugby training ball). AKQA is looking to make Speedgate a thing in communities around the world. There’s a page on the Speedgate website where you can sign up to get materials on how to start your own league. Unless that’s just what the machines want us to do.
- MIT Develops Algorithm to Accelerate Neural Network Evaluation by 200x
- Modular AI Wheelchairs Can Watch for Obstacles, Incorporate Head Tracking
- Engineers Are Using AI to Predict How New, Unknown Materials Will Perform