Until now, humans have had one significant advantage over computers and robots: We were the only ones who could teach the machines how to function. Now, researchers at Washington State University have created computers that can teach other computers. And they're using Pac-Man to do it.
Dr. Matthew E. Taylor, professor of Artificial Intelligence at WSU, developed a method that allows a "teacher" computer to give advice to a "student" computer in a way that mimics how you might teach a fellow human, except with an algorithm. The trick is in how often the teaching computer gives pointers—too frequently means the student isn't learning, and too sparse makes the learning process slow.
"We designed algorithms for advice giving, and we are trying to figure out when our advice makes the biggest difference," Taylor says. With the right timing, the "teacher" machine was able to guide the "student" machine through learning how to play Pac-Man and a modified version of StarCraft—eventually leading to the student machine playing more skilfully than the teacher.
The implications here are huge. In the future, machine teaching could, for instance, allow outmoded factory robots to teach their replacement equipment, no downtime required. Eventually, the long-term goal is to create robots that can teach humans.