Towards a Worldwide Knowledge Base Extended Abstract

The paper proposes that the basic idea of the WWW can be generalized to the domain of data structures or information structures. This means that \packages" of structured information are represented as short text les, stored in web-accessible locations (URL:s), and are loaded as needed by the web-oriented database tool. We suggest that this technique may help to get done what has not been achieved before in the history of A.I. research: a comprehensive, universally usable knowledge base of common knowledge. 1 Questions Why doesn't there exist, after 35 years of research in artiicial intelligence, a com-prenhensive knowledge base of common-sense knowledge which everyone can use as a starting-point for her projects? Since "symbolic", "representation-oriented", and "knowledge-based" approaches have dominated throughout the history of A.I., why is it that every research group and every project builds its own knowledge base more or less from scratch, to t its particular application area? If common-sense knowledge is truly universal and can not be cut into small self-contained pieces, why is it that almost every project that professes to use a knowledge-based approach does exactly that? And, if we agree that the present situation is unreasonable and that a comprehensive universal knowledge base is a necessity for A.I., what are the possibilities of changing the situation by using the new software technologies that have emerged in recent years, in particular, the WorldWide Web technology? The present paper attempts to give a brief answer to the rst set of questions, and a more extensive answer to the second set. 2 Why don't we have a comprehensive knowledge base today? The present absence of one or more widely used knowledge bases is of course due to a combination of several factors. I will therefore only identify a few aspects of the relevant background.