Abstract
Domain knowledge is viewed here as consisting of two parts: A domain theory, TH[D], for a domain D, which can be used to decide whether a given instance, S,belongs to the domain D or not, and domain heuristics, H[D], which can be used by a machine to solve problems in the domain D using the theory TH[D]. A machine that can build theories TH[D] and learn H[D] from experience of using TH[D] to solve problems in D is called a Knowledge Based Learning machine. The theory formation aspects of a knowledge based learning machine are illustrated here with a simple example. It is argued that such a machine should be endowed with the right kinds of a priori "biases" in order to be able to examine given instances of a domain and formulate domain theories. The example is used here to present a typical set of generally applicable "biases" of a knowledge based learning machine.