Task one - integrating agents and the web

We believe that, as a first order theory, the applied intelligent systems of the next decade will be based on the rapidly evolving base of web technology. This is the key insight of the DAML program -- to promulgate a universal web markup language that is sufficiently rich to support intelligent agents and other applications. Thus, our first task, headed by UMBC, will focus on developing the support and infrastructure required for intelligent agent software to integrate seamlessly into a web-based environment of web browsers, web servers, web application server platforms, and associated supporting languages (e.g., WEB/SQL, WEBL), protocols (e.g., SSL, S/MIME, WAP, eSpeak), services (e.g., LDAP) and underlying technologies (e.g., Java, Jini, PKI, etc.). Our approach will be to build on the theory, languages and software base of agent research over the past ten years (in which we have been a significant player). This will involve adding concepts and components at the "agent communication language" layer and the "agent services" level to DAML, and migrating, enhancing and building on our extensive work in developing agent software such as is being done in CoABS and in our own NIST sponsored work (Cost, 1998, 1999). The result will be an environment in which agents can exist and work easily on the web -- in browsers, in servers and in ASP environments -- and using their own ACLs but adapted to the web environment.

Task two -- DAML meets practical rule-based technology

DAML is envisioned as a web-based markup language that has the expressive power of at least a large subset of first order logic. We expect that it will be, or at least aim to be, largely based on a fairly pure, declarative foundation. However, nearly all of the systems which are widely used in practice for knowledge-based and rule-based reasoning have some critical (and in most cases essential) features which will be in conflict with such a pure foundation. In this task we will develop a software component which is capable of mapping expressions in DAML into practical executable form in conventional rule-based languages supporting (and requiring) such features as conflict handling, default reasoning, rule prioritization, rule naming, procedural attachment, etc. Moreover, we will develop capabilities to map other rich markup languages (e.g., non-DAML XML, or XML Query) into a form suitable for knowledge-based processing. We will also develop an ontology and knowledge base to address the problem we call "distributed belief transfer". We see a critical need to be able to identify an agent as an authenticatable source for statements expressed in DAML markup so that agents can reason about such issues as "Should I believe this statement". "How much should I trust it?", "Is their any contradictory evidence?", and "Where can I get additional information to help to answer these questions"? This work will build on our initial attempts to characterize similar problems in the context of security and authentication, performed as part of our NIST sponsored EECOMS project.

Task three -- Adding knowledge to information retrieval

In task three we will develop a new kind of intelligent search and retrieval agent that works in an environment in which web pages have a combination of free text, low level XML markup and higher level DAML markup. Conventional search engines are complicated systems (Brin and Page, 1998) and many of the constituent modules will have to be modified to handle DAML or will benefit from DAML-ization. We propose a comprehensive framework and work plan for exploring this evolution from the traditional web-based IR task to the DAML-based future. We believe that such a hybrid information retrieval system will not only achieve better results than a text-only system, it will also outperform a DAML-only retrieval system.

Task four -- DAML.ORG as a vertical portal and testbed

Finally, we will design, develop and maintain a vertical portal for DAML (http://daml.org) that will simultaneously serve as a useful resource for the DAML research community, gather raw information required for our own (and other's) evaluation efforts, and serve as a testbed for components of our own work. We will draw on our experience in building and maintaining community-building portals in other areas, such as UMBC's agentWeb (http://agents.umbc.edu) and ectechWeb (via http://www.igec.umbc.edu). This portal will include a Yahoo-like directory as well as a search engine. The directory will include features that support the automatic or semi-automatic indexing of new pages based on our previous work on document characterization (Labrou, 1999). We will take the basic results of task three and realize them in a scalable Internet search engine. A rudimentary portal will be operational by the second quarter of GFY2001 with additional enhancements, capabilities and optimizations made over the next two years.

In Summary

In summary, we offer an integrated set of tasks that will result in innovative key software components to help advance the state of the art in building intelligent agents in an DAML-enriched World-Wide Web. Our team is experienced in all of the underlying technologies, has a proven track record of significant research accomplishments, and is has demonstrated an ability to work well together on complex team-oriented projects.