Biased Group Selection

Group-level selection, although controversial in biology, has been shown to be very effective for obtaining cooperative behavior in artificial systems. A key issue is evolving cooperative task completion is group composition. Prior studies have addressed two ends of a spectrum with homogeneous groups on one extreme and heterogeneous groups on the other. In this paper we explore the space in between by using biased group selection. Under this selection model, subpopulations compete with respect to a cooperative task, while an extrenal bias may favor the genes of specific individuals, depending on whether they actually participated in the task or not. We evaluate this approach on a cooperative predation task in digital organisms, where feasible solutions can be carried out by homogeneous or heterogeneous group configurations. Our results show that, consistent with earlier studies, homogeneous teams tend to find better overall solutions than their heterogeneous counterparts. Howerver, we also observed that populations comprising groups with some degree of heterogeneity found solutions more frequently than in the homogeneous case. Effectively, while evolution pushed heterogeneous groups toward functional homogeneity for this particular task, heterogeneity with a selection bias proved more effecive at exploring the search space.


The source code for the main branch of Avida, along with Avida documentation, can be found at

The specific source code for this project can be obtained here.

Finally, the specific set of configuration files used for each of our treatments in this project can be obtained here.