Genetic Algorithms

According to the preceding road map the genetic algorithms (GAs) will be the first formal devices which we will investigate on our way to engineer intelligent evolutionary semiotic systems. We will distinguish here several sub cases.

In a first step we will give a formal account of the main properties of the evolutionary mechanism as it is presented by nature. In this formal account we already do abstract away from the biochemical details (although these are the ingenious matter out of which everything emerged) and focus on the general principles independent of the used material. This is called the 'evolutionary model'.

In a large part of the computer science literature this formal model of the evolutionary mechanism is reduced further by cutting off the 'phenotype' and concentrating only on the 'genotype'. This is the version known as 'genetic algorithm'. This reduction is very elegant and highly effective. But it has it's prize: all the knowledge which is encoded in the phenotype and its interaction with the environment has now to be attached to the fitness function, which makes the fitness function like a 'god' who knows everything. But because the fitness function as such is completely void of any higher knowledge the GA-version of the evolutionary model has to assume that the 'authors' of the fitness function know 'everything which is needed'. Thus this simplification presupposes a lot of 'covered' knowledge which makes this kind of formalization appearing a bit like 'magic', making modeling less rational.

If one does not strip away the phenotype then one works with the genotype as a primary 'blueprint', which can be changed, and a 'phenotype' built from the genotype as that unit which does the 'real work' through real interactions with the environment. In this variant of an evolutionary model no explicit fitness function is needed. The phenotype with it's interactions realizes implicitly a fitness function. This version - often called 'genetic programming' (GP) - has no magic and therefore is less irrational.

Gerd Doeben-Henisch 2013-01-14