Subscribe
 
 

It's a Bot-Eat-Bot World

22 February 2007 (All day)
Comments

(top) Walter Karlen; (inset) Sara Mitri

Talkin' 'bout an evolution.
Armed with a virtual "genome" and a simple brain, the s-bot (right) evolves to talk to other robots with light.

Alliances, deceptions, and even some shoving: It could be reality television, or it could be insect expert Laurent Keller's lab at the University of Lausanne in Switzerland. Keller and his interdisciplinary team of researchers have condensed thousands of years of evolution into a weeklong battle of the bots that demonstrates for the first time how social creatures evolve to communicate--and how, in a pinch, they evolve to deceive as well.

Experts disagree over exactly when and how communication arose among social animals. Evolutionary biologists suspect that early communication may have developed as a way for closely related individuals to boost each other's chances for survival. Studying such evolution in the lab is practically impossible, however, because most socially sophisticated creatures, such as bees or monkeys, can take hundreds of generations to show substantial behavioral changes.

Enter the s-bots, robots fated to live, reproduce, and die within 2 minutes. Keller and company equipped these 15-centimeter-tall subjects with wheels, a camera, a ground sensor, and a virtual "genome"--a computer program that dictated their responses to their environment. Some of the robots also had blue lights they could turn on or off. The robots then entered a foraging environment consisting of a "food" source and a "poison" source. Robots that found food were "mated" with other successful robots: Their genomes were recombined into new programming for the next generation. Robots that didn't find food, or that found poison, saw their genomes vanish from the game.

In one set of experiments, robots entered the game as part of a larger colony. When most members of the colony found food, individuals from the entire group stood a good chance of having their genome make it to the next generation. In another set of experiments, it was every bot for itself.

During the course of 500 generations, or about a week, the robots evolved to use their blue lights to communicate. Some groups flashed them to tell others where the food was; other groups used them to warn of the presence of poison. As the tactic worked and the genomes of successful communicators survived, the robots became more and more efficient at foraging.

The researchers expected the lone bots to largely ignore each other. But they were surprised, says Sara Mitri, a graduate student involved in the experiment. Bots acting alone developed the same communication strategies, along with some strategies of deception. When surrounded by their kin, the incentive of trying to get their genome--or one similar to theirs--into the next round of the game kept the cooperation going. But when surrounded by "stranger" bots with dissimilar genomes, they flashed their blue lights far from food to sabotage the nonkin bots' chances for survival. "We did not expect that they would evolve such a sophisticated system of communication," says Keller. He says the results--presented online today in Current Biology--confirm that kinship and pressure to succeed as a group help give rise to social behavior, even the unsavory kind.

"I think this is really, really stunning," says Lee Dugatkin, an evolutionary biologist at the University of Louisville in Kentucky. Using robots to understand the evolution of communication opens the door to testing more complicated aspects of social behavior, such as reciprocity, he adds. "It has tremendous potential ... to address all sorts of questions that haven't been answered yet."

Related sites

Posted In: