The list of games at which humans can still defeat robots is growing distressingly short. We’ve lost chess, checkers, and Jeopardy. Computer programs can challenge top human backgammon players. Now add to that list rock-paper-scissors. Of course, there are those who would say it’s not even a game, since it’s supposed to be pretty much random—a way to force a difficult decision when there are no coins around to flip or straws to draw.
Yet, in video posted online by engineers at the University of Tokyo, viewers can watch a robot hand they developed repeatedly trounce its human antagonist in Janken (the Japanese name of the game). When the human goes rock, the robot goes paper, when the human goes paper, the robot goes scissors, when the human goes scissors, the robot goes rock. Over and over and over.How does the robot do it? It cheats. Using super-high-speed image processing, it divines what shape its opponent’s hand is in the process of forming, and then, before the human eye can comprehend what’s happening, arrange its three “fingers” into the shape that beats it.
In that way the Janken robot has less in common with IBM’s (IBM) Deep Blue or Watson—computers that beat human beings at games involving something like thought and complex decision-making—and more in common with robots that walk and scurry around bumpy terrain and imitate the facial expressions of people looking at them. It doesn’t really think, it reacts. What’s striking about the Janken robot, and what makes it so deadly at its job, is that it reacts very, very quickly.
According to Mashatoshi Ishikawa, the professor who heads the lab that built the robot, it is able to process images at 1000 frames per second, as compared to 30 frames per second for humans. And the hand moves at speeds that put the most accomplished human gunslingers and pickpockets to shame: Its three fingers can make ten 180-degree motions and 40 45-degree motions per second. Ishikawa claims that both in processing and movement, the Janken robot is the fastest in the world.
Ishikawa and his researchers see the Janken robot as a test case for technologies with wide applications. His lab is working on using the robot’s image-processing technology to make phone and computer interfaces faster and more intuitive, as well as to improve medical imaging. The robot’s ability to link processing and movement could help make factory robots faster and able to keep track of and manipulate more objects—and make computer-controlled vehicles more nimble.
Since there is no human roshambo player of the stature of a Kasparov or a Ken Jennings—or for that matter, a John Henry—with which the Janken robot can do public battle, it is possible that it will never lose. Unless, of course, it meets another Janken robot. When I asked Ishikawa by e-mail what would happen in such a match-up, he said he thought they would simply cycle through the different forms; each would predict what the other was going to do and react accordingly, only to trigger the opponent to, in turn, react and change shape, flitting from rock to paper to scissors to rock every 10 milliseconds. To us, it would appear just as a constant blur, a robo sign language we are too slow to understand.