Evolve III: a discrete events model of an evolutionary ecosystem

Symbiosis in Digital Evolution: Past, Present, and Future

Symbiosis, the living together of unlike organisms as symbionts, is ubiquitous in the natural world. Symbioses occur within and across all scales of life, from microbial to macro-faunal systems. Further, the interactions between symbionts are multimodal in both strength and type, can span from parasitic to mutualistic within one partnership, and persist over generations. Studying the ecological and evolutionary dynamics of symbiosis in natural or laboratory systems poses a wide range of challenges, including the long time scales at which symbioses evolve de novo, the limited capacity to experimentally control symbiotic interactions, the weak resolution at which we can quantify interactions, and the idiosyncrasies of current model systems. These issues are especially challenging when seeking to understand the ecological effects and evolutionary pressures on and of a symbiosis, such as how a symbiosis may shift between parasitic and mutualistic modes and how that shift impacts the dynamics of the partner population. In digital evolution, populations of computational organisms compete, mutate, and evolve in a virtual environment. Digital evolution features perfect data tracking and allows for experimental manipulations that are impractical or impossible in natural systems. Furthermore, modern computational power allows experimenters to observe thousands of generations of evolution in minutes (as opposed to several months or years), which greatly expands the range of possible studies. As such, digital evolution is poised to become a keystone technique in our methodological repertoire for studying the ecological and evolutionary dynamics of symbioses. Here, we review how digital evolution has been used to study symbiosis, and we propose a series of open questions that digital evolution is well-positioned to answer.

Symbiotic interactions in the EVOLVE III ecosystem model

Parasitic feeding mechanisms were introduced into the EVOLVE III ecosystem model. The modified system exhibited a number of interesting behaviors, including obligative and facultative parasitism, transient facultative mutualism and nonsymbiotic feeding. Speciation and massive die-offs were also observed. The introduction of a symbiotic capability altered the co-evolutionary development of traits that exhibited an arms race when the symbiotic capability was not present. Stable mutualisms were not possible in the present version of the model due to the restriction that organisms were not allowed to feed both on other organisms and from the environment at the same time. The ecosystem biomass increased with increased symbiotic activity. Biomass was influenced by the balance of virulence and resistance traits in the population and decreased as these traits became more disparate.

The artificial worlds approach to emergent evolution

Artificial worlds models of evolutionary systems are computer models that map the essential logical structure of ecological systems, defined as self-sustaining biological organizations. The artificial world comprises an artificial environment, with mass components, energy input, and physical states. It also comprises artificial organisms, including a genome, a phenome, and a (developmental) map that connects the genome to the phenome. Mass components are cycled and space is limited. The evolution process results, as in nature, from genetic variation combined with natural selection imposed by the finiteness of the environment. The selection criteria (fitness values) are not imposed, but rather emerge from the interactions of the organisms with each other and with the environment. The dynamics at the population level also emerges from these basic interactions. In this paper we describe the comparative properties of the EVOLVE family of artificial worlds models.

Knowledge-based design of artificial worlds

We review the concepts of knowledge representation and modelling and simulation methodology which facilitate computer exploration of alternative artificial worlds, such as self-sufficient human habitats. An object-oriented computer environment which supports such simulation studies is briefly described. A simplified example of an artificial world model is given to demonstrate the power of the approach.

Simplicity and complexity in MIRROR universes

The scientific simplicity principle (OCCAM's razor) has always been strongly enforced by the available modelling tools. Moreover, the concept of simplicity itself is shaped by these (classical) tools. Computer models are less subject to simplicity constraints than other models are. It may be argued that complexity is the preeminent property for biological systems to study. In this paper we discuss our MIRROR modelling methodology in which (a concept of) simplicity is reconciled with biological complexity. Simplicity resides in the simple "TODO" ("do what there is to do") of the "individuals" (molecules, cells, organisms) which inhabit the model universe. The complexity appears in the multiple (levels of) individuals and the multiple levels of observable behavior of the universe. Examples are given of the development of complex, self-regulating social structures by simple interactions of individuals, and the adaptability of TODO based entities is compared to that of evolving entities. On the basis of these examples we sketch a slightly unconventional image of the evolution of complexity in biotic systems and discuss observations on the molecular record of biotic evolution which seem to fit this image.

Simulated evolution and artificial selection

A highly simplified evolving system was investigated by computer simulation. The genetic complement of each simulated organism in the population was represented by a single chromosome that consisted of a string of symbols. Individual fitness was measured as the number of symbols that corresponded to a specified rule. Reproduction was simulated with a non-breeding algorithm and two variants of a breeding algorithm, and was subject to random point mutations. In each generation, selection was effected by replacing the less fit members of the population with offspring of the more fit. The size of the population and the fraction replaced, though under experimental control, were constant for each simulation run. It was found that even such a simplified system is able to mimic a variety of properties observed in natural systems. In addition, the effect of the simulation parameters on the course of fitness increase provides a basis for using a genetic algorithm as an optimization technique.