Synthetic Neutrality for Artificial Evolution

Recent works in both evolution theory and molecular biology have brought up the part played by selective neutrality in evolution dynamics. Contrary to the classical metaphor of fitness landscapes, the dynamics are not viewed as a climb towards optimal solutions but rather as explorations of networks of equivalent selective genotypes followed by jumps towards fitter networks. Although the benefit of neutrality is well known, it is hardly exploited in the genetic algorithm (GA) field. The only works about this subject deal with the influence of the inherent neutrality of a fitness landscape for evolution dynamics. In this paper, we propose a very different approach which consists to introduce “handmade” neutrality into the fitness landscape. Without any hypothesis about the inherent neutrality, we show that a GA is able to exploit new paths through the fitness landscape owing to the synthetic neutrality.