GENETIC ALGORITHMS THEORY
Breeder Versus Environment
A GA has two parts: the breeder and the environment. The environment decides which organisms reproduce and what their relative fitnesses are. The breeder uses the fitness information to select organisms and mate them.
The environment can be either simulated or real. You use simulated environments to test how good the breeder is. You then place the breeder in a real environment that you would like to learn about.
To successfully move your breeder from a simulated to a real environment, you must maintain a solid abstraction barrier between the breeder and the simulated environment. A breeder should only have access to the organisms' fitnesses and genes.
The fundamental process of genetic evolution is mutation. It alone would eventually produce a perfect organism. In real life, it took billions of years just for a mutation breeder to evolve. Earth first had to evolve:
The list goes on. In reality, the genetic instructions for making these mechanisms has to be encoded in the organism's genes. But we can instead model the mechanisms much more simply with a breeder.
- a constant set of alleles (CGAT) to encode our genetics;
- a complex system of RNA, enzymes, and proteins which could translate genes into an actual body;
- a mechanism which could replicate DNA with mutation.
There are many other mechanisms:
- maturation: organisms must mature before they can reproduce. This assures the organism is fit enough to live for a certain time.
- crossover: any two living organisms exchange their genetics to produce a child.
- fitness-proportional: organisms with better fitness produce more children. In real life, this requires organisms to be smart enough to discern fit mates from unfit mates.
- normalized fitness: organisms with better fitness relative to the rest of the population produce more children. This simulates a scarce-resource environment to maintain selective pressure.
- species-only: organisms only mate with sufficiently similar organisms. In real life, this is accomplished with looks, pheromones, matching genetalia, and so on; but your breeder can compute similarity by looking at the organisms' genes.
- no-incest: organisms only mate with sufficiently dissimilar organisms.
- market-niches: each gene-space maximum has a fixed amount of fitness it can give to organisms. This produces a market, which, when saturated, will no longer appeal to organisms.
- groupism: organisms can adopt a set of groups they belong to and mate with. Hierarchies of groups should form naturally.
- synergy: organisms give part of their fitness to the groups they belong to. A larger or smaller fitness is then returned, depending on the normalized average fitness of the groups.
- enemies: one group competes for fitness resources with another group. They maintain selective pressure on each other. This should form naturally with groupism and market-niches; however, it is only useful with meta niches (too much to explain here).
March 1, 1999