While it may be getting easier for humans to teach robots how to perform new tasks, there's still one potential problem – when a new robot is introduced to a work environment, its user may have to teach it the task over again, from scratch. That might soon no longer be the case, however. Researchers at Washington State University have devised a method by which computers can teach each other, freeing humans from having to do so.
The technique mirrors the way in which a human teacher and student would interact, and has so far been used to teach "student" computers how to play Pac-Man and StarCraft.
UPGRADE TO NEW ATLAS PLUS
More than 1,500 New Atlas Plus subscribers directly support our journalism, and get access to our premium ad-free site and email newsletter. Join them for just US$19 a year.UPGRADE
The algorithms utilized are based around telling the "teacher" computer – the one that already knows how to play the games – when to advise the student computer to take action, and when to let it simply learn by doing. This is a crucial component of any teacher/student relationship, as the student won't know what to do if too little advice is offered, yet they'll never truly understand what's involved in a task if they're always simply following instructions.
Lead scientist Matthew E. Taylor is currently designing a curriculum for the teacher computers that starts with relatively simple tasks, but builds up to ones that are more complex. He's also looking at incorporating principles from dog training into his technique.
Taylor acknowledges that it's sometimes possible to simply transfer the trained "brains" of one robot into another, although this sometimes doesn't work due to software or hardware conflicts. Additionally, he hopes that the technology can be used not only to let computers/robots teach one another, but also to let them teach humans.
A quick overview of his research can be seen in the video below.
Source: Washington State University