Trip Report

Meeting at Sun Microsystems

February 19, 1999

Frank Manola
Object Services and Consulting

I attended a meeting at Sun Microsystems, Burlington, MA on FIPA- and CoABS-Grid-related activities.  The other participants were Jeff Bradshaw (Boeing, working on the JumpStart project), and Kate Stout and Geoff Arnold (Sun Labs).

Jeff briefly ran through his JumpStart presentation from the last CoABS workshop, discussing their work on designing and verifying conversation policies between agents, as well as their work on mobility and resource management.  He noted that he had just been at MIT talking about a CoABS TIE involving integration of JumpStart work and MIT's work on exception-handling .  Boeing is interested in various low-level Java technology in their work (e.g., virtual machine modifications to help with mobility and resource management, Jini + services), which is one of the reasons they are working with Sun.  Jeff said that agent languages need to be compositional (their language approach is illustrated in [BDBW97]).  He said they are now using XML as a representation for everything (each agent has IBM's XML parser built into it).  He also noted the need to be able to construct narratives that explained agent actions in human terms, so that humans could understand what the agents were doing or had done (involving the need to "unoptimize" and inverse-map agent interactions into human-understandable descriptions).  This is highly relevant to the "tracking" issue that Jim Hendler mentioned as being important in our meeting earlier this month. They are also working with Cyc (to some extent), along with MIT, on a formal semantics for control.  In connection with the tracking issue, I mentioned to Jeff that there was a famous (among fans) Bugs Bunny cartoon which provided a funny illustration of the importance of this type of "tracking" (Daffy Duck reruns through a conversion with Bugs in order to figure out why, when Daffy was urging Elmer Fudd to shoot Bugs, Elmer shot Daffy instead.  Jeff later emailed me to get the title of the cartoon, which I gave him;  he said he may want to use it as an illustration).

Sun has joined FIPA, and has (along with IBM) submitted a proposal (which was accepted) to create a formal FIPA architecture, to be the basis for FIPA 2000 work.  The architecture would define a set of services, communication model, etc.  They feel that FIPA's specifications are currently inadequate to handle heterogeneity, large-scale agent systems, disconnected operations, agents that persist for long periods on the network, and other issues that will need to be addressed in industrial-strength agent systems.  In particular, the FIPA specifications need to explicitly consider distributed object systems, the Web, E-commerce, and the fact that agents will be part of larger systems.  (Kate and Geoff ran through the presentation they and IBM had made to FIPA, which included the "stick" that, if FIPA didn't address these issues, major players like Sun and IBM would go ahead and develop specifications independently of FIPA, and FIPA would eventually be marginalized).  A new FIPA technical committee has been formed to address the architecture work.  The architecture would be defined as a set of language-independent specifications, along with instantiations in Java (done by Sun), CORBA/C++ (done by IBM), and possibly others.  They mentioned that there were possibly going to be some problems relating to FIPA's current procedures not being very well adapted to commercial standards development (they would like to see FIPA move toward something more OMG-like, e.g., formal change proposals, more incremental work between meetings).  They would like to see Java agent standards for platform services by the end of the summer, and expect to have initial work done on a set of abstract descriptions of the current FIPA model and additional services (and other things) needed for presentation to the April FIPA meeting.  Other players in this FIPA activity include David Levine (IBM Watson Labs) and Charles Pow (Lockheed-Martin).  Sun has also been involved (mainly as onlookers) in OMG's agent work, and are working on an Agent Green Paper.

We also briefly discussed the CoABS Grid.  Jeff distributed copies of Brian Kettler's presentation from the January CoABS workshop.  I described the papers Craig and I had written on trying to characterize the Grid concept (and gave the Sun folks the URL to our Agility page).  Geoff said that the Grid sounded a lot like the idea behind the Java Jive work which later got boiled down to Jini (as being a concrete starting point on top of which the higher-level grid-like capabilities could be developed).  This seems to fit in with the approach Brian is taking in defining basic Grid services (although there seemed to be general agreement that you want both bottom-up work on identifying basic capabilities and top-down work, like ours, on identifying the ultimate capabilities you are after).  There was some mention of tying up the Grid idea and the FIPA architecture work.

A further meeting between Boeing and Sun (and I don't know who else, or where) may take place the week of March 15.

In the course of the discussions, I briefly described the Advanced Logistics Program, and the interest in identifying technology transfer opportunities between ALP and CoABS.  Jeff expressed interest in this, and although we didn't have time to get into this further, we plan to interact further by phone or email.  Jeff mentioned that he had talked to Steve Milligan at the CoABS workshop, and got the feeling that Steve felt that there wasn't a lot in CoABS that seemed immediately useful.  We'll have to see how that stands up after further discussions with the ALP folks.

[BDBW97]  J. Bradshaw, S. Dutfield, P. Benoit, J. Woolley, "KAoS:  Toward An Industrial-Strength Open Agent Architecture", in J. Bradshaw (ed.), Software Agents, AAAI Press / MIT Press, 1997, 375-418.