When you think of Facebook, a few things come to mind—pioneering the world of social media, privacy and data scandals, the staggering likeness of Mark Zuckerberg and Jesse Eisenberg. It’s hard to go a week without some sort of mention of Facebook in a major headline, but Wired recently took a look at a Facebook department that doesn’t attract as much press attention—Robotics.
Labs across the world are taking different approaches to make their robots as human-like as possible. A team at UC Berkeley, for instance, is using a method called reinforcement learning in order to teach their two-armed robot how to put a square peg in a square hole. Success equals digital reward. Failure equals digital demerit.
Facebook’s approach, however, is to focus on the human being’s innate curiosity, which is a major contributor to a person’s ability to learn and process new ideas. So instead of providing a reward or punishment for certain actions, the team allows the robot to explore freely and learn from non-optimal movements and actions. It treats the robot like a toddler who needs to fall off the couch once or twice before it realizes head plus hardwood equals ouch.
This method is known as self-supervised learning. Rather than the team constantly keeping tabs on the robot and disciplining it to a degree, the robot has to figure out for itself what the best movements are to complete its goal. Of course, the Facebook team is not a neglectful parent that lets its children run wild without any supervision. A major goal of the process is to collect more data to develop more effective software updates.
“Although it didn’t achieve the task, it gave us more data, and the variety of data we get by exploring like this is bigger than if we weren’t exploring,” says Franziska Meier, an AI research scientist at Facebook. The robot may not be meeting its immediate goals as fast as its contemporaries, but fitting a peg into a hole is not the point. Facebook is trying to craft a creative and flexible robot that approaches problems with an open mind.