Representations for Artificial Organisms

we are interested in simulations of biological evolution, i.e. simulations of populations of organisms over many generations living in a complex and dynamic environment. Our simulations are microanalytic, meaning that each individual organism and gene is separately represented, and the biologically significant events in an organism's life are all separately simulated in detail. Although we have been successful with simple models, we have encountered fundamental difficulties when scaling up the complexity of the organisms and the complexity of behaviors we expect of them. These difficulties all lead to a single question: What is an appropriate representation for an organism, i.e. what is an appropriate programming paradigm in which to express the complex behavior of organisms, and how should such programs be encoded into strings so that genetic algorithms will be successful over them? The project that brought these issues to the surface is a complex evolutionary simulation called AntFarm, in which we are attempting to evolve cooperative foraging behavior in a population of colonies of artificial “ants.” In this paper we survey a number of candidate representat. ions for organisms, that we have considered for AntFarm, all of which have been used in the past for simple evolution models. We show that none of the representations are well-suited for AntFarm. From their inadequacies we abstract a number of principles that we believe are necessary tor successful evolution of complex artificial life. Finding a representation that has all of the propert.ies we identify is still an open problem.