Improving quality of interference in multilevel secure knowledge-based systems

The issue of providing security to knowledge-based systems has been addressed in the literature. Most of these works aim at protecting secure information from leakage by enforcing security constraints. For security reasons, the knowledge available at high security level is not disclosed to subjects holding lower security clearance levels. The inherent inability of multilevel secure systems to utilize the high security knowledge for inference may lead to poor Quality of Inference at low security levels. The paper explores the issue of improving Quality of Inference without significantly compromising security. We first develop a secure Match-Resolve-Act algorithm. We then illustrate the Quality of Inference issues arising in a secure knowledge based system, and suggest an inter-level inference engine based architecture for improving the Quality of Inference. We also study the impact of inter-level inference on security by applying the Sphere of Inference notion.