The DARPA Knowledge Sharing E ort : Progress

Building knowledge-based systems today usually entails constructing a new knowledge base from scratch. Even if several groups of researchers are working in the same general area, such as medicine or electronic diagnosis, each team must develop its own knowledge base from scratch. The cost of this duplication of effort has been high and will become prohibitive as we build larger and larger systems. Furthermore, lack of methodology for sharing and communicating knowledge poses a signiicant roadblock in developing large multi-center research projects such as DARPA/Rolm Laboratory Planning and Scheduling Initative 21]. To overcome these barrier and advance the state of the art, we must nd ways of preserving existing knowledge bases, and sharing, reusing, and building on them. an initiative to develop the technical infrastructure to support the sharing of knowledge among systems. 27] The goal of this eeort is to develop a technology that will enable researchers to develop new systems by selecting components from library of reusable modules and assembling them together. Their eeort will be fo-cused on creating specialized knowledge and reasoners speciic to the task of their system. Their new system would inter-operate with existing systems, using them to perform some of its reasoning. In this way, declarative knowledge, problem solving techniques and reasoning services could all be shared among systems. The reusable modules in the library themselves will beneet from reenements that are only possible through extensive use. This would facilitate building larger systems cheaply and reliably. The infrastructure to support such sharing and reuse would lead to greater ubiquity of these systems, potentially transforming the knowledge industry. The work in the Knowledge-Sharing EEort began with the identiication of the impediments to knowledge sharing and corresponding needs for the development of technology to overcome these impediments. Four key areas were identiied for the initial eeort. They are: (1) mechanisms for translation between knowledge bases represented in diierent languages; (2) common versions of languages and reasoning modules within families of representational paradigm; (3) protocols for communication between separate knowledge-based modules, as well as between knowledge-based systems and databases; and, (4) libraries of \ontolo-gies," i.e., prefabricated foundations for application-speciic knowledge bases in a particular topic area. A detailed discussion of the impediments, and an analysis of the issues that motivated us to focus on these four types, appears in 27]. That article also describes the working groups (comprised of researchers from the DARPA AI community and other …