Linked production rules: controlling inference with knowledge

A key insight in artificial intelligence, which has been the foundation of expert systems and now business-rule systems, is that reasoning or inference can be separated from the domain knowledge being reasoned about. We suggest that the knowledge acquisition and maintenance problems that arise, might result from too great a separation of knowledge and inference. We propose Linked Production Rules, where each rule evaluated directs the next step of inference and the inference engine has no meta-heuristics or conflict resolution strategy. We suggest that this loses none of the power of conventional inference but may greatly improve knowledge acquisition and maintenance since various Ripple-Down Rule knowledge acquisition methods, which have had some success in facilitating knowledge maintenance can be described as specific instances of Linked Production Rules. Finally the Linked Production Rule approach suggests the possibility of a generalized Ripple-Down Rule method applicable to a wide range of problem types.