The Design of Adaptive Systems: Optimal Parameters for Variation and Selection in Learning and Development

Some aspects of learning and development are based on evolutionary change within the organism. In trial and error learning, variant ideas or behaviours are generated and selective filters (learning rules) choose among the population of variants. Development may, in some cases, proceed by selection within a population of variant cellular lineages. This paper analyses abstract properties of selective systems to understand the evolutionary dynamics that occur within organisms. The Price Equation and Fisherʹs fundamental theorem of natural selection, two of the most powerful concepts in evolutionary genetics are applied in a general way to internal selective systems in learning and development. This analysis emphasizes generative mechanisms and selective filters as genetically controlled phenotypes of individual organisms. Generative mechanisms create the variation on which selection acts. Selective filters determine the extent to which selection within the organisms optimizes organismal performance. The methods of Price and Fisher provide a general way in which to partition evolutionary change into improvements caused by selection and the tendency of high performance variants to deteriorate because of competition or environmental change. This balance between selective improvement, at a rate equal to the variance in fitness, and a matching deterioration in performance, provides general insight into the common properties of adaptive systems in genetics, learning and development. These ideas are applied to a model of honey bee foraging. This example clarifies the relation between genes and phenotypes controlled by internal selective systems.