DAML and Agents
Breakout Session
DAML PI Meeting
17 October 2002

Goals
Find out who is doing what
Enumerate issues and open problems
Suggest potential new infrastructure, e.g., ontologies, software components, tools, …
Identify how agents+DAML can help in DOD applications
Recommend next steps

Starting points
Current experience: How has DAML been used in what agent frameworks and what is the outcome.
Future: How should DAML fit into existing agent infrastructures (DARPA Grid, FIPA, Cougaar, OAA, ...)
Agent varieties: How is/can DAML be used with mobile agents, interface agents, learning agents, etc.
Ontologies: what common ontologies are available or needed for agent-based systems (e.g., agents & their properties, speech acts, user models, trust, policy, …)
DAML-S: How well does DAML-S fulfill needs for service description & discovery in MAS (e.g., FIPA DF, Retsina, OAA).
Tools: What tools, software components, techniques are available or needed (e.g., DAML to KIF/SL translators)
DOD: How can agents+DAML support DOD applications, e.g., by providing needed ontologies (temporal, spatial, geospatial, ...) or aiding semantic integration with DOD legacy systems.

Sharp definition of an agent?
… in contrast to traditional web services?
Nope.
 But there are some characteristics…
Agents do things and can be autonomous – focus on decisions and actions (describing and constraining)
Agents are (often) part of a community of peers – focus on communicating knowledge
Agents represent people and organizations – focus on modeling their clients

Paradigm shift
Most agent concepts and technology came from the client server and message passing
The web offers new paradigms
Shared memory vs. messaging

No dominant agent framework
There is, as yet, no dominant, general agent framework
Some standards are
FIPA (http://fipa.org/) standards for agnet languages, protocols and services
Major DOD R&D frameworks: DARPA Grid, Cougaar
Smaller communities: OAA, Retsina, …
Vaguely agent like frameworks: Jini, …

DAML’s impact on agent systems
There is considerable interest and much preliminary in using DAML in existing agent frameworks, such as FIPA.
It’s being used as a KR language, as an interlingua, for ontologies, to describe policies, for service description and matching, etc.
None of this has been codified as part of any standard.

FIPA Agent Platform

FIPA Standards Overview

Things to do

Ontologies for core agent services
We should develop common ontologies to support core services which facilitate multi-agent systems, such as:
Service registries
Authentication, security policy checker
Trust and reputation server
Etc.

More ontologies important for agent
Speech acts (inform, request, query-if, …)
Conversational policies and interactions protocols
Policies for behavior norms
Security and trust concepts
User models/profiles for people, roles, stereotypes, and organizations

Protocols for agent interactions
In developing things like DQL we should use or at least be informed by protocols designed by agent systems
These have been explored in several contexts and some reasonably mature protocols have been developed and specified, e.g.:
FIPA IP library

Tools
Some ideas for tools that would be useful in a MAS context were discussed, e.g.:
FIPA DF -> DAML-S translator could enable access to Agentcities services
Well defined and documented reasoning module (in Java) for DAML and RDF to use with an agent.

Ending point
The envisioned grand convergence of Web, Agents and legacy applications
Additional possible actions:
Evangelize in agent communities (DARPA Grid, Cougaar, FIPA, OAA, etc) to get DAML used for … content language, ontologies, service descriptions, etc.
Evangelize in W3C community to consider incorporating agent concepts and languages