An interesting find by Science Blogs. Basically, a bunch of robots were programmed to work together in finding a resource and avoiding a “poison”. They would wander around looking for the resource, and when they found it, they were initially programmed to turn on their light to indicate to other robots that the resource was over there. Then they’d gather around it.
The programming was a little bit variable, though, and would change as time went on. Every so often the most successful resource-getting poison-avoiding robots were “mated”, making that behaviour selected for. While the robots got more successful by working together, they began to get crowded out of the resource when they let other robots know where it was.
So 60% of them learned to lie. They’d leave their light off when they found the resource, hoarding it for themselves. That was the randomly generated behaviour that was selected for.
It would be interesting, however, to see the same experiment done on a larger scale, with competing groups of robots, rather than competing individuals, where the groups were rated collectively and selected for on that basis. I suspect cooperative behaviour would trump the liars.