Agents and the Semantic Web: Papers from the AAAI Fall Symposium
Terry Payne and Valentina Tamma, Cochairs
The semantic web is based on the idea of dynamic, heterogeneous, shared knowledge sources providing machine-readable content in a similar way to that in which information is shared on the World Wide Web. Integral to this vision was a synergy with multiagent systems technology; agents could use this knowledge to achieve their own goals, producing new knowledge that could be disseminated or published within a common framework. Conversely, the semantic web would benefit from autonomous, distributed agents responsible for gathering or aggregating knowledge, reasoning and inferring new facts, identifying and managing inconsistencies, and providing trust and security mechanisms.
Previous workshops and discussion devoted to this topic have mainly focused on either the semantic web aspect or the agent aspect of the problem, and have failed to achieve an agreement on the common research issue, leaving several open problems unaddressed, such as the following:
Knowledge Sharing: The agent paradigm is successfully employed in those applications where autonomous, heterogeneous, and distributed systems need to interoperate in order to achieve a common goal, however this is possible if agents are able to share knowledge. Ontologies are a powerful tool to achieve semantic interoperability among heterogeneous, distributed systems.
Syntactic Unification: Data exchanged between service providers are typically based on different syntax, raising the problem of data mediation for interoperability. Ontologies, and mechanisms for mapping and translating across ontologies can address these problems.
Discovery of Agent Capabilities: Semantic-based discovery mechanisms and languages/ontologies for describing agent capabilities and predefined coordination mechanisms are needed to make the automatic discovery of services offered by agents and other providers.
Agent Coordination: Goal-directed composition typically involves planning across a space of existing actions, ensuring that data and control flow constraints are satisfied. Model checking techniques are required to ensure valid compositions, as well as temporal reasoning to validate control flow dependences. Such techniques need to accommodate semantic descriptions as well as avoiding live-lock situations that may lead to failure.
Interaction Protocols: Different agents expect specific messages to be choreographed in a precisely defined manner. Integration has to guarantee and enforce the communication protocols. Interoperable description frameworks are thus required to ensure that both parties understand and adhere to interaction protocols. The semantics of the terms used in these protocols is made explicit in ontologies.
This symposium promoted and fostered a greater understanding of the synergy between multiagent systems and the semantic web.