Google’s DeepMind team is doing some amazing things with artificial intelligence. Namely, they are teaching artificial agents to solve problems with everything from motor control to high-level thinking. Now, this is normally where I’d make some “holy hell, robots and AIs are gonna take over the world” joke, but this is so impressive that I can’t muster up the cliché terror just yet.
Essentially, the DeepMind team is teaching programs to think and problem solve like humans do, and in a manner humans learn. Just like you and I learn problem solving skills by actually solving a problem, DeepMind is doing the same thing: giving an AI agent a problem to solve, and allowing it to learn through a trial and error process called reinforcement learning.One such skill they’ve been working with AI on is robotic manipulation and locomotion — how an agent can make something else move. And just like a human, movement and coordination can often come best when matched with a physical activity with clear indicators of success.
So how is Google teaching its AI locomotion? ant soccer, obviously! The agent is tasked with moving a tiny “ant” (it’s really more of a circle with four legs) around a playing field, and kicking a ball into a goal zone. While it seems easy, there are lots of things for the AI to control: leg movement, ball contact, ball movement, and scoring. Seeing this AI successfully play means it has learned not only how to move, but how to move effectively to achieve a goal and overcome problems like the ball moving farther than planned or not traveling far enough.
Okay, so now that we’ve given props to the geniuses at Google DeepMind, now we can freak out about the robot soccalypse.
What do you think about what DeepMind is doing? What other tasks would you like to see the team teach artificial agents? Let us know in the comments.
Image credit: Google DeepMind/Youtube