The following are notes taken during two discussion meetings of the new IEEE -FIPA launch at AAMAS, Utrecht, July 2005. Comments on this document can be emailed to FIPA-ROFS-Chair@ieee.org
Participants
Bernard Burg, Panasonic, burgb@research.panasonic.com
Monique Calisti, Whitestein, mca@whitestein.com
Jonathan Dale, Fujitsu U.S., jonathan.dale@us.fujitsu.com
Misty Nodine, Telcordia, nodine@research.telcordia.com
Michael Luck, Southampton University mml@ecs.soton.ac.uk
James Odell email@jamesodell.com
Lin Padgham linpa@cs.rmit.edu.au
Peter McBurney, University of Liverpool, p.j.mcburney@csc.liv.ac.uk
Steve Willmot, UPC, steve@lsi.upc.es
Stefan Poslad, Queen Mary, University of London, stefan.poslad@elec.qmul.ac.uk
Discuss how to continue FIPA / the need for agent standards.
Identify success criteria / visions.
Describe aims / milestones in the short-term.
Describe FIPA processes.
May not be able to define the roadmap or the vision to justify FIPA at the start. It is a chicken and egg situation. Participation by the masses should drive FIPA not the directors ,however FIPA needs a direction. Participants need guidance and a defined scope. FIPA as a space that has flexible work process vs. having very defined work-plans with customers.
The idea of FIPA was born in the second half of December 1995. At the root of FIPA are a set of facts:
Some agent technologies have reached a considerable degree of maturity;
To be of use some agent technologies require standardisation;
Standardisation of generic technologies has been shown to be possible and to provide effective results by other standardisation for a (MPEG and DAVIC).
History 1995-2005 (Short version):
1995: FIPA Root formed based upon - agent technologies useful, some are mature, standardisation useful, standardisation of generic technologies possible;
1997: FIPA focus along dimensions of agent management, message transport & ACL 1st set of 7 specifications & 7 core implementations resulted.
1998-9: core specifications revised and some new specifications developed.
2000: FIPA adopts an overall architecture model with less fragile abstractions, dont break as technology changes & mappings to commonly used technologies (CORBA, JINI etc); support alternate mechanisms, e.g., transports, content encodings; Explicit definition of implicitly used agent terms; new life-cycle model for standards
2002: 25 specs standardised & new activities started on Semantics, adhoc, Security, Services, Modelling, Methodologies.
2005: FIPA no longer autonomous becomes 11th IEEE SA (standards activity). IEEE has technical activities that move things to the standards stream. this research does not need to be done in the IEEE FIPA SA
Reflections on 1999-2005 wok focus on processes:
FIPA Focussed too much on plumbing.
Standards were research driven. Early FIPA was mainly research started some basic research don't want to do this again in a standards organisation.
FIPA focussed on addressed interoperability but many proprietary systems are in the market-place. Let market-place in vertical domain do the interoperability
FIPA did not have the capacity to interlink with the other communities that overlapped.
GRID has a common framework that is more generic than FIPA and has vertical tracks that are domain oriented like vertical markets that define their own VOs & Ontologies. FIPA model differs from this model in that it tried to harmonise the vertical domains as well - this may not be necessary. Don't need to be generic?
Web Services (WS): Some companies view it with some apprehension - WS acts as a direct tunnel to database, companies may not want this. Some things missing / not complete such as contracts and commitments firmly grounded in a WS are a syntactical framework they will walk away from anything that is a semantic framework.
OASIS is chaotic - can produce inconsistent specs FIPA spent a lot of time trying to make specs. consistent OASIS / W2C doesn't do this? Do FIPA Specs need to be normative? - this should be determined by market-place not the standards body.
FIPA must produce concrete things in only a few area and not do too many things, e.g., FIPA ACL is the most well-known . FIPA started off with specific application tracks and then broadened into logic and generic stuff and did not develop any further application tracks.
A Standards Activity needs to provide something concrete to exist.
What is the value for others to be articulated to participate in FIPA?
What agent stuff is needed for P2P, need protocols to download? Example of set-boxes, how to standardise to share content and to interoperate?
Need to attract the main players in a vertical domain to work together. FIPA provides a space to bring together larger group of players. But manufacturers who see a need to work with other companies to get harmony of product interfaces may no look to standards organisations such as FIPA to go to do this , they will go and do these themselves. FIPA must have a critical mass to participate and have resources to put in..
FIPA Can't do things in isolation must know the work of other communities and be able to connect. Does this mean FIPA needs to know not to do this, to know to add value to these, to draw things together (maybe lightweight). Some duplication of effort always occurs / of doing the same things in several ways using several technologies.
Ground specifications with practical implementations / technologies Drive things from technologies, e.g., success of EU IST triggered Agentcities project. Generating specs and then waiting for people to implement these is a slow, more uncertain process.
Implementations are necessary for firms to take up new ideas. FIPA agent platform seems too 'fat'. it designs services as agents that have already matured as non-agent services, FIPA-DF (directory facilitator service) vs. UDDI / LDAP
FIPA moved to IEEE. What are milestones expectations from the IEEE? None.
Promote market-places where agents can make a difference vs. promote an abstract architecture to support interoperability between groundings of technologies.
Can get interoperability through SOAP but this is a very simple model and has things missing.
Many standards grew out of research. Need implementations of standards to drive them to be used.
Originally academic researchers / platform builders / application developers
Industry
Other: Security needed in an emerging environments, need to build trust.
Organisational interoperability suggested as new focus for FIPA.
Start small, let things happen.
Step 1: Start with road map of FIPA how it can be used, limitations, of where FIPA was in the world. Business more interested in specifications.
Assess where standards that cover agents to identify gaps some of these may be so big to influence them. Need critical subjects / application domains.
Step 2: reach out & engage community (within FIPA, outside FIPA) to get support, FIPA could issue for call from community for work-plans and commitments? Do a questionnaire, too late to hand-out at AAMAS is too late but could it add onto the web-site?
How to collaborate: during meetings, in between meetings, virtual meetings
How to meet? Setting meetings not tied to conferences is good for industry but academic would prefer to co-locate FIPA meetings at conferences.
Some procedural things need to be sorted out.
How to become a member?
Who are the FIPA contacts?
Need to collect a new email address at the main FIPA meeting at AAMAS. Next meeting is for inform audience about the start rather than to give no information up-front and to brain-storm. Lead from front give a straw man.
~ 40 participants
A short overview of what happened to FIPA in the last year or so,
Explain the new constellation (do we have new statutes???),
Anticipate membership fee (can we agree on this),
Announce upcoming meetings.
Next FIPA meeting co-locate with Agentlink meeting in Budapest
Jim (Odell) will update us on US?
What about Japan and/or Australia/New Zealand?
Anticipate key technical aspects the new FIPA IEEE will be focusing o
What does IEEE provide at what cost to FIPA? Visibility via the IEEE organisation, management.
How to move specifications to IEEE?
How does voting work? IEEE candidate recommended proposals (FIPA voting), IEEE vote on standard.
When is something appropriate to standardise? Implement first then standardise, what is optimum point? Also impacts on organisation introduce concepts of WGs or study groups that examine things to standardise.
Technical advisory of IEEE has several activities that cover Agents.
Why go to IEEE not OMG or OASIS? IEEE was very enthusiastic to have FIPA? Provides awareness of technology? If we fail to standardise agents they will fail Charles Petri quote.
What does meaning FIPA compliance? What will happen to the original standards? they will remain public despite IEEE standards not being publicly assessable , but only accessible to the (sub-)group of members.
What about new proposals they will start being public as proposals, in limbo, who do they belong it WG?
What do we do if IEEE rejects a standard? Keep advertising; review whether or not the reasons are valid, decide to try again or not.
Need to get a position on the IEEE IPR issues if standard not accepted.
What about the name? parts of FIPA acronym not the focus anymore. N.B. Also IEEE not about EE so much anymore? Check what name is currently on the FIPA web-site?
GGF: driven by application Virtual computer, now virtual organisations.
Standards must have a driving force? Difficult to get agreement. In GGF have strong rules to form a group and to move things to a standard, so FIPA could adopt similar rules. Drive from needs.
Why come to FIPA and not the other standardisation? Something to do with agents. Got to make a clear point of identifying this.
Work between meetings? Yes
Quality of standards: Quality of some FIPA specifications was questioned but others refuted this. Is it worth industry trying to standardise the FIPA specs? Yes. Not everything may have a reference implementation, e.g., methodology. Maybe rules about this need to be formulated or let market-placed decide? Experts who work on proposal should be given free reign to propose something. Others are not experts. Group may not include all the experts? Could identify experts to review it that weren't working on it. N.B. can be difficult to engage the community if they feel excluded. Some people must respond to review in a formal way must be able to defend this.
ACL standards have well-known flaws FIPA was aware of these. Because of the structure, only few people committed to do these. Need procedures to make sure this is sufficiently resourced this is often a source of poor quality standards.
Purely industry model for developing standards may not work? May not include all the stake-holders or a critical mass.
Can FIPA survive? Support from industry dried out? What can convince industry to participate in FIPA? What are the relevant compelling problems / business cases? Other organisations are good at allowing small piecemeal solutions. Make it easy to build on a part of specifications. Need business use-cases in particular market-places.
Need producers and consumers (these are different?) in SA. Company X develops Tech B for customer Z.
At this meeting: industry representatives here are mainly from research parts of these organisations not from business departments.
Small companies have closer links between developers and business users. Customers do not need to be aware of the technologies, only that they can solve a problem. Standards help justify the particular technology approach proposed. Telcos / vertical markets have they own vertical technology stacks, e.g., not OMG? Need to analyse vertical domains.
What drives OMG? Can we learn from this? OMG has some application domain groups, E.g., healthcare, and many others. Need these people within FIPA. At OMG, healthcare end-users are driven by insurance companies, product activities.
How to engage industry? Develop some example applications. FIPA tried that. Was very political about who got the tender to build this.
Lot of things depend on critical mass and that takes time. Need to start somewhere smaller from some specific seed needs just one thing?
People had confidence from FIPA to produce and use FIPA platforms - Is that still there? Standards were a signal to take something seriously. FIPA helped.
Survey / classification of user groups on the FIPA Chat email list.
Don't use standards just because they are there. People don't use FIPA, they use software tools that implement the standards such as JADE.
What are the drivers to make standards? Interoperability Others?
Have there been trials of getting different FIPA implementations to work together? Yes, FIPA trials, some projects such as FACTS, Agentcities etc. NIST suggested that they could help with conformance testing and helping with specifying procedures.
How to get some of these people back working from other standards group? How to get the health people over? Bridge
Will FIPA define something what is not an agent, scope it? This is a philosophical question? Standards can only define observable qualities semantics of ACL are not normative.
How many do commercial agent platforms comply with FIPA?
Agentlink changed its way of working from SIGS to adhoc on-demand TFs (Technical Forums). However, this can leads to fragmented technical work.
Put slides on web-site and email other discussion material on the web-site, start a WIKI to get collaborative input.
Solicit call for input in particular business cases and technical working groups within FIPA and outside FIPA: email, web-site,
Start a road-map document and circulate? Agentlink roadmap? Talk to Agentlink about case if Agentlink IV does not happen.
Draft Discussion document for policies and procedures? Some expectations of realistic time-out to halt things.
Agenda for next meeting
Consider as a discussion forum to bring forward best practice. Industry needs standards. Be realistic.