| Previous | Table of Contents | Next |
Steve Wand
sewand@amp.com
ABSTRACT
Each species that exists on this planet is the product of millennia of natural selection. Competition for finite resources has produced varied species, many of which exhibit specialized behavior that allows them to survive. Genetic algorithms acting upon a randomly chosen population, and competing for finite resources should produce a near-maximal biomass, with several distinct species exploiting different levels of the biosystem. Utilizing reproduction, crossover, mutation, and niching operators, the coding scheme could preserve diversity in predator/prey populations and mass, while maximizing biomass and sensory performance within the population, particularly in a static environment. This chapter describes an investigation in which a genetic algorithm is used to simulate an artificial environment in which various species compete with one another.
INTRODUCTION
Darwin’s theory of evolution concludes that natural selection is the key factor in the origin of species. Within species, an individual that reproduces passes on its genetic characteristics. Individuals that possess ‘favorable’ traits are more likely to survive, hence future generations increasingly exhibit ‘favorable’ traits. Given time, a population’s characteristics can diverge significantly from their original makeup. Examining the machinery of natural selection can lead to a keener appreciation of complex interactions that shape life. Because genetic algorithms are based upon the mechanisms of reproduction, they provide a clear analogy to how real populations can evolve over successive generations.
Field observations yield glimpses of natural selection’s capacity to produce populations that fully exploit their environment. In the real world, this process takes millennia. A simulated ecosystem, with a diverse initial population, offers a means to view the effects of evolution over hundreds or thousands of generations. The recombination of individuals via a genetic algorithm provide an elegant means of rewarding variations that maximize their environment.
This chapter will examine a simulated ecosystem of herbivores and carnivores. Each individual will have several characteristics that shall determine the relative success or failure of each organism within the environment. The GA operators of reproduction, crossover, mutation, and niching will operate on a multi-parameter coding. Organisms that can successfully adapt to their environment will be favored within the reproductive pool. ‘Winning’ populations will have the greatest increase in mass (numbers * size), with the population existing near the ideal carrying capacity of the environment.
LITERATURE REVIEW
There have been numerous works published on simulated organisms and ecosystems. Mechanisms of cell chemistry were examined by Rosenberg [1], which simulated enzyme reactions using genetic algorithm-like operators. The Avida simulated ecosystem shows support for the punctuated equilibrium view of evolution, as opposed to a more Darwinian gradual model of evolution [2].
An artificial life program called Tierra is used to model both small and large scale ecosystems. Tierra utilizes genetic algorithms to simulate evolutionary change[3]. The Tierra system creates a diverse population of organisms, but does not optimize resources by the population as a whole. To examine this problem, it is useful to look at models of population interaction. Two primary engines of ecological change are predation and competition. Ten components of functional response to prey and predation [4] are:
Each factor has sub-factors. For instance, successful search involves sensory facility, reaction distance, speed of predator, speed of prey, and capture success. Relationships are drawn between density of predators, density of resources, probability of attack, time spent in attack, expected gain, and number of attacks to derive a success ratio of predation. Smith [5] draws the conclusion that species that spend most of their time searching for food that takes little effort to capture, will be generalists (e.g., hyenas), while species that have abundant prey that takes much effort to capture will be specialists (e.g., cheetahs). Specialization leads to speciation. Sub-populations which converge at multiple ‘peaks’ along the spectrum of the initial population will eventually stop sharing genetic information with other sub-populations.
| Previous | Table of Contents | Next |