Abstract

Are all systems knowledge-based? What special characteristic(s) separate knowledge-based from non-knowledge-based systems. Our introductory remarks are aimed to stimulate discussion q{ what exactly constitute knowledge-based systems. We briefly explain and discuss three disparate knowledge based system applications under development in universities. One application falls under the heading of constraint oriented logic programming expert systems, the intelligent graphics interface (IGI) project. We describe System X, our natural language interface designed/or executive information access and manipulation as a knowledge-based system. We then explain how DB-Discover our system for discovering the implicit classifications and categorizations in a relational data base was designed and how it works. We conclude by discussing some ramifications of knowledge-based systems.

Introduction

Artificial Intelligence (AI) researchers noticed early on in the development of expert systems technology that expert systems derive their power from the knowledge they possess rather than on particular inference mechanisms or representational formalism (notwithstanding arguments of expressive power and notational efficacy). These later considerations, however, are certainly important for overall system performance and efficiency. This observation led to more robust problem solving and knowledge representation schemes and these, in turn, led to more efficient constraint propagation algorithms and effective inference implementations in modern approaches to building knowledge-based systems.

At one time the terms expert system and knowledge-based system were synonymous; contemporary usage generally favours the more general and encompassing term, knowledge- based system. Architectural principles for knowledge based systems emerge when certain characteristics are identified. Knowledge is the central ingredient powering expert systems. Knowledge is often inexact and incomplete and often poorly specified. Novices become knowledgeable incrementally. Knowledge-based systems should be flexible as well. Many early AI systems simply did not qualify as knowledge- based systems even though their performance may have been admirable. For example, the general problem solver.

To address these requirements, IGI has evolved the architecture shown in Figure 1. An IGI is a knowledge-based mediator between multiple real-time asynchronous control systems and multiple human operators working within an information bandwidth limited display environment. The mediation paradigm views IGI as an intelligent negotiator which uses various knowledge sources to manage the limited user interface resources. These limited resources include:

  • host workstation(s) display capabilities

  • existing interface protocol with the application database(s); and

  • most importantly, the attention limitations of the human operator(s).

Figure I illustrates this definition schematically. IGI uses a number of specialized knowledge bases to mediate the interaction of operators with the control system database. The mediation process is dialogue-based among operators, user interface objects and control system objects. The dialogues arise asynchronously by operator, program or application control. Dialogues internally establish a synchronous communication path between objects which may persist indefinitely. A dialogue consists of a sequence of one or more messages exchanged synchronously and bidirectional between objects. Figure 1 - Functional Overview of the lGI Architecture. (Available in full paper)

Dialogues are persistent communication media required for IGI reasoning. We compare dialogues with the message pa

This content is only available via PDF.
You can access this article if you purchase or spend a download.