How do you teach your robot to learn? This is the question that Sergey Levine, an assistant professor in the department of electrical engineering and computer science at UC Berkeley, is trying to answer.
“One of the most important things is that you have to somehow communicate to the robot what it means to succeed,” Levine said in a conversation with AI Podcast host Michael Copeland. “That’s one of the most basic things …You need to tell it what it should be doing.”
During Levine’s research, he explored reinforcement learning, in which robots learn what functions are desired to fulfill a particular task. He’s also quick to point out that it’s important that the robots don’t just repeat what they learn in training, but understand why a task requires certain actions.
“So you observe people doing a task, try to infer what are the goals these people are optimizing, and then the robot can attempt to optimize the same goal itself,” said Levine.
It’s not easy, though. Teaching a robot to learn, instead of to just recognize images, is more difficult than building a deep learning system that can recognize images. In reality, the learning process is a lot of “trial and error.”
“If you want to get a robot to do interesting things, you kind of need it to learn on its own,” Levine said. “This is not something where people can just give you these perfect labels like they do for image recognition.”
Levine’s fascination with robots and his current area of research stemmed from two things: his work in computer graphics, and a Disney Animation film titled Big Hero 6.
One of the film’s characters is Baymax, a robot built by a graduate student to assist with people’s medical needs.
“The thing that I really liked about that movie…[is] this idea that scientists can build a robot as an experiment but for the purpose of actually helping people,” said Levine. “The fact that something like that makes it in popular culture and a lot of people see it, that to me is actually quite inspiring. I think we should be thinking more about robots in this way.”
AI Podcast: Move Over Sherlock, AI Is Here
And if you are a frequent PayPal user, give last week’s episode a listen: Vadim Kutsyy, a data scientist at PayPal, discusses how the online payments company uses AI to crack down on suspicious transactions.
How to Tune in to the AI Podcast
The AI Podcast is available through iTunes, DoggCatcher, Google Play Music, Overcast, PlayerFM, Podbay, Pocket Casts, PodCruncher, PodKicker, Stitcher and Soundcloud. If your favorite isn’t listed here, email us at aipodcast@nvidia.com.
