From Charles Darwin on, evolutionary biologists have struggled to explain self-sacrificing behavior. If evolution is all about the survival of the fittest, then why do animals from bees to people help others when it can hurt them or their chances to reproduce? Simulations of miniature robots that "evolve" helping behaviors have now provided a possible answer, confirming a 47-year-old theory that recently has come under attack: We help those who are most related to us because they are able to pass some of our genes to the next generation.
For all organisms, the ultimate goal is to pass on one's genes. The problem with altruism is that sacrificing individual gains for the greater good can compromise that goal. In the 1960s, biologist W. D. Hamilton pointed out that, actually, one could still pass on one's genes, or at least some of them, by helping a relative. According to his kin selection theory, the closer the relative, the greater this indirect benefit and, therefore, the more the helper should be willing to sacrifice in assisting that relative.
Studies of ants, wasps, bees, and termites, among others, have borne out this idea. But researchers could never quite verify the theory because they couldn't pin down exactly what the cost and benefits were or study them over the many generations needed to see evolution in action. And recently a few researchers have challenged the idea that relatedness is necessary for altruism's evolution, though they have their own critics.
Laurent Keller of the University of Lausanne in Switzerland wondered if he could resolve the debate using a computer simulation. He and roboticists Markus Waibel and Dario Floreano, both from the Ecole Polytechnique Fédérale of Lausanne, started with real-life robots that are just a couple of centimeters high. The robots have two independently operating wheels and a "nervous system" composed of sensors and a camera, which allow them to detect small discs—a stand in for food.
The researchers then created virtual representations of these robots on a computer so that they could observe the robots' evolution over time. In real life, random mutations build up over many generations, leading to adaptations that help organisms better survive in their environment. In the simulation, the researchers replicated this process by randomly varying the strengths of the various connections that made up the robots' nervous systems. Some of these "mutations" helped the robots better gather the food disks, while some made the robots less efficient at the task.
The simulations ran for hundreds of rounds, each time selecting the best food gatherers and culling the others. The process mimics natural selection, as only the most "fit" robots multiply (in clone form) in subsequent rounds. Every so often, the researchers checked their simulation results by programming a real robot with the same set of mutations. The virtual and real robots behaved similarly, says Floreano.
Once the team was comfortable with the virtual evolution environment it had set up, it added a new twist: It allowed the robots to share food disks with each other. If Hamilton's hypothesis was correct, "successful" virtual robots were likely to be those that were closely related and shared food with each other; that would help to ensure that at least one of them -- and some of the genes of both—would make it to the next round. (Two robots with a modest amount of food disks would both be more likely to be cut from the simulation, but if one robot gave all of its food to a second robot, that second robot would likely make the next round.) And indeed, altruism quickly evolved in the simulation, with greater food-sharing in groups where robots were more related, the researchers report online today in PLoS Biology. The more closely related the robots, the quicker they cooperated. "It shows how general the [theory] is, whether you are an insect, a human or a robot," says Floreano.
"This is a very original approach," says evolutionary biologist Jacobus Boomsma of the University of Copenhagen in Denmark. "The results are remarkably clear cut, given all the messy dynamics that might have appeared ... and show convincingly how tremendously robust Hamilton's rule is."
But Harvard University theortician Martin Nowak is more cautious about drawing conclusions based on computer simulations. Virtual robots are not a stand in for real life, he says. "[The work] tells us nothing about whether Hamilton's rule makes a correct prediction for actual biological systems," he says.