Some Agent Theory for the Semantic Web Leona F. Fass P.O. Box 2914, Carmel CA 93921 <lff4 [AT] cornell [DOT] edu> Abstract We take the position that for any goal achievable on the Semantic Web, there will be a “best” system of Web-dwelling software agents to realize that goal, and that such a system may be discovered effectively. The process of determining the “best” agent system may be overseen by a distinguished Manager Agent. But with realistic time and space constraints, and the dynamic nature of the Semantic Web, finding an approximating system may be acceptable. The approximation then may be adapted iteratively, to approach the ideal. We show that very practical researchers have looked at software agents and Semantic Web problems in a similar way, determining approximating sub-optimal systems and subsequently adapting them. Their applied research confirms that theory provides a good foundation for practice. Keywords: Software agents, multi-agent systems, Web services, the Semantic Web, theoretical foundations. Introduction The aim of Semantic Web designers and developers is to enhance the existing Web, supplying structure and meaning that will facilitate machine-to-machine interaction. With agent and multi-agent system research and development integrated into Web development, it should be possible for software agents to utilize the Web environment, fulfilling complex tasks that human users have specified. Given a human-specified task (e.g., “Schedule my appointments for the East Coast trip”, or “Get me contact information for the Cornellians and Penn grads who live in this area”) capable agents could produce a satisfactory result. They would access machine-readable information, discover fulfilling Web services and automate their use, making logical decisions and invoking and composing services as needed. In inceptive articles about the Semantic Web James Hendler [13] and Tim Berners-Lee, James Hendler and Ora Lassila [2] described some of its proposed features including languages, communicating ontologies and inferencing mechanisms. They provided illustrative examples of “agent-based computing” and the role that agents can play in Web applications, interacting with each other and with human users. There is typically an Artificial Intelligence emphasis on Semantic Web and agent research. Agents often are depicted as lifelike, as they may navigate the Web and interact with other agents or with people. (The latter aspect also brings a Human-Computer Interaction perspective to the field.) While they may communicate, learn and adapt, these agents are really just software. And so, as well, is the Semantic Web. Hence the development of the Semantic Web and Web-navigating agents must involve substantial Software Engineering. In a project as massive as the Semantic Web there is room, and a need, for multiple and coordinated research emphases. We ourselves take an automata theory approach to agents and the Semantic Web, motivated by some intriguing software agent- and Web-related research we identify as theoretically grounded. Thus we look at tasks to-be-fulfilled or problems to-be-solved on the Web as behaviors to-be-realized. Viewing the Semantic Web itself as a multi-agent system we look at sub-systems constructed of Web-dwelling agents to fulfill specific tasks, or to solve specified problems, as “behavioral realizations”. With this very theoretical perspective we propose that any goal achievable on the Semantic Web will have a “best” multi-agent system behavioral realization. We take the position that this realization may be constructed or adapted effectively from Semantic Web-dwelling software agents, perhaps by a distinguished Manager Agent. The system agents may invoke and compose Web services, as needed, to achieve behavioral goals. Since this theoretical result may disregard real constraints on time and space, we next discuss the necessity of accepting approximating Semantic Web-dwelling multi-agent systems. These can adequately fulfill tasks or effect problem solutions that are “good enough”, relative to time, space, and information they may obtain. We describe adaptation of such approximating systems, when tasks, goals, information, or available components change, or when anomalies or defects are found. Despite the abstract approach we take when considering these issues we find other, very practical, researchers have looked at multi-agent systems and the Semantic Web in a similar way. We describe their related work with multi-agent system managers, approximating results, iterative approaches to the “optimal”, and constraints applied so that features of the dynamic Semantic Web can be represented to software agents effectively. Certainly, we believe that there are problems for which optimal results may never be found. But we also believe, from the practical work we have seen, that when considering agents and the Semantic Web, theory provides a good foundation for practice. Research Motivating Our Approach We were motivated to consider theoretical problems related to the Semantic Web and Web-dwelling software agents from our analysis of some very interesting research in several related areas. These included fairly recent work in: emergent semantics; automated composition of Web services; deduced interaction; collaborative learning; agent coalition formation; and an agent’s locally-closed view of its world. Each bears some relation to our automata theory view of behavioral modeling, and each has convinced us to extend our theoretical work into problems of the developing Semantic Web . In describing emergent semantics Steffen Staab [22] noted the possibility that semantic links to Web pages could be inferred from observations of actions of human users. Luc Steels [23] described knowledge representation for agents, so that they could interact. Equivalent external representations that agents needed to communicate did not imply equivalent internal representations. In our own inductive inference research (such as described in connection with model-based validation [5]), we had shown that the structure of a language (or device) could be inferred by observing examples of behavior. We’d also shown that behaviorally equivalent results, with different underlying structures, could be found. Thus we were interested in the concept ACM SIGSOFT Software Engineering Notes Page 1 September 2005 Volume 30 Number 5of inferring