Report on the 2nd Meeting of AgentLink SIG on
Intelligent Information Agents
=============================================

Matthias Klusch

DFKI GmbH, Deduction and Multiagent Systems Lab, Saarbruecken, Germany.
klusch@dfki.de


The AgentLink SIG on Intelligent Information Agents (I2A) aims for
providing resources and a unique forum to initiate and foster collaborations among
researchers and developers across Europe in the area of information agent technology.
Its main objectives are the following:
- Provision of a forum for advanced R&D on I2A in Europe.
- Initialize collaborations and joint projects among SIG participants and 
  relations to other AgentLink SIGs.
- Set up R&D projects in european research programs, like IST, and EUREKA.
- Support related events, like workshops and conferences in I2A area.

The I2A SIG had its kick-off meeting in late September last year in Brussels, Belgium,
with 27 participants from academic area and industry which considered it as 
a success and highly stimulating event. To facilitate convenient communication among 
SIG participants an open public majordomo mailing list infoagents@gmd.de, which anyone interested in
is invited to submit contributions, as well as a (protected) shared working space
(http://bscw.gmd.de/bscw/) for exchange of more sensitive documents, like pre-proposals 
on joint projects among SIG participants, have been created and maintained.
Several contacts to related research communities, like that for database and information systems,
and information retrieval have been established. Information on any activity of the I2A SIG 
may be found in the SIG's pages on the Web at http://www.dfki.de/~klusch/i2a-SIG.html.


The Second Meeting
******************

The second SIG meeting was held in London, UK, on April 21 - 22, 1999. Its purpose
was threefold: Firstly, to report on the proceedings of collaborations which have been
initiated and set up among SIG participants in the meantime since the first meeting, 
secondly, to present new advanced work of R&D groups active in the I2A area, and finally,
to discuss potential working groups within the SIG focussing on few main topics and
significantly relevant issues of research and development of information agents. 
Accordingly, the agenda of the two-day meeting comprised two sessions for selected presentations
of projects and groups, one session for software demonstration, a keynote talk, 
one common session with the AgentLink SIG on Agent-Mediated Electronic Commerce (AMEC),
and four panel sessions for working groups.

After a brief welcome and some general remarks from the SIG coordinator Matthias Klusch, 
the morning session of the first day started off with a keynote talk of Norman Sadeh
from the European Commission, DGXIII, about  the pro-active 
initiative Universal Information Ecosystem (UIE) in the european research program IST 
(Information Society Technologies). He emphasized the intent of the UIE initiative 
to encourage in particul all kinds of 'revolutionary' projects towards R&D of new technologies 
used by highly dynamic population of infohabitants of emerging, adaptive, scalable and open
information 'ecosystems' including future cyberspace (http://www.cordis.lu/ist/fetuie.htm).
This talk attracted much interest and stimulated intense discussion on project ideas of
attendees partially even late night the same day. The session continued with a block of four selected
presentations on industrial and research collaborations as well as perspectives and 
concrete proposals for joint projects among SIG participants.

Vadim Ermolayev (Uni Zaporozhye, Ucraine) presented motives, perspectives and potential
of R&D on intelligent agents for organisational management and teaching in eastern europe,
especially the Ucraine. Dieter Fensel (Uni Karlsruhe, Germany) contributed a short description of the 
ESPRIT project called IBROW3 on Internet-based brokering services for knowledge reuse
including, for example, the development of ontology-mediated facilities to
bridge heterogeneous information systems, an ontology editor, and a fairly plain
graphical user interface of OntoBroker system. Chihab Hanachi (Uni Toulouse, France) 
reported on first steps towards a joint project with another SIG participant, 
the DAKE Center of the University of Keele, UK, on collaborative agent systems facilitating
cross-organizational workflow among different companies/corporations. To end the 
presentation session, Klaus Fischer (DFKI GmbH) gave an outline of the so-called CASIMA
project which deals with research and development of an agent-based pan-European intelligent information
and trading network infrastructure for agriculture industry; the project is a collaborative
effort of six AgentLink members. 

In the afternoon, two panel sessions on the key topics of two working groups on agent-based 
mediation among information systems, data and knowledge management, and communication,
coordination and collaboration (C3) among information agents were held and chaired by Dieter Fensel
and Paolo Petta (Austrian AI Institute, Austria), respectively. The first panel started off with 
an overview from Sonia Bergamaschi (Uni Modena, Italy) about past and current R&D activities 
towards utilization of coupled AI/Database techniques for integration of heterogeneous information sources
including known approaches and systems like TSIMMIS, InfoSleuth, InfoMaster and MOMIS.
Leon Sterling (Uni Melbourne, Australia) gave an account to XML-based integration
of ontologies and sketched some of the work of his group on single, adaptive information
agents for information retrieval in the Internet, like SportsFinder.
Hans-Juergen Mueller (Deutsche Telekom AG, Germany) presented his group's work on 
composition and configuration platforms for Internet-based applications, in particular
how to support effective knowledge management based on the concept of a so-called 'knowledge factory'.
Hans Weigand (Uni Tilburg, Netherlands) shortly described the most prominent R&D projects
carried out by his lab, like TREVI, MEMO and DECOMATE/CIA, in the areas of basic language
ontology, electronic commerce, and use of description logic for intelligent integration 
of information, respectively. Finally, Richard Benjamins (Uni Amsterdam, Netherlands)
briefly presented an approach for intelligent web page annotation and information search
which was very controversily debated. Due to the discussion and comments on given
presentations the following three key topics and issues were identified for upcoming joint
effort by SIG participants in that working area:

(1) Methods and standards for interoperability among heterogeneous, distributed
    information sources of different corporations,

(2) Intelligent information brokering and matchmaking; this includes, for example,
    agent capability description languages and tools, (re-)Use of common, minimal or 
    domain-specific ontologies and metadata tools, and

(3) Agent-based support of (semi-automated) knowledge management in different enterprises
    and corporate virtual private networks or Intranets.

The second panel session was devoted to the working area of communication, coordination and
collaboration (C3) among information agents. The panelists were Paolo Petta, Franco Zambonelli 
(Uni Modena, Italy), Luigi Serafini (IRST Trento, Italy) and Monica Divitini (Uni Trondheim, Norway).
Coordination was seen as the sum of conventions and commitments in a multi-agent environment,
means as the process to ensure coherency (unity) of multi-agent systems. The presentations
given by the panelists comprised, for example, the relation of C3 to CSCW in real-world context, 
the analytical distinction among basic C3 capabilities of an agent, the role of artefacts 
and possible approaches for metrics of C3 as well as the increasing need of learning and
dynamic management of coordination in steadily changing collaborations of information agents.
In general, the discussion mainly suffered from the fear of potential overlap with the objectives 
of the AgentLink SIG on Multiagent coordination and control. At the end several key topics and 
issues to be addressed by joint efforts uniquely in the I2A SIG have been worked out; the main topics are

- Communication methods and standards, like intelligent, agent-based CORBA compliant
  service facilities for coordination and collaboration, basic shared ontology for C3,
  and advanced interfaces for standardized agent communication language (FIPA ACL),

- Formal coordination models and reference scenarios for collaborative information agents,

- Learning and metrics of quality of C3 in the context of cooperative information agents and systems.
  
The second day of the meeting started off with a short classification of intelligent information
agents by the SIG coordinator who even briefly sketched some project proposals as one already accomplished 
goal of this meeting after the first day. The morning session was divided into one presentation
from Martin Schneider (Siemens AG, Germany) describing the work of his group on a
client/server-based agent broker for project matching, and two software demonstrations.
The first demonstration was given by Pascal van Eck (Vrije Uni Amsterdam, Netherlands), standing in
for his colleague Catholijn Jonker, on an intelligent Website architecture for information agents.
Even interesting was the second live demonstration which adressed collaborative user profiling 
performed by the CASMIR agent system which has been developed at Uni Salford, UK.

The afternoon was devoted to the discussion of two more working groups, the third one focussing
on the areas of Human-Agent interaction (HAI) and interfaces for information agents, the fourth one
dealing with the relation of information agents and electronic commerce. The panel session on 
the third working group was chaired by Jeremy Pitt (Imperial College, UK), the panelists were 
Daniela D'Aloisi (FUB Rome, Italy),  Mathias Bauer (DFKI GmbH, Germany), Vadim Ermolayev and
Ian Dickinson (Hewlett-Packard Bristol, UK).
After each panelist briefly presented the work of his group and even some ideas on the topics 
to be addressed by the working group, the following main issues and problems have been worked 
out in a discussion:

- Convenient inspection of agents by user, more transparency of agents' activity, clear
  impact of user feedback given in more native language, gestures or media input

- Standard of interface design for different kinds of information agents

- Need-driven not technology-lead products for agents in interfaces (which are not equal to
  complex agent architectures) and to cope with the danger of raised expectations of
  anthromorphisation.

- Shared context between user and intelligent interface agent including a common ontology for 
  meaningful understanding in both directions, in particular to avoid that any single agent 
  deployed on the Web will be just an individual curiosity to the user.

Since the last panel session intended to discuss one obvious key topic of the AgentLink SIG on agent-mediated 
electronic commerce, it was mutually agreed to held a common session of both SIGs. This common session
was chaired by Walter van de Velde (Starlab Ltd., Belgium), the panelists were
Markus Schwehm (Uni Stuttgart, Germany), Eugenio da Costa (Uni Porto, Portugal), Mikail Matskin (Uni
Trondheim, Norway), Ygge (Uni Karlskrona, Sweden) and Klusch (DFKI GmbH, Germany).
The discussion attracted much interest for several different reasons and was sometimes even a bit
emotional concerning the evaluation of future vision and current state of the art of 
personalized agents doing everydays business on the Web including homebanking and online shopping for the average user.
Some main questions and issues discussed by the panelists were, for example, how to increase the
awareness of users regarding the potential of agent-based computing not only in business-to-customer 
electronic commerce, the mutual impact of electronic and material commerce, profit
maximization by online retailers vs. users' cravings for fun and social events, e.g.,
collaborative shopping (adaptive ShopBots for WebTV, etc.), notions of rationality not only in 
traditional terms of micro-economic based decision-making behavior of trading agents.
More detailed information on the outcome of this panel discussion will be available in a separate
contribution by the panel chair. The status of a joint working group among AMEC and I2A SIG remains
to be clarified. At the plenary meeting on April 23 the preliminary results of the meeting
of every SIG and some additional news and admistrative issues related to AgentLink as a whole have been
reported.


Outcome of the Meeting 
**********************

The second I2A SIG meeting had 43 registered participants, means a significant increase of interest 
since the kick-off meeting. Though, again twice as much attendees came from universities than from industry.
But the large number of participants even showed the problem of satisfying different needs of active and
passive attendees at the same time during a packed two-days meeting. 
However, the meeting was widely regarded as a successful event. In fact, much productive 
efforts towards collaboration and joint projects to propose for the recently started european IST
research program have been put together, especially by fostering and establishing new
inter-disciplinary contacts among attendees of this meeting in London. 

The core activities of the two-day meeting were strongly determined by the panel sessions on
the working groups intended to focus the work of the I2A SIG on a few important topics and 
to produce significant results theoretically and in practice in reasonable time. Actually, three 
important subareas of information agent technology have been discussed for future joint efforts 
among SIG participants, thereby forming kind of domain-specific working groups, at the meeting. 
Those groups focus on

- Agent-Based mediation among information systems, data and knowledge management (WG-1)

- Communication, coordination and collaboration among information agents (WG-2)

- Human-Agent interaction and interfaces for information agents (WG-3)

In mid-term it is expected that each of these working groups will participate in national 
and european research programs, especially by joint projects among SIG participants
in the respective working areas. First steps were taken already to accomplish this objective.


Future Action Items
*******************

To further improve the productivity of upcoming I2A SIG meetings it is intended, for example,
to restrict the number of participants to a much lesser amount, and to try to 'recruit' more industrial 
attendees. Besides, discussion on current or potential joint projects within the
SIG working groups, means collaborative work and efforts among SIG participants in the mentioned working
areas, will be the dominant factor of the next meeting. 
The next I2A SIG meeting will be held in Barcelona, Spain, in late September this year (probably
September 20 - 22, 1999) and co-located with the SIG on agent-mediated electronic commerce,
methodologies and software engineering for agent systems as well as the SIG on agent-based 
social simulation which had its kick-off meeting in London. 

The technological roadmap of the I2A SIG for the forthcoming period including that of each of 
its three working groups will be available as a kind of comprehensive green paper in late summer. 
In addition, it is intended to publish the preliminary results from substantial efforts within 
the I2A SIG as a journal article or book at the end of this year. 
More up-to-date information about the I2A SIG, to obtain reports on R&D of SIG participants and
the SIG meetings, check out the SIG's home page at: http://www.dfki.de/~klusch/i2a-SIG.html