Using Robots To Understand The Genetics Of Selflessness
Swiss scientists searching for the "altruism gene" use robots to show why creatures, human and otherwise, sacrifice themselves for the sake of others.

LAUSANNE - "I would give my life for two brothers or eight cousins' is how a British evolutionist once described the kin selection theory. In nature, examples abound of social animals willing to sacrifice themselves for the wider good of their group, provided that the genetic closeness (how many genes they share) between them is high enough. By contributing to the survival of their genetic relatives, these altruistic individuals will propogate, passing on their genes within the species.
Still, quantitative tests on this theory have been difficult to perform in nature. But thanks to the use of robots, a team of Swiss scientists has been able to overcome the problem, in a study whose conclusions have been published in the PLoS Biology magazine this week.
Why do meerkats accept to go up a rock and keep an eye on potential predators? Not only does this kind of altruistic action prevent them from looking for food, but their conspicuous position also puts their life at risk. Worker ants are even more generous: incapable of reproducing, they toil for the benefit of their sisters, one of whom will someday become the queen. In the 1960s, biologist William Donald Hamilton formulated his famous rule of kin selection, which states that this kind of altruistic behavior can only occur if the degree of genetic relatedness is superior to the cost-to-benefit ratio for the altruist and the beneficiary of the action.
"We can speak of altruism when an action has a cost for the individual performing the action," says Laurent Keller, co-author of the study and member of the Department of ecology and evolution at the University of Lausanne in Switzerland.
In nature however, it is very hard to quantify the advantages and disadvantages of a certain action, which means that the phenomenon cannot be accurately measured. "There are many parameters that influence the way genes are passed on from one generation to another," says Michel Milinkovitch from the Department of genetics and evolution at the University of Geneva. "Scientists have a very hard time trying to pin down a single one of these parameters. In a colony of ants, climatic variations can completely change the equation."
Robots, on the other hand, can be controlled much more easily. Equipped with several infrared distance sensors, the robots move around with the help of two motorized wheels, explains Dario Floreano, co-author of the study and director of the Laboratory of Intelligent Systems at the School of Engineering of Lausanne. The robots were placed into groups of eight inside a square arena, and were asked to push as many disks as possible – or food items – towards the wall. At the end of the exercise, they could either keep their gains (selfish behavior) or share it with others (altruistic behavior).
The robots have a network of 13 neurons, and an artificial DNA comprised of 33 genes connecting the neurons. Exterior signals received through the sensors are amplified from one neuron to the next on a scale ranging from 1 to -1, as determined by the genes. "The genome influences the way the information is processed," Laurent Keller says. "Different connection weights may result in very different behaviors," Floreano adds.
In the first generation of robots, the values were randomly set and the robots – which were divided in 200 independently evolving groups – behaved completely arbitrarily. At the end of each test, only the best performers were kept by the researchers. Their genetic codes were then recombined – as in the case of sexual reproduction – introducing a basic evolutionary variation mechanism of random mutations. The researchers conducted 500 generations of selection: the robots quickly became more efficient and learned to work as a group without any human interference.
The Swiss team also tested different parameters by playing with the cost-to-benefit ratio, and the level of relatedness between individuals. In groups formed by perfect clones – robots shared the same genetic code – the relatedness was equal to 1. For groups of four clones of the same model and four of a different kind, that number falls to 0.5, just like in the case of brothers. This allowed the scientists to put Hamilton's rule to the test.
Somewhere between computer simulations and in vivo observations, the robot tests are another step towards the real world, says Michel Milinkovitch. "It allows us to show that even simple systems can provide the necessary conditions for the evolution of altruism."
Laurent Keller says that the study has also proved the validity of the theory in the case of complex genetic architecture systems. One of the criticisms often raised against the Hamilton rule was that no one had been able to identify the altruism gene. "Our study shows that this feature is not actually related to a single gene but can emerge from a system," says the Swiss biologist.
Read the original article in French
photo - Ed Yourdon