Modeling the effects of interaction between learning and evolution
Scientific Research Institute for System Analysis, RAS
Nakhimovskiy pr., 36-1, Moscow, 117218 Russia
*e-mail: vgredko@gmail.com
A model of interaction between learning and evolution is constructed and investigated. The modeling of the main effects of the interaction between learning and evolution is carried out. These effects are as follows: 1) Genetic assimilation of skills, which are acquired via individual learning, during many generations of the evolutionary process. Due to genetic assimilation, individually acquired skills become inherited; 2) The hiding effect, which means that strong learning can inhibit the evolutionary search for the optimal genotype, as this learning increases the chances of finding a good phenotype independently on the genotype of the model organism; 3) The effect of the learning load, which means that the large change in phenotype in the learning process can decrease the evolutionary fitness of an organism. The learning load can accelerate evolutionary optimization.
Using computer simulations, the functioning of all three effects is demonstrated. The evolution of the population of model organisms is considered, each organism has a genotype and phenotype, which are long chains of symbols. Each generation of evolution lasts a certain number of time steps. The genotype of an organism determines its phenotype at the beginning of a generation. During a generation, phenotypes of organisms are optimized through learning. The genotypes of organisms throughout the generation remain unchanged. At the transition to the next generation, organisms are selected into a new population with probabilities that are proportional to their fitness. The fitness of organisms is determined by the phenotypes of organisms at the end of the generation. Genotypes of parental organisms modified by mutations are inherited by descendant organisms. In our earlier studies, we analyzed the effects of interaction between learning and evolution for a simplified version of the model, when there is only one maximum of organisms' fitness. In the present work, these effects are investigated for a more complex variant with a large number of fitness maxima.
