Interactions between learning and evolution: the outstanding strategy generated by the Baldwin effect.

The Baldwin effect is known as an possible interaction between learning and evolution, where individual lifetime learning can influence the course of evolution without using any Lamarckian mechanism. Our concern is to consider the Baldwin effect in dynamic environments, especially when there is no explicit optimal solution through generations and this solution depends only on interactions among agents. We adopted the iterated Prisoner's Dilemma as a dynamic environment, introduced phenotypic plasticity into its strategies, and conducted computational experiments, in which phenotypic plasticity is allowed to evolve. The Baldwin effect was observed in the experiments as follows: First, strategies with enough plasticity spread, which caused a shift from defect-oriented populations to cooperative populations. Second, these strategies were replaced by a strategy with a modest amount of plasticity generated by interactions between learning and evolution. By making three kinds of analysis, we have shown that this strategy provides outstanding performance in comparison with other deterministic strategies. Further experiments towards open-ended evolution have also been conducted so as to generalize our results.