Agent-Based Composite Services in DAML-S: The Behavior-Oriented Design of an Intelligent Semantic Web ∗ Joanna J. Bryson Primate Cognitive Neuroscience Laboratory Harvard University Cambridge, MA 02139, USA joanna@ai.mit.edu Lynn Andrea Stein Computers and Cognition Group Franklin W. Olin College of Engineering Needham, MA 02492, USA las@olin.edu ABSTRACT Many researchers are working towards the goal of a semantic web — a web that provides information useful to artificial intelligences. A semantic web would allow artificial agents to do the work of searching for and organizing services required by humans or organizations. DAML-S is a web service on- tology intended to facilitate the semantic web by describing the properties and capabilities of Web-available services in an unambiguous, computer-interpretable form. In this paper, we propose that the content of the semantic web should be thought of not as knowledge, but as intelli- gence. The features encoded in DAML-S would then be specifications either for extensions of the agents attempting to exploit the composite service, or of independent, collab- orative agents that can be ‘woken’ to assist the user agent. We draw on our experience in agent development to elabo- rate the specification particularly of the process ontology of DAML-S in order to support this vision. General Terms Paper ID: 554 Keywords Standards for Agents and MAS; Electronic Commerce; Ac- tion Selection and Planning 1. INTRODUCTION: INTELLIGENCE AND THE SEMANTIC WEB ∗ Effort sponsored by the Defense Advanced Research Projects Agency (DARPA) and Air Force Research Labo- ratory, Air Force Material Command, USAF, under agree- ment number F30602-01-2-0512. The U.S. Government is authorized to reproduce and distribute reprints for Govern- mental purposes notwithstanding any copyright annotation thereon. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Copyright 2001 ACM X-XXXXX-XX-X/XX/XX ...$5.00. The World-Wide Web has revolutionized the communi- cation of information between humans. However, we crave more than rapid access to information and services. What we want are intelligent schedulers, planners and searchers that, with minimal direction, can serve as an omnipresent staff of advisors, secretaries, agents, brokers and research as- sistants. We want agents without issues or personal lives to plan everything from vacations to colloquiums, product de- velopment cycles to birthday parties. In short, we want com- petent artificial agents to do the mundane organizational tasks in our lives. The explosion of electronic commerce has brought this vi- sion tantalizingly near. The general public has done the hard work: they have connected an enormous variety of products and services to the Internet, making them accessible to com- puter programs via simple communication protocols. Unfor- tunately, this puts our community squarely on the spot. In order for intelligent agents to use the web as it now stands, they must understand ordinary language web pages and pre- sumably a host of other cultural protocols which reduce the ambiguity rampant in such language. There is some hope that parsing ordinary web pages might succeed in sufficiently stereotypical transactions: see for example Koller [21] for a discussion of the information beyond natural language avail- able in web pages. In this paper, we focus instead on the alternative ap- proach: changing the web in order to make it accessible to existing AI. This new “semantic” version of the web would consist of pages marked up in accordance with standardized conventions in order to reduce ambiguity and facilitate au- tomated reasoning [5, 27]. This approach has been the focus of a large number of research initiatives [1], which has lead to the prediction that “Soon it will be possible to access Web resources by content rather than just by keywords” [2, p. 411]. Ankolekar et al. [2] describe a formalism, DAML-S, de- signed by the DARPA Agent Markup Language (DAML) Services Coalition to facilitate accessing services via the web. In this paper, we work to extend and refine this work. We propose that in designing the semantic web, we should be thinking not only about ways to type, flag and advertise the information on the web, but about how to make the web actively usable. We should expect the semantic web not just to extend the knowledge of artificial assistants, but to extend their intelligence. This is because a service is not in- formation, it is behavior. And behavior is the fundamental attribute of intelligence.