Deriving conclusions in expert systems when knowledge is incomplete

A number of recent research articles have pointed out several connections between topics in propositional logic, artificial intelligence and MP. In particular, Jeroslow and Wang demonstrated how inferencing in a Horn clause knowledge base can be carried out by solving a linear programming problem and showed how the dual to this problem could be used to explain how a given conclusion was reached. This paper extends these results to develop a model that allows heuristic inferences to be made when there is insufficient information available in a knowledge base to reach a definite conclusion and additional information cannot be obtained. This is done by identifying the conclusion that is most consistent with the knowledge at hand.