Summary and discussion Session 4: Communication Tim Finin University of Maryland - Baltimore County Slides: PowerPoint (242K)

Relevant links

Summary

A software agent is an autonomous, goal-directed process that (1) is aware of and reacts to its environment and (2) cooperates with other agents to accomplish its task. This cooperation is the key to realizing the potential of the agent paradigm, and cooperation can only be achieved through effective agent-agent communication.

How do agents communicate? There are four levels at which agents can communicate: (1) object sharing - shared objects, procedure calls and data structures; (2) knowledge sharing - shared facts, rules, constraints, procedures and knowledge; (3) intentional sharing - shared beliefs, plans, goals and intentions; and (4) experiential sharing - shared experiences and strategies. KQML (combined with KIF and appropriate ontologies) is an example of knowledge sharing. It is described further in Tim's slides.

Communicating mobile agents? There has been no serious integration of the mobile-agent and software-agent paradigm, but there appear to be no immediate deep issues (an initial integration of TKQML and Dartmouth's Agent Tcl has already been done). A key practical issue, however, is to develop protocols for finding other agents and for forwarding messages to an agent's current location.

What should agents communication about? The Web! The SHOE system has two main components: (1) semantic markup tags that Web authors can use on their web pages and (2) "URL agents" that understand small groups of related web pages (and communication with other agents with similar pages). Other agents can ask the URL agents questions about their web pages (via KQML).

Tim closed with a key question: How do we make mobile and communicating agents relevant to the commercial world? Are our current systems too theoretical or too academic? Can we build on emerging standards such as CORBA? Will be able to define common communication languages, common ontologies for particular domains, and common agent protocols, so that all the different systems can work together?

Discussion

Q:	Are mobile agents different from stationary agents that communicate?
Q:	In practice, they are implemented differently.

Q:	What about XML? (XML is another semantic markup language for Web pages.)
A:	It is similar to SHOE.

No real world applications for KQML? Several people in the audience said that they were developing commercial applications with KQML.

Someone raised the issue of Named Universal Locators. The response was that creating a global namespace is hard, and is not applicable to rapidly-changing agents.

In reference to KQML, someone raised the issue that logic programming doesn't scale. Does this imply that there is a problem with large numbers of agents based on KQML?