On the evolution of learning: Representation of a stochastic environment (Q1909283)

Learning is frequently proposed as a mechanism that enables an individual to adapt to changes in its environment. These studies generally take as their point of departure the ``absolute fixity argument.'' That is, in an absolutely fixed environment, a genetically fixed pattern of behavior should evolve (assuming some cost is associated with learning). On the other hand, in an absolutely unpredictable environment, where the past and present states of the environment offer no information about the future, there is nothing to learn and, assuming some cost to computation, there is again no driving force for a learning capability to evolve. We shall construct a population genetic model in which different genotypes have different capabilities to form representations of a randomly changing environment. The stochastic properties of the environment will first be outlined. Then a haploid genetic model with two alleles, one of which allows a limited representation capability, will be described. This is extended to a two-locus framework. The analysis concludes with a generalization to a diploid one-locus model with different genotypes conferring on their carriers different abilities to represent the environment. We discuss the results of a stochastic simulation study of this model and compare these with the local stability analysis of the mathematical model. In particular, we shall see that conditions favoring initial increase of learning may not guarantee its eventual fixation in the population. The ability to represent an environment requires internal structures within the individuals that possess this capability, and some metabolic and energetic cost is required to maintain such structure. We shall regard this as the cost of learning. Our objective is to understand how the learning characteristics of the different genotypes correspond to fluctuations in the environment. We document differences between haploid and diploid genetics as well as the effect of assumptions about the cost of learning.