Deception is something that people do all the time — and it
plays an important role in military strategy. Now some researchers are trying
to figure out how to get robots to do it, by looking at the behavior of
squirrels and birds.
At Georgia Tech, a team led by Ronald Arkin, a professor at the
School of Interactive Computing, studied the literature on squirrels hiding
their caches of acorns. Squirrels will hide their food in a certain place, but
when they see other squirrels trying to steal from them, they attempt to fool
the thieves by running to a false location.
Ronald Arkin, a professor in Georgia Tech's School of Interactive
Computing, and his Ph.D. student Jaeeun Shim, used that as a model for robot
behavior. They programmed the robot into tricking a "predator" machine by doing what a squirrel does: showing the adversary a false location for an important resource.
DNEWS VIDEO: Who’s Your Daddy? Sparrow Gets Around
The team also looked at how other animals -– in this case, a species of
bird called an Arabian babbler –- drive off predators. Babblers will make an
alarm call when they see a predator and other babblers will join the bird and
make more calls. They then mob the predator, all the while flapping wings and
making noise. The babblers don't ever actually fight the animal they want to
drive off; they just make enough noise and flap around enough that attacking a
babbler seems like it isn't worth it.
Arkin and and Ph.D. student Justin Davis found that the deception works when the group reaches a certain
size — essentially, when enough backup arrives to convince the adversary
that it's best to back off. Davis modeled that behavior in software using a military scenario and found that it worked even if the group didn't have the firepower to confront the enemy directly.
The military is interested in this because a robot that can fool an
opponent is a valuable tool. It could lead an enemy down a false trail or make
itself look more dangerous than it actually is.
The work is an extension of similar
research Arkin started in 2009, developing a kind of 'ethical governor' for
robots. In 2010 he worked with Alan Wagner to develop deception
algorithms using a kind of hide-and-seek game.
If robots can fool other robots – or people – that does
raise interesting ethical problems. When does fooling people become dangerous?
How do you tell the robot when the right time to do that is? We won't be seeing
anything like the Terminator anytime soon, but we already have drones, and the