Increasingly, knowledge, as well as information and data, is being transferred over the World Wide Web. There is great potential in linking traditional knowledge-based systems (KBS) technology with the Internet because each technology can overcome limitations in the other. As a result, it might enable expert knowledge that has hitherto been confined to those who possess the correct computing platforms to be made available to small enterprises and people in developing countries.
Five types of KBS-Internet integration are outlined (Intelligent Agents, Active Web Pages, Local KBS Accessing Web-distributed Information, Web-distributed Knowledge Bases, and Knowledge Servers). This paper discusses knowledge servers in detail. It examines the issues and problems that must be addressed if existing KBS inference software is to be integrated with the World Wide Web, and discusses, in depth, solutions as implemented in the Istar knowledge server. The paper shows how technical design and implementation decisions can be influenced, not only by the technical characteristics of the Internet, but also by a range of other, 'softer' issues. In particular, it shows how real life styles of WWW browsing, and a desire to make knowledge available to developing countries, influences both overall architecture and detailed implementation decisions. Early experience of actual usage shows Istar to be a highly efficient knowledge server. Directions for future research are discussed.