A robot in Georgia has been running around telling lies.
It says it’s going one way then heads another. It says it’s over here when it’s really over there, hiding.
In the course of investigating trust between robots and humans, researchers at the Georgia Institute of Technology developed algorithms that tell a robot when and how to deceive.
“It dawned on us that trust and deception were intimately related,” said Ronald Arkin, a professor in the Georgia Tech School of Interactive Computing. “But you need to find a time and place to deceive. No agent should be constantly deceiving.”
Arkin and his research partner Alan Wagner arranged a game of hide-n-seek with two WowWee Rovio robots available for $135.95 on Amazon.com. The hider attempted to “lie” about its location by knocking down a marker in front of certain pathway before heading in another direction. The downed marker fooled the other robot into searching in the wrong place. “It’s the same as leaving a muddy trail,” Arkin said.
In 20 trials, one robot successfully deceived the other 15 times. The deceiving machine felled the wrong marker a few times.
The Office of Naval Research funded the studies. Experts believe these experiments open up a world of possibilities and ethical considerations for using deceptive robots. Trickery is a regular part of warfare, but the military may not be the only place for such technology.
The research could influence video games, search-and-rescue efforts and even personalized care, Arkin said. Robots may need to mislead patients to get them to take their medicine, he said.