Tuesday, May 08, 2007

Simulating group selection

Wilson's approach to group selection is nicely illustrated by some of the early experiments in genetic programming. (Click here for all my posts about Evolution for Everyone, which I liked a lot.) For example, in an ant "food foraging" experiment, simulated ants were bred (through a simulated evolutionary process) to release simulated pheromones when they found simulated food. (See, for example, this paper by Liviu A. Panait and Sean Luke.) Other simulated ants follow the simulated pheromone trail to retrieve more of the simulated food. The simulated ant colony as a whole works as a cooperative unit to retrieve more simulated food (or at least to retrieve the simulated food faster) than would have been retrieved had there been no communication via simulated pheromones.

From an evolutionary perspective, one might ask why such pheromone releasing behavior began. Since the release of pheromones (like the dance of bees) is costly to the ant (or the bee), the argument must be that it evolved because those groups in which it occurred did better than those groups in which it didn't occur — and hence individuals in those groups had a better chance of surviving and reproducing notwithstanding the cost to themselves as individuals of releasing pheromones (or of dancing).

As far as I know, these sorts of questions have not been investigated with GP. In the GP experiments, the ants didn't pay a price for releasing pheromones. In fact the ants didn't pay a price for anything. The only question was how much food was brought home. Ant survival and individual ant reproduction weren't considered.

It would be interesting to see what happens if ants are charged for the generation of pheromones. Suppose that all food is shared equally at the nest, but that it requires energy to generate pheromones and to go out and forage for food. Would free rider ants evolve?

What if one had multiple colonies competing with each other? And what if ants were able to interfere with each other's access to food? And what if ants were marked as having released pheromones or as having brought food to the nest? Would a colony policing mechanism evolve that restricted access to the common food source for ants that didn't do their share of the work? Since this sort of evolution is a purely random process (especially in GP) the only way this would happen is if an ant's "program" included the operation to interfere with another ant if that other ant didn't "smell" like it had done its share of the work. But once that happened, would that colony be more productive? If so, that mutation would survive. It sounds plausible. It would be interesting to do the experiment. It would also be interesting to play with the parameters that made releasing food and going out and bringing food back to the nest costly to ants. How much should that cost? What are the break points at which it makes a difference? Lots of interesting experiments could be done.

No comments: