COMPUTATIONAL MORPHOGENESIS

Evolutionary Algorithms

Natural Selection in Biology and on-line

The following idea of evolution by means of natural selection goes back Darwin and Wallace in its original form, who proposed it as the major mechanism for evolution of living things.

Three factors are required:

Heredity

Offspring are like the parents.

Variability

The offspring may be exactly like a parent, but may also differ. Sources of variability include mutation and recombination (crossover).

Differing Fitness

The likelihood and number of offspring of a given parent appearing in the next generation depend on the success of the parent. Success can be depend on adaptedness to the environment, interaction with other creatures, luck, and/or other factors. Reproductive success must depend (at least in part) on characteristics inherited from the parent(s).

In nature, finiteness of the environment and of resources are major factors in yielding differing fitness. (Both Darwin and Wallace came to this realization after reading the economist Thomas Malthus).

For natural selection, two additional properties are needed:

Self-Reproduction

Creation or construction of offspring is a function of individuals in their environment.

Instrinsic Fitness --- No External Fitness Function

Fitness -- determining reproductive success -- is not given as an external function, but emerges as a property of self-replication, and of the interaction among individuals and between an individual in its environments. That is, it arises as an effect on reproduction in terms of reproductive success.

Evolutionary methods have been applied to optimization, design and programming problems in the areas of Genetic Algorithms, Genetic Programming, Evolutionary Computation, Artificial Neural Networks, and other computationally-driven approaches. Self-reproduction is missing from all genetic algorithm applications of evolution, although some genetic algorithms pit individuals against one another in direct competition. Almost all current implementations of evolutionary algorithms on-line are instances of artificial selection, and even most of these use an external fitness function.

1. Many (computational) researchers would do not require self-reproduction as a criterion for natural selection. Indeed, requiring it is somewhat problematic since it can, in a trivial form, usually be incorporated in a trivial way. Yet non-trivial self-reproduction is present in all natural evolving systems.
2. I would like to thank to Dr. T.S. Ray for articulately clarifying for me the distinction between natural and artificial selection. Historically human beings have been applying artificial selection for roughly 10,000 years in agriculture and animal breeding. Artificial selection has only produced changes of degree in qualities already present in nature. It seems natural selection is necessary for more radical evolutionary transitions (e.g. changes in genetic systems, advant of eucaryotes, multicellular differentiation, segmentation, changes in body-plan, and even minor qualitative change).