Reliability and novelty: Information gain in multi-level selection systems

Information about the environment is captured in human biological systems on a variety of interacting levels - in distributions of genes, linguistic particulars, concepts, methods, theories, preferences, and overt behaviors. I investigate some of the basic principles which govern such a hierarchy by constructing a comparatively simple threelevel selection model of bee foraging preferences and behaviors. The information-theoretic notion of "mutual information" is employed as a measure of efficiency in tracking a changing environment, and its appropriateness in epistemological applications is discussed at some length. In particular, information accumulated in mid-level preference distributions exhibits suggestive properties for the purposes of naturalistic epistemology. It also appears that the novelty of scientific objects and representations and the rapidity of scientific change relative to genetic change need present no obstacle to the use of such models in explanations of scientific progress and the reliability of scientific judgement. © 1997 Kluwer Academic Publishers.