Origins

by Elizabeth Pennisi

In my essay, On the Origin of Cooperation, I describe experiments in which people are asked to play computer games that help reveal our cooperative tendencies, and I discuss other studies involving the use of microbes to get at the basic principles of working together. But Laurent Keller has gone a step further to work out details of social interactions. An evolutionary biologist at the University of Lausanne, Switzerland, Keller and his colleagues use evolving robots in experiments looking at the evolution of communication, an essential and complex ingredient of cooperation. The robots help him address questions that cannot be addressed through his studies of ants.

The 15-centimeter-diameter robots (see left) can move in all four directions, like tanks, and are rimmed with LEDs that can emit blue light either randomly or under control of a neural network, depending on the experiment. Each has an omnidirectional camera that sees red and blue light, as well as downward-facing infrared sensors that can distinguish gray from black. The robots are tasked with finding “food,” which emits red light. “Poison” also emits red light and can be distinguished only from up close, when the infrared sensors are able to detect a black paper circle under the poison—the food sits on a gray circle. As the experiments begin, robots flash blue when they find food, a signal others can tap into to locate the food as well

The robots have “brains” and “genomes.” The brains are the software running on onboard computers, with neural networks consisting of 10 “sensory neurons" that convey what the camera and sensors perceive to “activation neurons" that make the motors turn and that light up the LEDs. There are 33 connections in all, with different strengths, each determined by a “gene.”

The researchers tested 100 colonies, each with 10 robots, in a 300-square-centimeter arena with food at one end and poison at the other. (See movie.) The robots got points for detecting and staying by food or lost points for targeting poison. Not everyone could fit by the food, so there was some jostling for position.

Using that system, the researchers scored each robot’s fitness and took the average of the colony's members to assess the colony’s fitness. Depending on the specific test, the researchers selected a subset of the best robots or the best colonies, tweaked their "genes," and ran the experiments again—up to 500 generations' worth. In some cases, they created groups of “related” robots—all with the same “brains”—to assess the affect of kinship on the social dynamics. They wanted to know what behaviors evolved when competition for food was between relatives versus between nonrelatives.

Over time, robots evolved when and where they emitted blue light, and they came to associate high-blue-light areas with food. Foraging efficiency increased, particularly in the highly related robots chosen from the best colonies, Keller and his colleagues reported in 2007. “Reliable communication evolves either when individuals in a group are highly related or when there is increased competition between groups, and so reduced competition between individuals,” explains Keller’s collaborator Sara Mitri.

In a second series of experiments, the researchers focused on “unrelated” robots selected for their individual fitness. Cooperation theory suggests that unrelated individuals have less motivation to share information and thus “one would expect unreliable communication to evolve,” Mitri explains. After 52 generations, the robots were much less likely to emit blue light by food and instead tended to light up around the poison, Keller and colleagues reported 15 September in the Proceedings of the National Academy of Sciences.

But the researchers also discovered that some robots continued to light up at the food source, even after 500 generations. It seems that once the meaning of the blue light became ambiguous, few enough robots were attracted that competition for food diminished to the point that there was little selection against having the blue light on at the food source, they reported. The net result: individual variation in light production and responses, says Keller. That parallels what is seen in animals: variable communication strategies.

The robots are allowing us to address questions that cannot be answered with real organisms, says Keller. With respect to the evolution of communication, “there are no fossils allowing us to study how it evolved, and it is not really amenable to experimental evolution.”

Photo credit: Sara Mitri and Walter Karlen

Movie credit: Stéphane Magnenat, Matthieu Bontemps, and Kevin Frugier.