Large-Scale Knowledge Systems

Abstract

AI technology has produced a variety of tools for generating knowledge-based systems ranging from Programming Languages to Expert System Shells. When used for large-scale applications, these tools suffer from several significant limitations. First, virtual memory size limits the volume of knowledge that can be handled. Second, there are no facilities for allowing the knowledge base to be shared among several applications. Third, inference mechanisms are computationally intractable when large knowledge bases are resident on secondary storage. Fourth, there are no facilities for distributing the knowledge base across multiple workstations to achieve better performance and reliability. Database technology, when properly integrated with these AI tools, has the potential for overcoming these limitations. However, the solution requires more than just coupling a DBMS as a back-end storage system to an AI tool. To achieve the best functionality and performance, the tool’s knowledge representations and inference engines must be designed to operate in cooperation with those of the DBMS. Knowledge processing must be modularized to provide a stable “working set” for the tool inference engine. Inferences that can be performed efficiently by the DBMS alone must be separated by the tool inference engine and delegated to the DBMS. Furthermore, DBMS facilities must be extended to provide more efficiency and flexibility in knowledge processing. A lot of design and experimentation will be required before the full integration of AI Knowledge Tools and DBMS technology is fully understood. This paper takes a first step towards this integration by identifying approaches and principles in the design of large-scale knowledge systems.