Studying the evolvability of self-encoding genotype-phenotype maps

We introduce a model of reproduction in which the genotypephenotype (G-P) map is able to evolve. In this model, Each organism implements a G-P map, determining how the organism is encoded in its genome. Crucially, it also determines how the G-P map itself is encoded. We call these maps 'self-encoding'. We relate this model to recent artificial life research, and back to the seminal work of John von Neumann. We simulate populations of organisms that have as their genome and G-P map the axiom and production rules of an L-system. The populations are given the task of optimizing a dynamic fitness function. Our purpose is to study whether the self-encoding property has any effect on the evolution of evolvability, and to look for other factors that lead to the evolution of G-P maps that confer evolvability. We find that evolvability does evolve, but only when we add constraints to the model.