What's an Agent?
Good question: the first answer is that there must be a protocol of message types and possibly sequences that are legally exchanged among software modules. Different protocols define different agents. In our opinion, and there is no consensus on this, an agent is a software module in a distributed collection of modules working together as a system conforming to a message protocol defined prior to the system integration and reusable for other applications. This is a loose definition so far, but a good example of it is the combination of SHADE KQML and Redux' protocols in Next-Link. The next obvious question is the relationship of Next-Link to SHADE.
Relation to SHADE
The SHADE project, part of the Knowledge Sharing Project , is addressing the problems of information sharing among disparate engineering tools. This work includes the development of common ontologies ( Ontolingua), finding the right protocol and performatives ( KQML) that computational agents should use to coomunicate with each other, and the problem of finding the right agents with which to communicate (content-based routing and facilitators). A typical use of this technology is a user attempting to find information on a network with two or more computational working together to provide the information. Also see the Agent Communication Language based on KQML/KIF.
We assume that these problems will be addressed by the research already underway in these projects. In fact, we will depend upon these research results, as shown in the NEXT-LINK deliverable description.
The cable design scenario illustrates some of the coordination level issues that are addressed by NEXT-LINK. In particular, we hope to manage the complex interactions among tasks refered above: reducing thrashing, inconsistencies, eliminating uneeded work, and finding opportunities to improve the design.
MAC users may retrieve
compacted self-extracting binhexed copies of a PowerPoint 3.0
overview of related SHARE projects and the relationship of
Next-Link to SHADE. (176K)
Related CDR Work:
In addition to the completed First-Link project, there are two current CDR projects strongly related to NEXT-LINK:
ACaPS - This is a related Lockheed project to provide an Agile Cable Production Service.
CABLER - This is an off-shoot of the First-Link project which focuses on interactive 3D routing issues.
Next-Link is related to these projects as a SHARE family of cable-harness related projects.
Check out the local Agent-Based Engineering (ABE) working group.
The Concur project covers a lot of basic ABEissues and technologies.
We are also working with Yan Jin and Raymond Levitt on the CIFE project ASCD: Agent-Supported Collaborative Design.
Other Related External Work:
See also the Sandia Intelligent Agents for Manufacturing project,
of which the CDR SIAM project is part.
The ASU Virtual Corporation is a distributed design project.
For a general guide to Distributed AI, see the UMass DIS Server.
For a general guide to Software Agents, see Ralph Becket's Intelligent Software Agents page.
David Brown has has developed a Webography of AI in Design resource.
He also has a Multi-Agent Systems Webliography.
The Constraints Archive is a comphresive index to research on constraints.
The Lockheed COSMOS system is using declared constraints to perform sensitivity analysis on proposed design changes. COSMOS could be perhaps used with Next-Link once COSMOS becomes a KQML-based agent.
Check out The Software Agents Mailing List FAQ
and the AgentNews mailing list.
Then there is Tim Finin's KQML agent news. Also see the Agent Sourcebook.
Further Reading on Next-Link and Redux:
A more detailed understanding of the approach can be found in Redux . Further detail on the general theory can be found in the paper outlining how NEXT-LINK will be using Redux: Abstract or Report . The cable scenario and how it is accomplished in Redux is described in a paper focusing on Pareto optimality: Abstract or Report (PostScript) / (HTML).
Industrial sponsors and applications for this technology are welcomed.