It's bad enough we have to worry about the machines beating us at ancient Chinese board games; now it appears we'll also have to worry about getting served by them in random dance-offs. The videos below might look like the shameful homework of a freshman animator, but in reality we're seeing some software working its way toward inventing moves that might make even Beyoncé weep with envy.
The concept is called generative choreography, and our previous coverage of it has mainly focused on programs that generate patterns and shapes for human dancers to follow. Now, thanks to a joint effort between Swedish developer Peltarion and the Scandinavian Lulu Art Group, there's software out there that can come up with its own sweet moves.
Peltarion calls the associated system "chor-rnn," and it sees the technology as useful for "collaborative human-machine choreography or as a creative catalyst, serving as inspiration for a choreographer." Peltarion and the Lulu Art Group gathered the foundational data by capturing over five hours of contemporary dance movements from one of the Lulu Art Group's living dancers with a Microsoft Kinect camera. It then input the data into one of the artificial neural networks Peltarion specializes in—a form of technology that, much like a living brain, analyzes information from multiple sources and creates dynamic responses based on that information. And for 48 hours, it taught itself how to dance.
It's beautiful, in a way. While greatly truncated, the footage shows a pulsating blob of lines and dots evolving into a creation that knows how its joints and limbs interact with each other. Six hours in, it's still unsure of its movements, but by the time two days go by, it's gyrating its hips, making a stab at moonwalking, and looking generally comfortable with itself.
For the most part, the moves are still simple stuff. But for two days of practice, that ain't bad. I've been around for years and I'm still trying to do what it did six hours in.