PRIME: a bottom-up approach to probabilistic rule development

PRIME (probabilistic rule induction mechanism), a program which demonstrates a bottom-up approach to developing a rule base, is described. It is a system to be used by an intelligent machine to allow it to operate autonomously in an abstract but uncertain (or stochastic) environment. The purpose of PRIME is to allow an intelligent machine to satisfy user-specified goals with maximum success probability. To achieve this objective, it maintains a probabilistic model of the machine's effects on its environment, in the form of a rule base, and continuously updates its knowledge on the basis of recent experience. To learn the rule probabilities, a two-level estimation procedure is used, which is shown to be effective at tracking nonstationary probabilities for certain choices of parameters. The planning mechanism in PRIME is also shown to perform its task of deriving optimal plans satisfactorily. The results clearly indicate that goal-directed exploration is a desirable, if not necessary, function of PRIME in order to generate, maintain, and use a rule base in a sizable environment.<
>