Building a Learning Bias from Perceived Dependencies

Abstract

This chapter explores the way in which perceived dependencies can provide a solution to problems raised by a rigid bias. Perceived dependencies are dependencies that, although not always provable, are satisfiable in a system's observed environment. The relation of perceived dependency possesses the formal properties needed to devise a learning bias. The notion of perceived dependency is not the solution to every problem related to the learning bias. Still, it provides under circumstances a simple and versatile tool for building a good bias. Besides its importance in relation to the learning bias, the notion of perceived dependency has a strong epistemological appeal, a simple definition, and an accordingly simple representation. It can under circumstances be reasonably efficiently built and maintained, and it has implications in the difficult question of representation change.