User Tools

Site Tools


Evolution as Algorithm

The idea of evolution by natural selection was originally developed by Charles Darwin to explain the change in form and function of plants and animals across generations, but evolution is a more general concept, something that happens whenever the conditions are right:

  • Agents can reproduce, with offspring fairly accurately inheriting traits from their parent(s).
  • Variation is introduced somehow, either by mutation or by inheritance of a mixture of parent traits.
  • Environmental limits prevent all of the offspring from successfully reproducing.
  • An agent's inherited form and function affects its ability to survive and reproduce.

Before Darwin, evolution only meant some sort of change over time, often a trend with an implied direction or even a purpose (see Orthogenesis.)

Evolution is an iterative interaction between organism and environment. The result of evolution is Adaptation – that organisms become better fitted to their place in the environment (their niche). Because the evolutionary system evaluates organisms by their ability to survive and reproduce, this increase in fitness is nothing other than a relative increase in successful reproduction.

In this abstract view, evolution is an Optimization algorithm, where the evaluation function is fitness. In mathematical global optimization we are only interested in peaks higher than the one we are currently on. In the evolutionary landscape the relative height of peaks is less clearly defined because escaping into a new part of design space is often associated with exploiting a new evolutionary niche.

In one view, you can say that a new niche has a very high fitness value for the pioneer organisms because they have no competition. The problem with this view is that they reproduce exponentially, rapidly filling the niche and restoring individual reproductive success to more normal levels. Fitness is then seen as declining (due to population pressure) even as the organism continues to optimize its design to better exploit the niche.

Of course, fitness does depend on the environment, but so far as understanding the evolutionary virtue of radiating into new niches it may make more sense to say that the relative fitness of organisms that do not compete as ill-defined. So the power of the evolutionary algorithm is defined by its ability to exploit new niches, without any emphasis on finding the highest peak in all Design Space.

See also Artificial life.

You could leave a comment if you were logged in.
analysis/concept/evolution_as_algorithm.txt · Last modified: 2015/04/12 16:54 by ram