May 2011

Swarms of tiny robots have given up their selfish ways and started sharing resources for the greater good. Though this might sound like the plot of a bad summer blockbuster, it is real news. This month, a team of Swiss researchers announced that they’ve used robots to simulate biological evolution. The simple, mobile robots — each a little larger than a sugar cube — began their lives directionless, meandering aimlessly into walls. But after a few generations of natural selection, their computer programs evolved so that they became efficient foragers, purposefully collecting disks that represent food. None of that is particularly surprising. Scientists have long been able to simulate evolution through computer programs that mimic the processes of genetic inheritance, mutation, recombination, and reproduction. What is noteworthy is that many of these robots eventually evolved to help one another, sacrificing personal success to aid other robots in their group. Such altruistic behaviors — which we observe in all sorts of organisms, from humans to slime molds — are a topic of debate among evolutionary biologists, and these big-hearted robots are allowing scientists to test ideas about how such behaviors evolve.

Where's the evolution?

What exactly is the conundrum regarding altruism? In evolutionary terms, altruistic behaviors are those that decrease the fitness of the individual performing the behavior (e.g., a meerkat that sounds the alarm about an approaching jackal, putting herself at a higher risk of attack) and increase the fitness of one or more other individuals (e.g., other meerkats in the colony that are able to escape the jackal attack). We’ve observed many examples of such selfless behavior across the natural world. The question for evolutionary biologists is, how do altruistic behaviors evolve to become common in populations? After all, natural selection decreases the frequency of gene versions that hinder their carrier’s ability to survive and reproduce. Any meerkat that happens to have a gene that causes it to risk its own life for the life of its colony mates should be less likely to survive to reproduce and less likely to pass that altruistic gene on to the next generation. Based on this line of reasoning, natural selection should drive altruistic behaviors out of populations! So what’s going on?

Biologists have considered many different hypotheses to explain the evolution of altruism. One of the most important is based on kinship. The basic idea is that if self-sacrificing behaviors help the altruist’s relatives survive and reproduce, altruistic genes can spread in a population because they are likely to be passed on to future generations through the relatives, although not through the altruist him or herself. To understand this hypothesis, consider the case of the meerkats described above. Imagine that the meerkat sounding the warning (let’s call her Flower) does so because she carries an “altruistic gene.” What happens to the gene if Flower dies but saves the lives of several other individuals in her colony? Well, if Flower is the only one with the altruistic gene, it dies with her. However, if the individuals she saves are her relatives — e.g., her siblings, nieces, nephews, and children — odds are that many of them also carry the altruistic gene and will be able to pass it on to their offspring because of Flower’s good deed. Altruism can spread in the meerkat population in this way. This is one example of kin selection — a form of natural selection that is caused by interactions among related individuals.

Some popular articles describe this hypothesis about the evolution of altruism as one animal helping a relative in order to pass on their shared genes; however, this is a bit misleading. Natural selection has no foresight and does not require any intent on the part of the organism undergoing evolution. Organisms that display altruistic behaviors are not necessarily “trying” to pass on genes through their relatives; they may be unaware of the effect of their actions or — for example, in the case of altruistic slime molds — may not have any sort of consciousness at all!

The hypothesis simply suggests that altruistic behaviors that happen to have the effect of promoting the fitness of relatives can spread — and specifically, that this is more likely to happen the more closely related the altruist and the recipient are, the less the altruist has to sacrifice to help the other, and the bigger the benefit the recipient gets. This makes intuitive sense. Closer relatives are likely to share more of their genes. So Flower’s altruistic gene is more likely to spread if it ends up helping her sibling (who shares half of her DNA) than if it ends up helping her cousin (who shares just 1/8 of her DNA). Similarly, if Flower’s altruistic behavior is very unlikely to get her killed, or is very likely to save the lives of her family, the altruistic gene is more likely to spread. The detailed form of this hypothesis (which gives numerical values to the relatedness of individuals and the benefit and cost of the altruistic act) is known as Hamilton’s rule, after evolutionary biologist W.D. Hamilton who proposed the idea in 1964.

The robots in the recently published study were used to test Hamilton’s rule. Each robot was controlled by a simple computer program (which you can think of as its DNA) with 33 different parts (i.e., with 33 “genes”) that influenced different aspects of the robot’s behavior — its movement, sensing ability, and how likely it is to share disks representing food with others (i.e., its altruism gene). Groups of robots were placed in arenas with food disks to collect. At the beginning of the experiment, each robot had random settings for its computer program. Some robots ran into walls, some stuttered in place, and some, just by chance, happened to move in a way that pushed a food disk. After allowing the robots to show their stuff, the researchers evaluated how well each robot did at obtaining food — a measure of evolutionary fitness. They formed a new generation of robots from the computer programs of the old generation (simulating recombination and reproduction), but with a twist: the better a robot was at obtaining food, the more likely it was that elements of its computer program would make it into the next generation. Each generation, the researchers also “mutated” the robot genomes, introducing small random changes into their computer programs. The researchers repeated this experiment over and over, simulating the robots’ performance — and in each trial, over 500 generations of natural selection, the robots rapidly evolved efficient foraging abilities.

In the most important part of their study, the researchers tested what would happen if the robots were more or less related to one another — i.e., if some of the robots in a group started off with computer programs that were genetically identical to one another. In these cases, the robots evolved exactly as we would expect them to based on Hamilton’s rule. The more closely related the robot group was (i.e., the more clones it contained), the more likely the whole group was to evolve altruistic behavior and ultimately wind up sharing all their food disks. And the less an individual robot lost by sharing a food disc and the more other robots benefited from shared food discs, the more likely altruism was to evolve. Hamilton’s rule did seem to hold in this simulation, which closely mimicked many aspects of a real, biological population.

This result is important because, in nature, it would be very difficult to perform similar experiments and impossible to perform them over and over again. So, while many different lines of evidence support the idea of Hamilton’s rule, this is the first time that biologists have been able to study it so exhaustively and with such accuracy — albeit in robots instead of actual living organisms. Nature is not always red in tooth and claw; it is often generous, kind, and unselfish — and these results help us understand why such traits have evolved over and over again.