Download here - AgentLink

European Coordination Action for Agent-based Computing
News
Features
Responsive and Generative Music for Public Installations
Semantic Web Services with the Web Service Modeling Ontology (WSMO)
Trust: Challenges and Opportunities
Why Argue Dialogue Types and Argumentation in Agent Interaction
Answer Set Programming and Agents
AgentCities.ES: Spanish network of agent-based computing (2003-2005)
19Agent Research Overview
ISSUE 19 | November 2005 | ISSN 1465-3842
Agentlink Technical Forum III
Budapest, Hungary, September 15-17, 2005
Towards a Standard Agent to Agent Argumentation Interchange Format (AIF)
Agent-Oriented Software Engineering (AOSE)
Coordinating Agent Standardisation Activities (CASA)
Environments For Multiagent Systems (ENV)
Multiagent Resource Allocation (MARA)
Programming Multi-Agent Systems (PROMAS)
Self-Organisation in Multi-Agent Systems (SELFORG)

November 2005
AgentLink News CONTENTS
Terry R. Payne
University of Southampton
AgentLink Co-coordinator
Editor in Chief
Rebecca Earl
University of Southampton
AgentLink Publications
Administrator
Serena Raffin
University of Southampton
AgentLink Publications
and Web Coordinator
Julian Padget
University of Bath
Event Reports Editor
Agentlink Report
Challenges for Agent-Based Computing, Michael Luck,
Peter McBurney, Onn Shehory and Steve Willmott
Features
Responsive and Generative Music for Public Installations,
Mark d’Inverno, John Eacott and Fredrik Olofsson
Semantic Web Services with the Web Service Modeling
Ontology (WSMO), John Domingue, Dieter Fensel
and Dumitru Roman
Trust: Challenges and Opportunities, Nathan Griffiths
Agent Research Overview
Why Argue Dialogue Types and Argumentation in
Agent Interaction, Katie Atkinson and Trevor Bench-Capon
Answer Set Programming and Agents, Thomas Either
and Jürgen Dix
AgentCities.ES: Spanish network of agent-based
computing (2003-2005), Antonio Moreno
Standards Reports
Latest News from the Standardisation World, Monique
Calisti
Agentlink Technical Forum III
Towards a Standard Agent to Agent Argumentation
Interchange Format (AIF), Steven Willmott, John Fox,
Dan Grecu, Simon Parsons, Iyad Rahwan, Chris Reed,
Dave Robertson, Nicolas Maudet
Agent-Oriented Software Engineering (AOSE), Massimo
Cossentino and Juan Pavón
Coordinating Agent Standardization Activities (CASA),
Monique Calisti, Giovanni Rimassa, Omer Rana, Stefan
Poslad, Terry Payne and James Odell
Environments For Multiagent Systems (ENV),
Danny Weyns and Tom Holvoet
Multiagent Resource Allocation (MARA), Ulle Endriss
Programming Multi-Agent Systems (PROMAS)
Editorial
3
5
12
18
20
Mehdi Dastani and Jorge Gómez Sanz
Self-Organisation in Multi-Agent Systems (SELFORG),
Giovanna Di Marzo Serugendo, Marie-Pierre Gleizes,
Anthony Karageorgos
Event Reports
Semantic Web Days Munich 6-7 October, 2005,
Andrea Kulas and Alain Léger
The 1 st International Workshop on Context for Web
Services (CWS’05), Djamal Benslimane, Chirine Ghedira
and Zakaria Maamar
The 8 th Biennial Israeli Symposium on Foundations of
Artificial Intelligence (BISFAI), Claudia V. Goldman
E4MAS 2005: Environments for Multiagent Systems,
Danny Weyns, Eric Platon and Fabien Michel
International Workshop on Agent-Based Models for
Economic Policy Design (ACEPOL05), Herbert Dawid
The 2005 IEEE/WIC/ACM International Joint Conference
on Web Intelligence and Intelligent Agent Technology
(WI’05 and IAT’05), Pierre Morizet-Mahoudeaux
3 rd Workshop on Agents Applied in Health Care, at
the 19th International Joint Conference on Artificial
Intelligence (IJCAI 2005), Antonio Moreno
Agents in Space, Michael Fisher
1 st International Workshop on Security and Trust
Management (STM 2005), Sjouke Rauw
Project Report
Site Report
Business Decision Making using Multi-Agent
Systems, Nadia Yakounina
Books
Agent Events Calendar
26
34
ATHENA: Advanced Technologies for interoperability
of Heterogeneous Enterprise Networks and their
Applications, Klaus Fischer
36
38
39
Chris van Aart
Acklin B.V.
Book Reviews Editor
Dominic Greenwood
Whitestein Technologies AG
Features Editor
Stefan Poslad
Queen Mary University
Industry Editor
The development of Agent Technology has found a place in a variety of application domains, from the Semantic
Web to The Grid, from Service-Oriented to Peer-to-Peer Computing. Given the advances and improvements in both
hardware and software in recent years, the objectives of developers has evolved, from minimising the complexity
and size of code to maximising the modularity and re-usability of code. As systems scale, the number of networked
hosts continues to rise and the number of intelligent consumer devices increases, issues such as autonomicity
and semantic interoperation are becoming ever more signifi cant factors in the way software systems should be
developed.
In this, the 2005 Christmas issue of the AgentLink Newsletter, we include an excerpt from the recently published
AgentLink Roadmap, which examines the fi eld of Agent Technology and its current trends and key drivers. d’Inverno
reports on a novel use of agents for interactive music spaces in public areas, whilst Griffi ths reviews many of the
challenges inherent in modelling trust amongst agent communities. The third feature introduces WSMO; an EUfunded
effort for representing and modelling services within the Semantic Web.
The Third AgentLink III Technical Forum Meeting, co-located with CEEMAS, attracted researchers from all over Europe
to the beautiful city of Budapest to engage in three days of stimulating discussion on topics old and new. Five
of the seven meetings followed on from earlier Technical Forum meetings held in Rome or Ljubjana, whereas one of
the new meetings, CASA (Co-ordinating Agent Standardisation Activities), followed on from the 2004 Paris meeting
on “Agent Technology and Standardisation Activities - The Role of AgentLink”, which was reported on in Issue 17.
As the year closes, so will the doors of the AgentLink III offi ces. Whilst the future of AgentLink IV is still unknown,
events such as the the Technical Forum meetings clearly indicate a promising and certain future for Agent-based
research in Europe over the coming years. Its been a pleasure to work with the AgentLink team, and to collaborate
with such a dynamic and insightful community, and so from the offi ces of AgentLink III, we wish you Seasons Greetings,
and a fruitful 2006.
Terry R. Payne
AgentLink Coordinator
University of Southampton

AGENTLINK REPORT
Challenges for Agent-Based Computing
Michael Luck
University of Southampton
United Kingdom
mml@ecs.soton.ac.uk
Peter McBurney
University of Liverpool
United Kingdom
p.j.mcburney@csc.liv.ac.uk
Onn Shehory
IBM Research
Israel
onn@il.ibm.com
Steven Willmott
Universitat Politècnica de Catalunya (UPC)
Spain
steve@lsi.upc.edu
Introduction
In its brief history, computing has enjoyed
several different metaphors for the notion of
computation. From the time of Charles Babbage
in the nineteenth century until the mid-
1960s, most people thought of computation
as calculation, or operations undertaken on
numbers. With widespread digital storage and
manipulation of non-numerical information
from the 1960s onwards, computation was reconceptualised
more generally as information
processing, or operations on text, audio or
video data. With the growth of the Internet
and the World Wide Web over the last
fifteen years, we have reached a position
where a new metaphor for computation
is required: computation as interaction. In
this metaphor, computing is something that
happens by and through communication
between computational entities. In the
current radical re-conceptualisation of
computing, the network is the computer, to
coin a phrase.
In this new metaphor, computing is an
activity that is inherently social, rather than
solitary, leading to new ways of conceiving,
designing, developing and managing
computational systems. One example
of the influence of this viewpoint is the
emerging model of software as a service, for
example in service-oriented architectures.
In this model, applications are no longer
monolithic, functioning on one machine
(for single user applications), or distributed
applications managed by a single organisation
(such as today’s intranet applications), but instead
are societies of components.
How should we exploit this new metaphor
of computing as social activity, as interaction
between independent and sometimes intelligent
entities, adapting and co-evolving with one
another The answer, many people believe, lies
with agent technologies. An agent is a computer
program capable of flexible and autonomous
action in a dynamic environment, usually an
environment containing other agents. In this
abstraction, we have encapsulated autonomous
and intelligent software entities, called agents,
and we have demarcated the society in which
they operate, a multi-agent system. Agent-based
computing concerns the theoretical and practical
working through of the details of this simple twolevel
abstraction.
Agent Technology: Computing as Interaction
Technologies
Trends and Drivers
Related Disciplines
Related Techniques
Mathematical
Modelling
Logic
Programming
User
Interaction
Design
ormal
Methods
Grid
Computing
Self
Systems
Ambient
Intelligence
Uncertainty
in AI
Robotics
Artificial Life
Programming
Languages
Simulation
Organisations
Comple Systems
PeertoPeer
Computing
A Roadmap for Agent Based Computing
Biology
Economics
Game Theory
Philosophy
Compiled, written and edited by
Michael Luck, Peter McBurney, Onn Shehory, Steve Willmott and the AgentLink Community
In the sense that it is a new paradigm, agentbased
computing is disruptive. As outlined
above, it causes a re-evaluation of the very nature
of computing, computation and computational
systems, through concepts such as autonomy,
coalitions and ecosystems, which make no sense
to earlier paradigms. Economic historians have
witnessed such disruption with new technologies
repeatedly, as new technologies are created, are
adopted, and then mature.
Logic
Anthropology
Sociology
Service Oriented
Computing
Organisation Design
Political Science
Decision
Theory
Trust and Reputation
Coordination egotiation Communication
Softare
Engineering
Interoperability
Infrastructure
Marketing
Reasoning
and Learning
Semantic Web
The concept of an agent has found currency in
a diverse range of sub-disciplines of information
technology, including computer networks,
software engineering, artificial intelligence,
human-computer interaction, distributed and
concurrent systems, mobile systems, telematics,
computer-supported cooperative work, control
systems, decision support, information retrieval
and management, and electronic commerce.
In practical developments, web services, for
example, now offer fundamentally new ways of
doing business through a set of standardised
tools, and support a service-oriented view
of distinct and independent software
components interacting to provide valuable
functionality.
In the context of such developments, agent
technologies have increasingly come to the
foreground. Because of its horizontal nature,
it is likely that the successful adoption of
agent technology will have a profound, longterm
impact both on the competitiveness
and viability of IT industries, and on the
way in which future computer systems will
be conceptualised and implemented.
Challenges
Hardware and software have improved
significantly in performance and availability
over the six decades of modern computing.
As these changes have occurred, the
objectives of programmers have also
changed. Initially, most programmers
sought to minimise memory usage and to
maximise throughput or processing speeds in
their applications. With increasing availability
and lower costs of memory, and increasing microprocessor
speeds, these objectives became far less
important. Instead, by the 1970s and 1980s, the
object-oriented paradigm sought to maximise
the modularity and re-usability of code, and to
minimise post-deployment system maintenance.
However, these objectives too have become dated.
Partly, this is because the development of proven
November 2005|AgentLink News 19
3

AGENTLINK REPORT
OO methods and support tools has enabled the
objectives to be readily achieved, and indeed,
taken for granted, over the last two decades. More
importantly, however, the rise to prominence of
the Internet has led to a new understanding of the
nature of computation, an understanding that
puts interaction at its centre. In this context, the
agent-oriented paradigm has sought to maximise
adaptability and robustness of systems in open
environments.
It is here that one can see how a new technology
may be a disruptive force. By tackling a different
set of objectives, agent technologies address
different problems and different applications than
do object technologies. It is not simply that the
rules of the game have changed, but rather that
a different game is being played. In a world of
millions of independent processors interconnected
via the Internet and, through it, engaged in
distributed cognition, a software design team can
no longer assume that software components will
share the same goals or motivations, or that the
system objectives will remain static over time.
Systems therefore need to be able to adapt to
dynamic environments, to be able to configure,
manage and maintain themselves, and to cope
with malicious, whimsical or just plain buggy
components. The power of the agent paradigm is
that it provides the means, at the appropriate level
of abstraction, to conceive, design and manage
such systems.
Broad Challenges
For agent-based computing to support the visions
of the Semantic Web, ambient intelligence, the
Grid, autonomic systems, etc, considerable
challenges remain, both broad, over-arching
challenges across the entire domain of agent
technologies, and challenges specific to particular
aspects. The broad challenges are as follows.
• Creating tools, techniques and methodologies
to support agent systems developers.
Compared to more mature technologies
such as object-oriented programming, agent
developers lack sophisticated software tools,
techniques and methodologies to support the
specification, development and management
of agent systems.
• Automating the specification, development
and management of agent systems. Agent
systems and many of their features are still
mostly hand-crafted. For example, the design
of auction mechanisms awaits automation, as
does the creation and management of agent
coalitions and virtual organisations. These
challenges are probably several decades from
achievement, and will draw on domainspecific
expertise (for example, economics,
social psychology and artificial intelligence).
• Integrating components and features. Many
different theories, technologies and infrastructures
are required to specify, design, implement and
manage agent systems. Integrating these pieces
coherently and cost-effectively is usually a major
undertaking in any system development activity,
a task made more challenging by the absence of
mature integration tools and methodologies.
• Establishing appropriate trade-offs between
adaptability and predictability. Creating
systems able to adapt themselves to changing
environments, and to cope with autonomous
components, may well lead to systems
exhibiting properties that were not predicted
or desired. Striking a balance, appropriate
to the specific application domain, between
adaptability and predictability is a major
challenge, as yet unresolved either theoretically
or practically. Associated with predictability
is the requirement for practical methods and
tools for verification of system properties,
particularly in multi-agent systems that are
likely to exhibit emergent behaviour.
• Establishing appropriate linkage with other
branches of computer science and with other
disciplines, such as economics, sociology
and biology. One task here is to draw
appropriately on prior research from these
other areas and disciplines. Another task is to
avoid reinvention of existing techniques and
methods, whether by agent researchers or by
others. Awareness-building between areas
and disciplines, and coordination of research
and development activities, are essential if the
appropriate linkages are to established and
maintained.
Conclusions
Agent technologies can be distinguished from
other programming technologies on the basis of
their differing objectives. For agent technologies,
the objectives are to create systems situated in
dynamic and open environments, able to adapt to
these environments and capable of incorporating
autonomous and self-interested components.
How quickly agent technology is adopted by
software developers, therefore, will depend at least
partly on how many application domains require
systems with these characteristics. Considering
the domains receiving attention from agent
software development companies such as Agentis,
Magenta, Lost Wax or Whitestein (among
others), the main areas are currently: logistics,
transportation, utility management and defence.
Common to many of these domains are multiple
stakeholders or organisations linked in a network,
such as a supply-chain, and with mission-critical,
real-time processing requirements. In other
words, there are both functional and technical
requirements for these applications, a divide that
agent technologies are able to bridge.
In application terms, we are already seeing the
deployment of agent-like systems (in the areas
of pervasive computing, the Semantic Web,
P2P networks, and so on). In the longer term,
we expect to see the industrial development
of infrastructures for building highly scalable
applications comprising pre-existing agents
that must be organised or orchestrated.
However, making the transition from research
laboratory to deployed industrial applications
is indeed a challenge, and it will be important
to make scientifically sound business cases for
implementations and descriptions that work as
stimulators both for industry adoption and for
further research.
For commercial and industrial systems, agent
technologies must emerge from the laboratory
with a focus on business issues, on quality and
on convergence with existing and emerging
industrial technologies rather than innovation.
Here, safety, reliability and traditional software
quality measures are equally important, and
must all be addressed to achieve wider adoption.
In particular, we need agent solutions for
distributed, enterprise-wide environments with
exacting development requirements. This might
be achieved through transition approaches by
which existing systems can be upgraded with
a successively increased agent presence in a
seamless fashion. Wrapping legacy systems
within autonomous agents situated in a larger
multi-agent system is one approach that is being
tried, for example, in connecting new and old
telecommunications switches together seamlessly,
allowing legacy switches to be gradually replaced
without major disruption to the overall system.
More generally, the adoption of agent
technologies in business environments depends
on how fast and how well agent technologies can
be linked to existing and proven software and
software methods. Agent technologies should be
targeted at those application domains to which
they are best suited, augmenting traditional
techniques that should be used when agents
are not applicable or appropriate. Ultimately,
achieving this aim requires a commitment on the
part of both business and research communities
to collaborate effectively in support of more
effective solutions for all. Such a dialogue is
already underway.
This article is based on extracts from the AgentLink III Roadmap, “Agent Technology: Computing as Interaction”,
available from www.agentlink.org/roadmap
4
November 2005|AgentLink News 19

ALSO IN THIS SECTION...
Semantic Web Services with the Web Service Modeling Ontology (WSMO)
Trust: Challenges and Opportunities
FEATURE
Responsive and Generative Music for Public Installations
Mark d’Inverno, John Eacott and Fredrik Olofsson
University of Westminster
UK
M.dInverno@westminster.ac.uk
john@informal.org
f@fredrikolofsson.com
Introduction
In a recent poll conducted by Time Magazine,
Muzak was voted one of the worst ideas of the
last century. To give some indication of just how
bad, it was only one place below the marriage
of Lisa Marie Presley to Michael Jackson. In
another poll conducted by Mori, 17% of those
questioned voted Muzak as the thing they most
detest about modern life. It was considered a
raise pollution from which there is no escape.
However, with developments in algorithmic and
generative music, and design techniques from
intelligent agent technology we are investigating
new possibilities for creating music generating
devices that can respond flexibly to the requests
of groups of users. And this is more than just an
on/off button or graphic equalizer; it can provide
users with the power to turn off tracks, change
the density of the drums, transpose or invert the
melody, re-harmonise,
add effects, and more.
In fact, the opportunity
is there to build a whole
new musical sound
system that can, in
theory, respond to our
every musical desire.
In this article we will outline the design and
implementation of an environmental sound
installation that is able to respond to the requests
of users interacting together. If you want to
hear some of the music it generates and see it in
action to help decide whether this article might
be interesting enough to continue reading, then
check out this movie: http://users.wmin.ac.uk/
~dinverm/istreet/shortistreet.mov.
The output from the installation is generated
solely using algorithmic processes written by
some of Europe’s leading composers of generative
music, using the SuperCollider language. Because
this music is represented solely (or very nearly
solely) in terms of algorithmic structures, it
allows for much more sophisticated and subtle
operations to be performed on the musical
performance in real time. In the installations we
have implemented, musical output is controlled
by the collective history of text commands from
users’ mobile phones (by some distance the
most common mode of communication at the
University of Westminster). In this way, people in
a shared environment can now directly, and more
importantly, collectively influence their sonic
environment and collaboratively create their
mutually desired soundscape.
The Installation
Of course there are countless mechanisms for
providing environmental or ambient sound, but
they are typically more closely entwined with the
natural environment (consider the most obvious
example of the wind chime, for example) and
not under the direct, explicit control of situated
users. It is now a real possibility to provide users
with the ability to collectively or individually
control their sound and music environment, and
with sufficient flexibility to mitigate boredom. At
the very least we wish for our work to highlight
It is now a real possibility to provide users with
the ability to collectively or individually control
their sound and music environment.
the possibilities users have to interact with their
sonic environment such that Muzak will become
simply unacceptable.
In order to offer music which is deeply
interactive, and that offers genuine scope for new
and personalised forms that reflect the tastes of
users situated within a common space, the sound
must be generated in real time using algorithmic
structures. Algorithmic music can be manipulated
very explicitly. Not only is it a simple matter to
reduce tempos or transpose particular parts the
piece up or down a tone, it can be achieved much
more effectively with algorithmic representations
of music rather than sampled and recorded ones.
Of course manipulating samples has made lots of
people very rich, but samples are not as readily
manipulated as code.
Just to get a flavour, here is what a pretty decent
sounding snare drum looks like.
Instr([\istreet, \synthSN],
{arg amp, pan, release, filter;
var out, env;
out= Resonz.ar(PinkNoise.ar,filter,0.5,2);
env= Line.kr(1, 0, release, amp);
Pan2.ar(out, pan, env); },
#[\amp, \pan, \release, \filter] );
Of course we wanted users to interact with the
installation space in the most comfortable way
and without inhibition. It didn’t take long to
come up with the notion of mobile phone text
messaging as an interaction metaphor. As long
as the user commands used to interact with the
software are kept simple and short, and problems
with spelling, typos, case and so on are filtered out,
it proves to be a successful means of interaction.
This was also partly helped by Vodafone, who
provided free handsets and SIM cards!
The set of pre-defined user commands was visually
projected into the space.
In the future there will
be scope to build a
mechanism that enables
users to dynamically
create a language for
interaction with one
another and with the installation, but this was
simply beyond the scope of the first prototype.
We also decided to use a mixture of words with
different semantics, some with explicit semantics
and some with ambiguous semantics, and also
include both emotional and non-emotional
words in order to investigate patterns of use. A
team of evaluators was employed to report on the
reaction and satisfaction of both participants and
non-participants in the space.
An additional motive for displaying user
commands was that it led to more interaction
between users and a collective sense of
expectation. All text commands were queued
in a buffer and executed simultaneously every
16 bars. Once the commands were executed the
buffer would be cleared and the process started
again. Executed commands were displayed; the
current output of the system being a function of
that command history. Future commands were
displayed separately with a timer that indicated
the time left until execution, which helped build
the sense of expectation.
November 2005|AgentLink News 19
5

FEATURE
The building blocks of the output came from
original algorithmic and generative compositions
commissioned from some of Europe’s leading
composers in the field. In order to ‘fuse’ these
styles in a meaningful way, the composers were
asked to use certain constant parameters: that it
be beat-based; structured into bars of 4-4 time;
with 10 tracks each relating to either a type
of instrument or a musical role; and that it be
structured into 4 different 16 bar passages (that
might, say, be mapped to intro, verse, bridge,
chorus) with some nominal structure between
the passages. These styles took the form of
compositions written within a common style
framework. It was absolutely necessary to have
a common framework for writing pieces (styles)
for this composition to ensure that intelligent,
musically sensitive operations could be made at
playback, especially when it came to combining
these styles for performance.
Global system parameters allow information
concerning the current output to be stored
such that decisions on how to alter the music
according to some user-command can be made
more effectively. These parameters affect the basic
sound as defined by the styles. We can have as
many global variables as we like in order to make
intelligent decisions on how to interpret any new
command. In the current system, these include
tempo, key, reverb, delay, pitch effects (high and
low), sampling rate, phaser effect, warp effect (a
cut-up effect) and pan. Each of these variables has
a predefined system default value.
A track can be thought of in the very traditional
sense of a track in a recording studio, or as in
standard sequencing. Each style has exactly ten
tracks. Moreover, we have pre-specified the role
and function of each of these tracks and given
them names as indicated below. We have also
associated each track with a “level” paramenter
which determines the “influence” a tack has on
the overall sound. This information can be found
in the table below.
A blend is a listing of which tracks are currently
associated with which styles, and thereby can be
viewed a snapshot of what is heard. To gradually
switch between different blends of styles we
found a simple, effective and musically satisfying
solution: Viewing the ten tracks as a stack, so that
styles could be pushed around in discrete steps
according to their levels. This retained a flavour
of the current structure whilst introducing new
elements from a different style.
Examples of some user command call methods
include simple operations such as muting or
unmuting tracks, transposing tracks and changing
tempo to adjust the complexity of the melodic or
harmonic possibilities for a given tune or chord
sequence as listed below:
• Increase density un-mutes the highest-level
muted tracks.
• Decrease tempo reduces the beats per second
by a given amount.
• Transpose tracks transposes melodic tracks
according to an array of order and amount.
• Low pass applies a low-pass filter to rhythm
tracks.
• Melodic filter progressively reduces the set of
allowable notes.
In total, there are twenty commands available
with different functions. Eight of them
correspond to styles composed in a certain genre,
their use adding elements from these styles to
the soundscape in the way described above. For
example, the command “Dark” emphasizes the
lower frequencies of the music, removing the
higher frequencies and diminishing focus on the
tempo to give the music a darker timbral quality.
When commands such as Dark are repeated, the
music is increasingly affected until an absolute
pre-defined lower bound is reached.
Final Thoughts
We are interested in designing software
architectures for sound that enables users
to interact, create and compose their own
sound landscapes. Our prototype architecture,
implemented in several sites around the UK and
Level Track Name Role
5 1 Percussion specific genre effect
4 2 Samplebeat sampled sound
3 3 Bass drum typically every beat
2 4 Snare off beat, patterns, fills
1 5 High hat key in defining style
1 6 Bass bass riff
3 8 Pad ambient backing
2 7 Chord harmonic sequence
4 9 Melody2 generic melody
5 10 Melody1 style related melody
Sweden, has been extensively evaluated by a team
of sociologists who found that the installations
essentially became a playground for students to
talk about music.
There are many potential avenues to explore new
practical and theoretical possibilities including the
user interface perspective (dynamically evolving
language) and more sophisticated techniques
drawn from AI, intelligent agents and generative
music. We intend to investigate the mechanism
by which users annotate their compositions in
order that user-commands and system-methods
can make more intelligent and musical use of the
structures. This work also suggests new avenues
of research exploring compositions consisting
of not simply a single linear piece, but ideas
and concepts that can be merged, tweaked and
manipulated by users in real time to determine
the actual performance.
Some of the continuing project goals are as
follows:
• To investigate the relationship between
algorithmic manipulation of music and sound
and individual/social emotional response.
• To map the space of possible mechanisms
for users to exert more flexible control over a
responsive sonic landscape.
• To create social spaces where users collectively
compete, discuss and evaluate the aesthetics
of the music and its responses.
• To provide a forum for algorithmic
composition and formulating challenges to
existing notions of musical composition.
• To develop the means for users to converse
and interact with musical performances in
real time.
We have provided an infrastructure where users
can explore the possibilities of blending music,
tracks and instruments, control the complexity of
melodic or harmonic sequences, and in general,
interact with the music for both aesthetic and
educational purposes. Potential application
environments for this project include nearly all
shared spaces, from the shopping mall, to the
office, to pubs and clubs, film and media. We are
also planning an online version of this installation
to allow users to “jam” with each other remotely
using our system. This should be available in
2006.
Further Reading
David Cope, Virtual Music: Computer Synthesis
of Musical Style, MIT Press, 2001.
John Eacott and Mark d’Inverno, ‘Embedded
intelligent music or ihifi the intelligent hifi’,
Digital Creativity, 14(2), 2003.
James McCartney, SuperCollider audio synthesis
environment, http://www.audiosynth.com, 2005.
6
November 2005|AgentLink News 19

FEATURE
Semantic Web Services with the Web Service Modeling Ontology
(WSMO)
John Domingue, Open University, United Kingdom, j.b.domingue@open.ac.uk
Dieter Fensel, DERI, Ireland and Austria, dieter.fensel@deri.org
Dumitru Roman, DERI Innsbruck, Austria, dumitru.roman@deri.org
In recent years, Web technologies have become
very relevant for agent technologies. For the
agent community, Web service and Semantic
Web technologies have the potential to increase
interoperability, to make use of Web standards,
and provide a link between the agent paradigm
and the newly emerging paradigm of Service
Oriented Computing.
In this context, the description of Web services in
a machine-understandable fashion is expected to
have a great impact in areas such as e-Commerce
and Enterprise Application Integration through
enabling dynamic and scalable cooperation
between different systems and organizations. Web
services provided by co-operating businesses or
applications can be automatically located, based
on another business or application needs. They
can be composed to achieve more complex, valueadded
functionalities, and cooperating businesses
or applications can interoperate without prior
agreements or custom codes. This leads to more
flexible and cost-effective integration with the
potential to create dynamic, scalable and costeffective
marketplaces and eCommerce solutions.
This has driven recent research efforts toward socalled
Semantic Web services which enrich Web
services with machine-processable semantics.
The Semantic Web and Web services are
envisioned as the enabling technologies for the
next generation of Web applications. The former
aims at enhancing the machine-readability of
WSML WG
A Rule-based Language for SWS
WSMO WG
A Formal Lenguage for WSMO
Web content, with ontologies identified as the
key technical building block. The objective of the
latter is to enable distributed computation over
the Internet by means of automated and dynamic
discovery, composition, and execution of services,
thus providing a new technology for Web-based
systems engineering. The current Web service
technology stack enables the exchange of messages
between Web services (SOAP), describes the
technical interface for consuming a Web service
(WSDL), and supports the advertisement of Web
services in registries (UDDI). However, these
technologies do not include explicit descriptions
of the functionality of a Web service. Moreover, the
existing descriptions are represented syntactically
and thus do not characterize the meaning of
the information to be interchanged. Semantic
Web services apply Semantic Web technology
to Web services which raises the level of this
discourse. More specifically, through the use of
semantic description frameworks, Semantic Web
services will support the provision of intelligent
mechanisms for the discovery, composition,
contracting, and execution of Web services.
To this end, the Web Service Modeling Ontology
(WSMO) provides the basis for Semantic Web
services by specifying a fully fledged framework.
This begins with a conceptual model, continues
with a formal language to provide formal syntax
and semantics for the conceptual model, and
is completed with an execution environment
that glues together the components that use
WSMX WG
A Conceptual Model for SWS
An Execution Enviroment
for WSMO
Figure 1. The relationship between the WSMO, WSML and WSMX Working Groups, that form a framework for
Semantic Web Services.
the language to enable automation of services.
The WSMO is developed in the context of
the WSMO Working Group [1], as part of the
SDK cluster [2]. Through alignment with key
European research projects in the Semantic Web
service area, the WSMO aims to further the
development of Semantic Web services through
work on a common architecture and platform,
and further standardization in the area of Semantic
Web service languages. As shown in Figure 1, the
WSMO Working Group includes the WSML
Working Group [3], which aims to develop the
Web Service Modeling Language (WSML) that
formalizes the WSMO, and the WSMX Working
Group [4], which aims to provide an execution
environment and reference implementation for
the WSMO. Figure 1 depicts these relationships.
WSMO Design Principles
The WSMO provides ontological specifications
for the core elements of Semantic Web services.
In fact, Semantic Web services are an integrated
technology for the next generation of the Web that
combine Semantic Web technologies and Web
services, thereby evolving the Internet from an
information repository for human consumption
into a world-wide system for distributed Webbased
computing. Therefore, frameworks for
Semantic Web services must integrate basic Web
design principles, both those defined for the
Semantic Web, and those defined for distributed,
service-orientated computing over the Web. As
a result, the WSMO is based on the following
design principles:
• Web Compliance
The WSMO inherits the concept of URI
(Universal Resource Identifier) for unique
identification of resources as the essential
design principle of the World Wide Web.
Moreover, the WSMO adopts the concept
of Namespaces for denoting consistent
information spaces, supports XML and other
W3C Web technology recommendations, and
supports the decentralization of resources.
• Ontology-Based
Ontologies are used as the data model
throughout the WSMO, implying that all
resource descriptions and all data interchanged
during service usage are based on ontologies.
Ontologies are a widely accepted state-ofthe-art
means of knowledge representation
and have thus been identified as the central
enabling technology of the Semantic Web.
The extensive use of ontologies allows
semantically enhanced information processing
and support for interoperability. The WSMO
November 2005|AgentLink News 19
7

FEATURE
supports the ontology languages defined for
the Semantic Web.
• Strict Decoupling
This denotes that WSMO resources are
defined in isolation, meaning that each
resource is specified independently without
regard to possible interactions with other
resources. This complies with the open and
distributed nature of the Web.
• Centrality of Mediation
As a complementary design principle to strict
decoupling, centrality of mediation addresses
the handling of heterogeneities that naturally
arise in open environments. Heterogeneity
can occur in terms of data, underlying
ontology, protocol or process. The WSMO
recognizes the importance of mediation for
the successful deployment of Web services by
making mediation a first class component of
the framework.
• Ontological Role Separation
Users, or more generally clients, exist in
specific contexts which will not always be an
exact match with available Web services. For
example, a user may wish to book a holiday
according to preferences for weather, culture
and childcare, whereas available Web services
may only cover airline travel and hotel
availability. The underlying epistemology of
the WSMO differentiates between the desires
of users or clients and the functionality of
available services.
• Description versus Implementation
The WSMO differentiates between the
descriptions of Semantic Web service elements
(description) and executable technologies
(implementation). While the former requires
a concise and sound description framework
based on appropriate formalisms to provide
concise semantic descriptions, the latter is
concerned with the support of existing and
emerging execution technologies for the
Semantic Web and Web services. The WSMO
aims at providing an appropriate ontological
description model, and complaince with
ONTOLOGIES
Figure 2. WSMO Top-level elements.
GOALS
MEDIATORS
existing and emerging technologies.
• Execution Semantics
In order to verify the WSMO specification,
the formal execution semantics of reference
implementations such as WSMX, as well as
other WSMO-enabled systems, provide the
technical realization of the WSMO.
WSMO Top-level Elements
As depicted in Figure 2, the WSMO [1] consists
of four main elements for semantically describing
Semantic Web services: (1) ontologies that
provide the concepts and relationships used
by other elements, (2) goals that define the
users’ objectives, i.e. the (potential) problems
that should be solved by Web services, (3) Web
service descriptions that define various aspects of
a Web service, and (4) mediators which bypass
interoperability problems.
Ontologies provide the formal semantics for
the terminology used within all other WSMO
components. A set of non-functional properties
are available for characterizing ontologies; they
usually include the DC Metadata elements.
Imported ontologies allow a modular approach
to ontology design and can be used if no conflicts
arise between the ontologies. When importing
ontologies in realistic scenarios, some steps for
aligning, merging and transforming them are
needed in order to resolve ontology mismatches.
For this reason, ontology mediators are used.
Concepts constitute the basic elements of the
agreed terminology for some problem domain.
Relations are used to model interdependencies
between several concepts (and respectively
instances of these concepts); functions are special
relations with a unary range and an n-ary domain
(parameters inherited from relations), where the
range value is functionally dependent on the
domain values, and instances are either defined
explicitly or by a link to an instance store, i.e., an
external storage of instances and their values.
The WSMO Web service element provides the
means to describe services through the following
WEB SERVICES
characteristics: The non-functional properties
and imported ontologies for Web services play a
role that is similar to that found in the ontology
class with the minor addition of a quality-ofservice
non-functional property. An extra type
of mediator to deal with protocol and process
related mismatches between Web services, is
also included. The final two attributes of the
definition are the two core WSMO notions for
semantically describing Web services: a capability,
which is a functional description of a Web service,
describing constraints on the input and output
of a service through preconditions, assumptions,
postconditions, and effects; and service interfaces
which specify how the service behaves in order
to achieve its functionality. A service interface
consists of a choreography which describes the
interface for the client-service interaction required
for service consumption, and an orchestration
which describes how the functionality of a Web
service is achieved by aggregating other Web
services.
A Goal as defined in the WSMO specifies the
objective that a client has when consulting
a Web service. In particular, aspects relating
to user desires with respect to the requested
functionality and behavior. Ontologies are used
as the semantically defined terminology for goal
specification. Goals model the user view in the
Web service usage process and therefore are a
separate top level entity in the WSMO. They are
mainly defined using the following properties:
the requested capability in the definition of a
goal represents the functionality of the services
the user would like to have, and the requested
interface represents the interface of the service
the user would like to have and interact with.
The concept of Mediator in the WSMO addresses
the handling of heterogeneities occurring between
elements that need to interoperate. This is achieved
by resolving mismatches between different
terminologies (data level), on the communicative
behavior between services (protocol level), and on
the business process level. A WSMO Mediator
connects elements and provides mediation
facilities for resolving such mismatches. The
WSMO defines different types of mediators for
connecting the distinct WSMO elements: OO
Mediators connect and mediate heterogeneous
ontologies, GG Mediators connect Goals, WG
Mediators connect Web services to Goals, and
WW Mediators connect interoperating Web
services to resolve mismatches between them.
Web Service Modeling Language
(WSML)
The Web service Modeling Language (WSML)
[3] is a language for the description of ontologies,
goals, Web services and mediators, based on the
conceptual model of the WSMO. The WSML
provides one coherent framework which brings
together Web technologies with different, wellknown
logical language paradigms: Description
8
November 2005|AgentLink News 19

FEATURE
Logics, Logic Programming, as well as F-Logic as
starting points for the development of a number
of WSML language variants: WSML-Core,
WSML-DL, WSML-Flight, WSML-Rule, and
WSML-Full. The WSML variants differ in logical
expressiveness and in the underlying language
paradigms. They allow users to make the tradeoff
between the degree of expressiveness and the
implied complexity on a per-application basis.
The WSML has two alternative layerings:
• WSML-Core → WSML-DL → WSML-
Full, and
• WSML-Core → WSML-Flight → WSML-
Rule → WSML-Full.
For both layerings, WSML-Core and WSML-
Full denote the least and most expressive layers
respectively. The two layerings are disjoint to a
certain extent in the sense that inter-operation
between the Description Logic variant (WSML-
DL) and the Logic Programming variants
(WSML-Flight and WSML-Rule) is only possible
through a common core (WSML-Core) or
through a very expressive superset (WSML-Full).
The WSML, can be seen as a testing ground for
the development of formal techniques for Web
service description.
Web Service Modeling Execution
Environment (WSMX)
The Web service Execution Environment
(WSMX) [4] is an execution environment which
enables discovery, selection, mediation, and
invocation of Semantic Web services described
according to the philosophy of WSMO. It is
thus a WSMO reference implementation. The
WSMX provides a tangible testbed for the
WSMO in order to prove its viability as a means
to achieve dynamic interoperability between
Semantic Web services.
In short, WSMX functionality can be summarized
as performing discovery, mediation, selection
and invocation of Web services on receiving a
user goal specified in WSML. The user goal is
first matched against the formal descriptions
of Web services registered with the WSMX.
If successful, one or more service descriptions
(ranked according to user preference) can be
returned. The most appropriate service as
selected by the user is then invoked and the
result returned to user. Prior to the invocation
step, the WSMX ensures that the data provided
for the service invocation is in the format that
the Web service expects. If necessary, a data
mediation process is performed to ensure interoperability
between different entities. Presently,
the WSMX architecture relies on a set of
loosely-coupled main components that provide
functionality for each step of the Web service
usage process: discovery, selection, mediation
and invocation.
Final Remarks
Semantic Web services, by combining recent
Web-related trends, constitute one of the
most promising research directions to improve
the integration of applications within and
across enterprise boundaries. In this context,
the WSMO aims to provide the conceptual
and technical means to realize Semantic Web
services, improving the cost-effectiveness,
scalability and robustness of current solutions.
The WSML provides a formal syntax and
semantics for the WSMO by offering variants
based on different logics in order to provide
several levels of logical expressiveness and thus
allowing tradeoffs between expressivity and
computability). Finally, the WSMX provides a
reference implementation for the WSMO and
the interoperation of Semantic Web services.
References
[1] http://www.wsmo.org/
[2] http://www.sdk-cluster.org/
[3] http://www.wsmo.org/wsml
[4] http://www.wsmx.org/
Trust: Challenges and Opportunities
Nathan Griffiths
University of Warwick
United Kingdom
nathan@dcs.warwick.ac.uk
Introduction
Trust is fundamental in distributed systems
where individual components interact to achieve
some overall objective. In small-scale or closed
systems this trust can be implicit, imbued to the
individual components and the system overall by
its designers and implementers. In open or largescale
systems however, it is becoming increasingly
common for trust to be explicitly represented and
reasoned about by the components, or agents, in
the system. In recent years trust has become a hot
research topic, with numerous conferences and
workshops attracting both academic researchers
and industrial representatives from solutions
providers in areas as diverse as telecoms,
logistics and e-business. This article gives a brief
overview of the alternative approaches to trust
and attempts to identify some of the important
research questions.
The current interest in trust creates an
active environment for trust researchers and
practitioners. However, it also raises some
challenges. Trust research has parallels with
agent research a decade ago — it is an exciting
area of clear value, but there is a risk that debate
about definitions and mechanisms might add
confusion and delay widespread adoption. Just as
there was (is) no consensus definition of agents,
there is similar debate over trust. For example,
how does trust relate to reputation Is trust an
individual (experience-based) notion, or should it
encompass others’ (potentially subjective) views
Are trusted agents secure and reliable, or do they
simply have “good” intentions These questions
are important, but it is equally important to
ensure that confusion is avoided, and that as a
community we have a clear overall view.
Psychological and Cognitive
Approaches
Many trust models take a cognitive view of
agents and trust, typically relying on folk
psychology notions such as belief and desire.
Agents trust others with respect to some activity
or the performance of some task, and consider
trustworthiness according to beliefs about such
aspects as competence, disposition, willingness,
dependence, and fulfilment [1]. The level of
trust is determined by these beliefs, along with
past experiences and possible recommendations
from others. Some cognitive approaches also
consider modelling the desires/motivations
of other agents, and incorporate this into
assessing trustworthiness. These approaches
give a powerful mechanism for reasoning about
interactions and the trust, power and dependence
relationships between agents. However, it can be
difficult to translate their richness into a practical
implemented system, since the data structures
and the functions needed to manipulate them
are expensive to maintain.
Numerical Approaches
Numerical approaches are perhaps the most
November 2005|AgentLink News 19
9

FEATURE
commonly used with numerous proposed
mechanisms, in which agents represent the
trustworthiness of others in numeric intervals,
typically [-1, +1] or [0, 1]. The lower bound
corresponds to complete distrust and the upper
bound to blind trust. Agents either keep track of a
numerical value that is updated by some function
after each interaction, or are equipped with a
function to transform a history of interactions
into a numerical value. Some approaches
decompose trust into separate values for different
situations, or according to
the different dimensions
of an interaction (such as
cost, quality, timeliness).
Numerical methods tend
to use a trust threshold
and only when trust is
above the threshold will
cooperation take place.
The main advantage
of numerical approaches is their simplicity,
since it is relatively inexpensive to incorporate
trust into an agent’s decision making. This
simplicity, however, is also their disadvantage.
Firstly, numerical approaches do not provide
the richness of reasoning available with other
techniques. Secondly, there is limited meaning
to the values themselves, which encumbers the
sharing of information between agents and
external reasoning about the system.
Probabilistic Methods
Probabilistic methods are a subset of numerical
approaches in which trust is represented in the
interval [0,1]. However, this number represents a
probability and has a clearer semantics associated
with it. There are many probabilistic approaches
ranging from those based on simple objective
and subjective probabilities, to those using more
complex Bayesian probability distributions.
Decisions are made in a similar manner to
numerical approaches by use of trust thresholds
Socially-oriented
and maximising the probability of success.
Reputation Systems
Many reputation-based approaches have been
proposed, ranging from centralised systems
that aggregate feedback (à la eBay), through
decentralised feedback systems, to numerical and
probabilistic approaches that are augmented with
recommendations from other agents. Reputation
systems generally use a combination of direct
experience, recommendations, and knowledge
Trust research has parallels with agent research a decade ago — it
is an exciting area of clear value, but there is a risk that debate
about definitions and mechanisms might add confusion and delay
widespread adoption. Just as there was (is) no consensus definition
of agents, there is similar debate over trust.
Service-oriented
of the social structure of the system to represent
and reason about trust. The incorporation of
reputation greatly enhances the richness of a
trust model, but typically increases complexity,
and opens up questions about issues such as
collusion.
Certificates and Keys
Where trust is viewed as a mechanism for
ensuring security, it tends to be achieved via
protocols, certificates or keys. Some approaches
define detailed interaction protocols that ensure
detection of any deviation from expected
behaviour, and define the actions or sanctions
that should be taken in such cases. Other
approaches use trusted third parties to provide
verification and authentication, and to act as
intermediaries in interactions. However, the
most common security-oriented approach is to
use certificates and keys: a certification authority
issues a certificate verifying that an agent’s public
key is owned by that agent. These public keys can
Security-oriented
then be used to sign and encrypt data to ensure
authentication and privacy. Certificates and keys
provide a powerful mechanism for achieving
security, but it can be difficult to combine them
with more general social- or service-oriented
approaches.
Challenges for Trust Researchers
Computational trust is a young and active
research area and numerous techniques have
been proposed, as introduced above. However,
there are a number of open
research challenges. Since
there is no overarching
view of trust it is difficult
for implementers to select
a trust mechanism for a
given environment. There
is a need for researchers to
frame their work within
the context of the general
trust landscape to enable simple comparisons
between approaches. A trust taxonomy would
begin to address this problem. Furthermore, trust
should be seen as a fundamental component of
any multi-agent system. Again, a taxonomy of
trust would assist in enabling this, especially
if aided by the provision of suitable software
tools to assist in incorporating trust into
agents. Finally, much trust research takes place
in simulated semi-closed environments, and a
number of issues must be addressed to enable
trust mechanisms to be effective in real-world
domains. We therefore propose three primary
challenges for trust researchers.
A Trust Taxonomy A taxonomy of trust would be
valuable, firstly to aid implementers in selecting
appropriate trust techniques for a particular
context and, secondly, to assist researchers in
positioning their work, and comparing it to
other approaches. The division of trust literature
into socially-, service- and security-oriented
trust might a starting point, but perhaps what
should be aimed for is a pattern library of trust
techniques. Ideally, an implementer should be
able to select a trust approach easily, based on
the characteristics of the domain.
logic economic certification
numerical/threshold-based
PKI
probabilistic
philosophical
protocols/policies
norms
psychological
Figure 1. Selected techniques and mechanisms for trust.
Agent Platforms Trust is fundamental to multiagent
systems. However, many existing agent
platforms do not incorporate trust by default. Trust
should be viewed as a fundamental component
of any agent platform, and implementers should
be able to select an appropriate “pattern” for a
given domain. Existing platforms need to be
augmented with “trust wrappers”, while new
platforms should include placeholders for trust
mechanisms by default.
Coping with the “Real-World” The real-world
is a complex place, with interactions potentially
failing in numerous ways for a variety of reasons,
including malicious motivations of cooperative
partners, competition between agents, and
10
November 2005|AgentLink News 19

FEATURE
unavoidable environmental change. Current
trust models are typically relatively simplistic
in updating trust after interactions. If trust is to
be effectively applied in real-world applications,
then there must be a mechanism to distinguish
between intentional and unintentional failure.
Furthermore, as Marsh proposes, there must
be a distinction between trust, mistrust and
distrust, i.e. trust, misplaced and incorrect
trust, and explicit distrust (c.f. information,
misinformation and disinformation) [2].
Summary
Trust is a rich notion and an area of active
research. In this article we have tried to give a
flavour of the breadth of trust research. As in any
young research area there are a number of open
challenges; this article has presented a personal
view in identifying three areas that the author
sees as meriting immediate attention. Clearly,
there are numerous other open questions, but
the author hopes that a focus on these challenges
will help facilitate more widespread use of trust
in agent-based systems.
[1] C. Castelfranchi. Trust mediation in
knowledge management and sharing. In C.
Jensen, S. Poslad, and T. Dimitrakos, editors,
Proceedings of the Second International
Conference on Trust Management (iTrust
2004), pages 304–318, 2004.
[2] S. Marsh and M. R. Dibben. Trust, untrust,
distrust and mistrust — an exploration of
the dark(er) side. In P. Herrman, V. Issarny,
and S. Shiu, editors, Proceedings of the
Third International Conference on Trust
Management (iTrust 2005), pages 17–33.
Springer-Verlag, 2005.
[3] S. D. Ramchurn, D. Huynh, and N. R.
Jennings. Trust in multi-agent systems.
Knowledge Engineering Review, 19(1):1–
25, 2004.
Diversity of Trust Research
There is a growing corpus of literature on trust, a detailed overview of which can be found in
the further reading identified below. There is no overarching taxonomy of trust research, and
the wide applicability of trust gives rise to a wide range of approaches. However, we can
broadly divide these approaches into three areas.
Socially-Oriented Trust - Typically influenced by social science, psychology or philosophy,
socially-oriented trust is viewed as a social notion for modelling and reasoning about the
relationships between agents. Socially-oriented trust often considers issues such as the
motivations of agents, and the power and dependence relationships between them.
Service-Oriented Trust - Taking a pragmatic view, service-oriented trust is a mechanism
for achieving, maintaining, and reasoning about quality of services and interactions. Agents
typically maintain their own trust information about others, possibly incorporating the
recommendations of others, and use this to inform their decision making processes.
Security-Oriented Trust - Taking the view of trust as a mechanism for ensuring security,
encompassing issues of authentication, authorisation, access control, privacy, etc. Securityoriented
trust also includes work on “trusted computing”, i.e. building trusted platforms to
ensure privacy and security.
In each of these areas a range of techniques and mechanisms have been proposed, drawing
on work in areas as diverse as logic and the social sciences. For any given situation there
are generally several alternative candidate trust models/techniques for an implementer to
choose from, as illustrated in Figure 1. (Note that this figure is not intended to be exhaustive,
and many other techniques exist.)
Trust literature can be further divided according to whether it is concerned with individual- or
system-level trust. In the former, individual agents model and reason about others, while in
the latter agents are forced to be trustworthy by externally imposed regulatory protocols,
and mechanisms [3]. Additionally, some trust models are centralised and have a single
repository of information, while others are decentralised with individuals maintaining their
own information. Finally, we can distinguish between models concerned with direct trust
where agents trust others directly based on their experiences, and recommendation trust
where trust is based on the recommendations of others. For each of these categories of trust
there are a number of alternative approaches discussed in the literature, the most common
of which are also briefly introduced.
Benchmark simulations for Multi-Agent Learning
Maarten van Someren (University of Amsterdam)
maarten@science.uva.nl
Evaluating methods and systems on publicly available datasets has proved to be a succesful methodology
in Machine Learning and in Information Retrieval. Evaluating multi-agent learning systems requires
simulation environments, even though this can be more difficult to achieve than for the other areas (e.g.
the traffic light simulation, see http://sourceforge.net/projects/stoplicht).
This effort may become part of the new network of excellence KDUbiq which will start in December or
January. If you are interested then please contact me by email (maarten@science.uva.nl).
November 2005|AgentLink News 19
11

AGENT RESEARCH OVERVIEW
ALSO IN THIS SECTION...
Answer Set Programming and Agents
AgentCities.ES:
Spanish network of agent-based computing
Why Argue Dialogue Types and
Argumentation in Agent Interaction
Katie Atkinson and Trevor Bench-Capon
University of Liverpool
Liverpool
UK
{katie,tbc}@csc.liv.ac.uk
Agents interact not only with their environment,
but also with one another. Like human agents,
software agents effect this interaction through
conversation and dialogue. Inter-agent dialogues
are typically founded on the theory of speech
acts, as developed by Searle [3]. Different speech
acts give different force to the same content.
Thus ask(“Paris is in France”) and tell(“Paris is
in France”) are two speech acts with the same
content, but with different forces, one seeking
and the other providing information. Agent
interactions, however, require more than one-shot
utterances, and may require the exchange of a
complex series of speech acts. Such dialogues need
to be sustained and coherent. This can be effected
through the use of conversation classes (see [1] for
an early example), or protocols which determine
which speech acts can be used to initiate a
particular kind of conversation, and which speech
acts can be
used at various
points in
conversations
of this sort.
This technique
certainly provides a means of co-ordinating
and structuring dialogues between agents. It is,
however, important to recognise that the speech
acts change their meanings according to the type
of dialogue in which they appear. There is a strong
pragmatic element to the meaning of speech acts,
and it is the type of dialogue which provides
the context which determines this pragmatic
element. Thus if agent A poses a question to
agent B, we usually expect the following to be
true: Agent A does not know the answer, but
wants to know and furthermore expects agent B
to know and provide the answer. But in a quiz
or an oral examination, the questioner will know
the answer and wishes to discover whether the
agent asked also knows it. Similarly if an agent
states information that it does not know, and
perhaps does not even believe, whether it be a lie
or a legitimate guess will depend on the dialogue
of which the assertion is a part. For this reason
discourse theorists e.g. [2] and informal logicians,
such as Walton and Krabbe [3], have seen the
importance of identifying and characterising
types of dialogue so that the pragmatic elements
of speech acts can be understood.
Walton and Krabbe’s typology of dialogues
in Commitment in Dialogue, is probably the
most influential classification. They identify six
dialogue types: persuasion, negotiation, inquiry,
information seeking, deliberation and eristic.
The last of these involves venting grievances
or quarrelling such that dialogue substitutes
for physical confrontation. This type seems of
little application to agent systems so we will not
consider it further here. The remaining five are all
plausibly useful in agent interactions.
Dialogue types are characterised by the initial
positions of the participants, the main goal of the
dialogue and the individual aims of the participants.
Persuasion starts from a disagreement with the
aim of bringing both parties into agreement, with
each party trying to encourage the other to change
their point of view. Negotiation starts from an
initial allocation of resources or tasks, and aims
to move to a better allocation, with each party
trying to achieve an allocation more favourable
to itself. Inquiry, which requires agents to pool
their information to solve a problem, begins
with neither party having information thereby
requiring cooperation via information exchange
to solve the problem. In information seeking
dialogues the parties start in an asymmetric
position, one with some information which the
other lacks. The aim of this dialogue type is the
spread of information. One participant has the
goal of acquiring this information. Another may
have the goal of sharing its information, but in
some circumstances may have the goal of hiding
it. Finally, deliberation dialogues concern actions:
the starting point is that some action must be
taken and the aim of the dialogue is to reach a
decision acceptable to the parties involved. The
goals of the participants are to influence the
decision such that it is acceptable to them.
These dialogues types are intended to be rather
broad, and will have a number of flavours depending,
for example, on the relationship between the
agents. Agents
six dialogue types: persuasion, negotiation, inquiry,
information seeking, deliberation and eristic
may have a
common cause
and be disposed
to co-operate,
or they may be
indifferent or even hostile towards one another.
Thus in a negotiation, an agent may be concerned
only with achieving the allocation which is best
for itself, or it may constrain its choices to ensure
that another agent improves its position to some
extent, or it may even be prepared to sacrifice itself
so as to achieve the best outcome with respect to
joint utility. Similarly, in a deliberation, agents
may have more or less concern for whether the
decision is acceptable to the other parties: contrast
a couple planning a holiday with a democratic
debate between political parties. Perhaps the
most striking contrast is the information seeking
dialogue where the agent with the information
may either be willing or reluctant to share its
knowledge.
12
November 2005|AgentLink News 19

AGENT RESEARCH OVERVIEW
An important fact to recognise is that an
interaction will typically involve several of these
dialogue types. Thus a deliberation dialogue
may well move through information seeking,
inquiry, persuasion and negotiation phases before
a decision can be reached. Walton and Krabbe
stress the importance of these dialectical shifts,
which may be sharp or gradual. In particular,
misunderstandings often occur when a shift has
taken place but one of the parties to the dialogue
has not noticed. Moreover shifts may be illicit,
and it is these illicit shifts that give rise to informal
fallacies. The important lesson from this is that we
cannot study particular dialogue types in isolation:
for example if we focus on negotiation dialogues
we will still need to consider the other types in
order to model the complete interaction.
Another point to note is that these dialogues can
be about beliefs or about actions. Deliberation is
tied to action and inquiry to belief, but persuasion
may involve convincing the other party that
something is the case, or that a particular action
is wrong. Negotiation may be about the division
of resources, but also about division of tasks.
Information may be sought about facts or about
options for action. This distinction will become
important as we come to discuss arguments.
The prevailing model, based on speech acts,
sees the content of utterances to be essentially
propositional, with the different speech acts
imparting different forces, such as assertive or
interrogative. To participate in the above dialogue
types, however, the agents will often need to
exchange information with more structure so as to
back up their assertions and choices with reasons.
In particular they will need to exchange arguments,
to justify themselves and to persuade other agents.
Proofs are the limiting case of arguments: where
information is complete, concepts are sharp and
conditionals are strict, utterances may be justified
by proof. Such conditions can be met in formal
domains such as mathematics, but in the domains
in which agents typically operate information is
incomplete, concepts are vague and conditionals
are defeasible. Thus justification has to take the
form of an argument, and as such is open to
challenge and defeat.
It has become accepted that an argument, often
expressed in the form “since P then Q” can be
challenged in three ways. Since the underlying
conditional is not strict, the argument can
be rebutted by an argument for the opposite
conclusion, say not Q. Alternatively the premise
may be rejected. Finally it may be argued that the
rule is not applicable in the particular case, perhaps
because the circumstances are exceptional, or the
offered rule is based on incomplete information.
Such criticisms are normally called undercutters.
These forms of attack do not, however, cover
all the ways in which an argument justifying an
action can be challenged. Philosophers have, since
the time of Aristotle, seen practical reasoning
– reasoning about action – as a separate subject
of study, and these additional attacks motivate
this. Arguments for actions can be challenged
not only in the above three ways but also by
proposing alternatives, other actions that will
achieve the same goal, or some more desirable
goal, and by pointing to side effects, inadvertent
and undesirable consequences of the action
which will occur in addition to realising the
goal. Note that these attacks do not, unlike the
attack on an argument for a belief, render that
argument invalid or inapplicable. The reasons
advanced for performing the action remain valid
reasons, but they do nevertheless provide reasons
for rejecting the conclusion. The resolution of a
discussion also differs when it concerns actions
rather than beliefs. Because beliefs are validated
against the objective measure of what is in reality
the case, ending the discussion with disagreement
cannot be considered successful. Because actions,
in contrast, relate to what we want to be the case,
what we shall attempt to achieve, the discussion
can rationally end in disagreement since the
interest and the aspirations of the parties may
differ.
Arguments play different roles in the various
dialogue types. Argument is obviously
fundamental to persuasion dialogues: it is
through argument that the parties can be brought
to change their positions. An important point to
recognise, however, is that the argument needs
to be persuasive not to the party advancing it,
but to the audience. An understanding of the
background assumptions and prejudices of the
audience is therefore essential for successful
persuasion. In negotiation, arguments are used to
indicate limits to flexibility, or other problems with
the proposal, and so to indicate what subsequent
offers may be successful. If I am buying a car,
and refuse the asking price, my arguments may
show that I cannot afford it, or that I find the car
unsatisfactory, and the seller’s response will differ
in the two cases. Argument is useful in guiding
the negotiation so as to improve efficiency and
perhaps also the outcome. In inquiry dialogues,
the participants are seeking a joint solution,
and so are in effect seeking to co-operate in the
construction of an argument to justify their
conclusion, and in the course of the dialogue
partial arguments leading towards the solution
may be proposed. Similarly an argument to justify
a line of action is jointly sought in deliberation,
and in deliberation a variety of arguments for
alternatives will typically be advanced to allow
the parties to determine the options from
which the best course of action can be chosen.
In information seeking dialogues the role of the
argument is essentially to justify the information
provided, to round out the understanding of
the information seeker. Arguments can play a
variety of roles, and these determine who needs
to accept them. They can justify, in which case
it is enough that the agent making the argument
accepts it. This predominates in negotiation and
information providing. Or they can convince,
when the agent to whom the argument is directed
must accept it, as in persuasion. Or they can be
used as part of a problem solving process, as in
inquiry and deliberation, in which case all parties
involved need to find them acceptable.
In this article we have drawn attention to some
important insights from discourse theory and
informal logic which we believe need to be absorbed
by those working on agent communication if the
aspiration of agents capable of interacting in a
rich and fruitful manner is to be achieved. There
are a number of important lessons:
• The pragmatic content of speech acts: it is
vain to seek a single meaning of a speech act,
such as tell, independent of the discourse
context in which it is used.
• The range of dialogue types, and the need to
recognise that interactions will shift between
them. It is not possible either to treat all
dialogues as alike without reference to the
initial starting point, the aims of the dialogue,
and the goals of the participant. Nor can any
single type be studied in isolation.
• The distinction between dialogues about
belief and dialogues about action. In
particular, arguments relating to action can
be challenged in additional ways, and rational
disagreement is always possible.
• The importance of argument and the
recognition that arguments play different
roles in different dialogue types has important
implications for what counts as a successful
argument in the various contexts.
These issues indicate a number of lines for
interesting research, and it is questions like these
that make agents and argumentation such an
exciting field of study.
References
[1] Barbuceanu, M., and Fox, M.S., (1995),
COOL: A Language for Communication
in Multi Agent Systems in Lesser, V., (ed),
Proceedings of the First International
Conference on Multiagent Systems, MIT
Press, Cambridge, Mass.
[2] Carlson, L., (1983), Dialogue games: an
approach to discourse analysis, Reidel,
Dordrecht.
[3] Searle, J.R. (1969). Speech Acts: An Essay
in Philosophy of Language. Cambridge
University Press.
[4] Walton, D.N., and Krabbe, E.C.W., (1995)
Commitment in Dialogue, State Univ of
New York Press.
November 2005|AgentLink News 19
13

AGENT RESEARCH OVERVIEW
Answer Set Programming and Agents
Jürgen Dix
Clausthal University of Technology
Germany
dix@tu-clausthal.de
Thomas Either
Vienna University of Technology
Austria
eiter@kr.tuwien.ac.at
Introduction
Agents need to make decisions based on what
they perceive and their model of the world.
Both sources are often incomplete and subject to
change.
Representing knowledge and reasoning under
incomplete information about the world is indeed
a very difficult task. Formal approaches extending
classical reasoning (which was mostly based on
classical logic) were first considered in the 80ies
with the seminal works of Reiter, McCarthy, and
McDermott/Doyle (see [1]) and led to the area of
nonmonotonic logics. In parallel, in databases, an
important milestone was the rise of the relational
model introduced by Codd back in 1974, which
revolutionised the area and found its way into
industry and became a standard. At about the
same time, Colmerauer and Kowalski introduced
PROLOG as a programming language (we refer
to [1] and the references therein).
Answer set programming (due to Gelfond
and Lifschitz [2]) can be seen as a hybrid of
these three approaches above: it combines a
true declarativeness (which PROLOG never
really achieved) with the efficiency of database
technology and the expressive power of
nonmonotonic reasoning.
ASP in a nutshell
What exactly is answer set programming (ASP for
short) Let us consider a simple example and try
to show the essential features of ASP. Consider
the following simple rules:
(1) col(X,red) v col(X,blue) v col(X,green) ←node (X)
(2) ← edge(X,Y), col(X,C), col(Y,C)
The first rule expresses that an object that is a
node, has to have at least one of three colours
red, blue or green. The second rule is a constraint
stating that if two objects are related (there is an
edge between them), then they can’t take on the
same colour.
Suppose now there is a database consisting of
facts of the form node(a), node(b), node(c), ...
and edge(a,b), edge(b,c), ... Such a database
encodes a graph with nodes a,b,c,... and edges
a→b, b→c, ... The question we would like to
ask is: what are all possibilities to satisfy the rules
(1) and (2) Note that X, C are uppercase letters
that are considered as variables (as in PROLOG):
all terms can be instances of them. It should
not come as a surprise that all such possibilities
(intuitively) constitute all 3-colourings of the
underlying graph.
The semantics of ASP (which defines the notion
of an answer set of a program) makes sure that
exactly the answer sets are 3-colourings. If we
used classical logic and considered classical
models, this would not hold. The reason is that
in classical models, nodes might have assigned
multiple colors; in answer sets, this does not
happen since rules are applied parsimoniously
and only one colour per node is included.
An answer set program therefore consists of
• rules of the form head ← body
• that can contain variables,
• the head can be a disjunction (or empty)
• the body is a conjunction (or empty).
An important feature is that the body of a rule
can also contain negation (a kind of negation-asfailure).
This creates a close relationship between
answer sets and extensions in default logic (and
thus allows to use methods from nonmonotonic
reasoning).
We are not presenting all the different features of
the language like disjunction, negation, strong
negation, in the head and in the body, cardinality
constraints, arithmetic, built-in predicates,
preference handling etc. These and other
extensions (surveyed in [17]) are very important
from a user perspective. However the main
features of ASP are as follows:
• the definition of “if-then” rules,
Figure 1. Schema for problem encodings with a separation between specification and data.
• which contain variables but no functions,
• an ASP program consists of finitely many such
rules,
• the semantics associates to each program a
number of answer sets (0, 1, or several),
• the semantics does not depend on the ordering
of the rules or subgoals in them,
• there exist nowadays several high performance
ASP engines to compute answer sets, including
DLV, Smodels, Cmodels, ASSAT (see [9]).
Programming with ASP then means the following.
Suppose you are given a problem which has
several solutions (e.g. a planning problem with
several valid plans). Write down an ASP program
such that the answer sets of it correspond to the
solutions of the original problem. Then compute
the solutions with an ASP engine.
ASP has been proven to be successful in a variety
of areas such as planning, diagnosis, configuration
and space shuttle control, which will be considered
more in detail below.
Extensions of the basic ASP-rule language
permit the use of constituents which are calls
to arbitrary legacy code (e.g. functions giving
back perceptions of the world (but encapsulated
in logic)). This makes it possible to use ASP as
the underlying program of an agent: the agent is
constantly computing the potential answer sets
and acting according to them. These answer sets
change as new perceptions are coming in. This
is the approach adopted in IMPACT, which is a
genuine agent platform project briefly discussed
below. Other generalisations of ASP to interface
legacy code, like DLVEX or DLVHEX, are
currently under development and will be released
soon.
ASP and Agent Programming
Thanks to its capability of dealing with incomplete
and default knowledge, ASP lends itself as a simple
yet expressive rule-based language for knowledge
representation, which may be utilised to develop
special plug-in reasoning modules for agent
architectures. For example, modules for providing
decision support, for reasoning about actions, or
for reasoning about security and trust may be
crafted and interfaced with other components in
the agent architecture. The modules may then be
activated in the thinking part of the prototypical
Observe-Think-Act cycle that has been proposed
14
November 2005|AgentLink News 19

AGENT RESEARCH OVERVIEW
for logic-programming based agents by Kowalski
and Sadri [10]. As mentioned above, extensions
of ASP which facilitate access of sensory input
and other agent data structures as first class
citizens in the language will be soon available.
Moreover, an extension accounting for possible
perception failure has also been recently devised
[18].
In the context of multi-agent systems, ASP has
been utilised in different projects already. In the
DALI project [11], ASP serves as a backbone for
an Active Logic Programming Language, which
is designed for executable specification of logical
agents, without committing to any specific agent
architecture, and provides reactive and proactive
features. An important aim in its definition
was introducing in a declarative fashion all the
essential features, keeping the language as close
as possible to the syntax and semantics of the
plain Horn clause language.
The work [12] deals with specifying and
verifying socially acceptable patterns of agent
interaction in multi-agent systems. Protocol
specifications which indicate exactly what kind
of messages agents should exchange and when
for achieving certain outcomes might be viewed
as compromising agent autonomy. A relaxed
approach thus uses a set of regulations or norms
which describe what sorts of agent behaviour are
acceptable and forbidden according to the society,
and how agents who violate these regulations
should be treated and sanctioned, with the
overall goal of discouraging but not necessarily
preventing socially unacceptable behaviour from
occurring. ASP is used for the formalisation and
Fig 2. IMPACT Agent architecture.
implementation of the model, as well as query
and verification language. A major advantage
here is the possibility to combine specification
and implementation into one formalism, and
flexibility. Using the same ASP program, one
can plan for a certain state, diagnose a state or
generate all acceptable states. Interactions among
logic-based agents using ASP are also respected
in the framework of Social Logic Programming
[19].
A full-fledged extension of ASP for agent
programming has been developed in the
IMPACT (Intelligent Maryland Platform for
Agent Collaborating Together) project, which
has been led by VS Subrahmanian. IMPACT
agents have a rule-based agent program which
governs the agent behaviour, in terms of actions
that the agent takes, on top of the agent’s internal
data structures, which include its message box
for communication (see Figure 2).
Access of the internal data structures is facilitated
through special code calls and atom that allows
checking whether a given value belongs to the
result of a code call or not. Action execution
is based on rules in language which features
deontic modalities such permittance, obligation,
or forbiddance of action execution, which might
depend on other actions. For example, the rule
Do(dial(N)) ← IN(N,phone(P)), O(call(P))
intuitively says that the agent should dial the
phone number N and is obliged to call P. The
general form of an agent rule in IMPACT is
Op_0 act_0 ← Cond, (~) Op_1 act_1, ... , (~)
Op_m act_m,
where each Op_i is a deontic modality, act_i
are action atoms, and Cond is a conjunction
of constraint atoms and atoms of the form
IN(X,f(Y)) resp. NOTIN(X,f(Y)), which access
the data structures of the internal agent state
through API functions f(Y) and test whether
X is in resp. not in the result. The semantics of
such an agent program, which consists of a set of
such rules is given in terms of a feasible status set,
which can be viewed as a set of status assumptions
for the action atoms compliant with the rules of
the agent program, deontic laws (e.g., an action
can not be both permitted and forbidden at the
same time), integrity constraints on the agent
state, and possible constraints on parallel action
execution. A suite of refinements of feasible status
set semantics has been defined, of which the so
called reasonable status set semantics implements
and faithfully generalises the semantics of answer
set programs to agent programs. For more details,
we refer to the chapter about IMPACT in [3].
Applications
ASP has been successfully deployed to a number
of applications in various areas including classic
AI fields such as planning, diagnosis, or decision
support, but also to other fields which gained a
lot of interest in the recent past such as security
management, information integration, and web
service composition. A comprehensive survey
of applications of ASP in these and further
areas is available at [7], and an ASP showcase
featuring difference applications in the areas of
configuration, information integration, security
analysis, agent systems, semantic web, and
planning can be found at [8].
Here, we just briefly touch three applications,
reporting the gist and the main benefits of the
ASP usage. An interesting application was
a knowledge-based decision support system
for controlling some functions of the Space
Shuttle [5]. Here, the task was to find plans for
the operation of the Reaction Control System
(RCS) of the Shuttle, which has the primary
responsibility for manoeuvring the aircraft while
it is in space. When failures in the system become
apparent, astronauts need to execute sequences of
actions which must assembled on the fly in order
to complete the mission. For that, a system based
on ASP has been conceived, which implements
a simplified model of the RCS (without loss of
necessary detail) and incorporates agent actions
and planning. The main benefits of ASP in this
application are a seamless integration of reasoning
about actions, change, and planning using a
natural declarative language; in fact, the new
system worked much better than its predecessor.
Another task in which ASP turned out to be
very useful is advanced information integration,
which is an important issue for information
November 2005|AgentLink News 19
15

AGENT RESEARCH OVERVIEW
agents. Here, the task is providing a uniform
interface to various pre-existing data sources, so
as to enable users to focus on specifying what
they want, rather than on thinking of how to
obtain the answer. Problems here are that the
integration of data from different sources to
a global user view, where source data is usually
heterogeneous, incomplete, and inconsistent. In
the recent EU project INFOMIX [6], ASP was
used as a computational backbone to implement
query answering from inconsistent information
integration systems. Here, an acknowledged
formal model for query answering has been
cast into an ASP program, which serves as an
executable logical specification of the semantics,
which aims at “repairing” inconsistencies. The
benefits of ASP in this application are that it
has expressiveness required for this application,
and that it provides a language for expressing
repair policies, which may be customised to a
particular scenario, in a fully declarative manner
rather than in a procedural way. This facilitates
reasoning about specifications, their properties
and behaviour much better since reasoning
about programs is a principal issue in logic
programming.
A third application which we briefly touch
upon here is monitoring of agent collaboration
in a multi-agent system. Agent behaviour has
to be judged on the basis of message exchange
between agents. The task is to find out whether
the collaboration is compliant with message
exchanges, collected in a message log, which
correspond to agent action sequences that will
accomplish an overall goal. To this end, agent
collaboration is modelled as an action theory, and
feasible action sequences reaching the goal are
determined by a planner, whose compliance with
the actual behaviour of agents allows to detect
possible collaboration failures, which may be due
to implementation errors (the code of some agent
is buggy) or due to design errors (the collaboration
is improper). In this way, the approach can be
applied to aid offline testing as well as online
monitoring. It has been implemented within
IMPACT and DLVK [13], which is a planning
system based on a declarative action language and
ASP. Thus, ASP comes into play here at different
levels. Figure 3 shows the message protocol for
the agents in the Gofish Postal Service system,
which collaborate for servicing package delivery.
Even in this small example, communication has
many steps and errors are (especially in the system
implementation phase) almost unavoidable;
monitoring helped a lot in debugging different
implementations of the system.
WASP
A working group on Answer Set Programming
(WASP) has been created in 2001 as a European
project within the 5th Framework Programme
(IST project IST-FET-2001-37004). It consist of
18 nodes spread all over Europe: they constitute
the main body of research about ASP done in
Europe.
WASP has greatly helped to keep European
leadership in ASP and to identify and realise a
set of target industrial applications. It has also
helped to develop a variety of language extensions
providing constructs and features which ease
problem solving in practice.
The web-page contains a showcase collection
which provides information about the use of
ASP in configuration, information integration,
security analysis, agent systems, semantic web,
and planning.
The interested reader is referred to the homepage
of WASP - http://wasp.unime.it -.
Conclusion
Answer Set Programming started as a particular
semantics for logic programs with negation in
the late 80s. It is now considered a declarative
paradigm for representing and reasoning about
knowledgein a rule-based way. Methodologies
have been (or are still being) developed to make
ASP useable also for the non-specialist.
ASP is useful in agent programming for several
reasons:
(1) to seamless integrate reasoning about
actions, change, and planning using a natural
declarative language;
(2) to combine specification and implementation
into one formalism (executable
specifications);
(3) to build decision support systems based on
clear policies;
(4) to get easy access to efficient database
technology (the DLV solver, e.g., offers an
ODBC interface to relational databases);
and
(5) to get access to a vast theory and methods
about knowledge representation developed in
the last decades.
[15] and [16] point to some recent workshop
series devoted to programming agent systems
(where ASP techniques might come into play).
Fig.3: Protocol of message exchange in the Gofish system (message types).
References:
[1] G. Brewka, J. Dix, K. Konolige: Nonmonotonic
Reasoning: An Overview, CSLI Lecture Notes
73, Stanford (California), 1997.
[2] M. Gelfond, V. Lifschitz: Classical negation
in logic programs and disjunctive databases.
New Generation Computing, 9:365--385,
1991.
[3] C. Baral: Knowledge Representation, Reasoning
and Declarative Problem Solving with Answer
Sets. MIT Press, 2002
[4] R. Bordini, M. Dastani, J. Dix, A. El
Fallah Segrouchni: Programming Multi
Agent Systems: Languages, Platforms and
Applications, Springer Series MASA, No. 15,
2005
[5] M. Nogueira, M. Balduccini, M. Gelfond,
R. Watson, M. Barry: An A-Prolog decision
support system for the space shuttle, Proc.
3rd International Symposium on Practical
Aspects of Declarative Languages (PADL
2001), LNCS 1990, pp. 169-183. Springer,
2001.
[6] N. Leone et al.: The INFOMIX system for
16
November 2005|AgentLink News 19

AGENT RESEARCH OVERVIEW
advanced integration of incomplete and
inconsistent data, Proc. ACM SIGMOD
2005 Conference, pp 915--917. Project
homepage http://www.mat.unical.it/infomix/.
[7] S. Woltran (ed), ASP Model Applications
and Proofs-of-Concept: http://www.
kr.tuwien.ac.at/research/projects/wasp/
report.html.
[8] ASP showcase: http://www.kr.tuwien.ac.at/
research/projects/wasp/showcase.
[9] ASP solvers: http://dit.unitn.it/~wasp/Solvers/.
[10] R.A. Kowalski, F. Sadri: From logic
programming towards multi-agent
systems, Ann. Math. Artif. Intell. 25(3-
4): 391-419 (1999)
[11] S. Costantini, A. Tocchio: The DALI Logic
Programming Agent-Oriented Language,
Proc. 9th European Conference on Logics
in Artificial Intelligence (Jelia 2004),
LNCS/LNAI 3229, pp. 685-688. Springer,
2004.
[12] O. Cliffe, M. DeVos and J. Padget:
Specifying and Analysing Agent-based
Social Institutions using Answer Set
Programming. Proc. Workshop on Agents,
Norms and Institutions for Regulated
Multiagent Systems (ANIREM) 2005,
AAMAS.
[13] DLVK Planning System: http://www.dbai.
tuwien.ac.at/proj/dlv/K/
[14] A European Working Group on ASP (http://
wasp.unime.it/) coordinates and represents
most of the work on ASP done in Europe.
[15] ProMAS Workshop series: http://www.cs.uu.
nl/ProMAS/
[16] CLIMA Workshop series: http://centria.di.fct.
unl.pt/~clima
[17] I. Niemelä (ed): Language Extensions and
Software Engineering for ASP: http://www.
tcs.hut.fi/Research/Logic/wasp/wp3/wasp-wp3-
web/wasp-wp3-web.html
[18] F. Buccafurri, G. Caminiti, D. Rosaci.
Perception-dependent reasoning and
answer sets. Working Notes AI*IA-RCRA,
pp. 119-126, 2005. http://www.ing.unife.it/
eventi/rcra05/articoli/BuccafurriEtAl.pdf
[19] F. Buccafurri, G. Caminiti: A Social
Semantics for Multi-Agent Systems. Proc.
8th Conference on Logic Programming and
Nonmonotonic Reasoning (LPNMR 2005),
LNCS 3662, 317-329. Springer, 2005.
AgentCities.ES:
Spanish network of agent-based computing
(2003-2005)
Antonio Moreno
University Rovira i Virgili
Spain
antonio.moreno@urv.net
AgentCities.ES 1 was a research network founded
by the Spanish former Science and Technology
Ministry from March 2003 to March 2005. It was
coordinated by Antonio Moreno, from University
Rovira i Virgili (URV) in Tarragona, Spain. The
network had 18 members (17 University research
groups on agent technology and the Artificial
Intelligence Research Institute-IIIA). Its main
aim was to join and coordinate the efforts of
the Spanish research groups involved in the very
succesful AgentCities European project 2 , which
intended to build a worldwide network of agentbased
platforms that provided interesting services
to citizens. It is worth mentioning that Spain
was one of the leading countries in AgentCities
activities, and there are also many Spanish research
groups which have been actively involved in
AgentLink III activities (e.g. promoting TFGs
such as the ones on Agents Applied in Health
Care, Networked Agents, Agent Oriented
Software Engineering and Programming MAS).
Some of the activities that have been carried out in
the network in its 2-years duration have produced
results that may be interesting to the AgentLink
III community. These activities have been the
following:
• A workshop called “Intelligent Agents in the
Third Millenium” was organised within the
Spanish Conference on Artificial Intelligence
(CAEPIA-2003) in San Sebastián. It included
the presentation of 9 papers, a demo session
with 7 implemented multi-agent systems, and
a panel on the use of formal methodologies in
the design of agents and multi-agent systems
(coordinated by Juan Pavón and Jorge Gómez,
from the Univ. Complutense of Madrid).
Michael Luck, the AgentLink-II coordinator,
gave a very interesting invited talk on the
status of agent technology in Europe at that
moment. The workshop proceedings, with
most of the papers in English, is freely available
through the network web pages 3 .
• The network organised the “First Spring School
on Agents. Development of MAS. Concepts,
methods and tools” at the University of Sevilla
in March 2005. 35 PhD students attended this
3-days school, in which the most prestigious
Spanish researchers on agent technology
lectured on fundamentals of multi-agent
systems, FIPA standards, JADE, principles of
complexity in MAS, electronic institutions, the
INGENIAS methodology and development
of agents in real-time environments. Some
of the material is available in English in the
school web pages 4 .
• The members of the network also identified
research areas of common interest and defined
Working Groups associated with them. The
most active WGs have been the following:
Methodologies for Designing MAS, Learning
and Planning, Trust and Security, and
Application of Agents in Tourism. It is worth
mentioning that the WG on Methodologies
has built a very detailed set of web pages
(in English) on Agent-Oriented Software
Engineering (maintained by Pedro Cuesta at
the Univ. of Vigo) 5 .
• The network has funded the development of
a document (in English) that describe the 18
Spanish research groups that are its members.
For each group, it provides information
such as the main research lines, software
products developed by the group, services
that the group may provide, experience in
technology transfer to industry, and current
funded research projects. This document is
freely available in the network web pages,
and it may be useful to those searching for
collaboration with Spanish partners in future
projects 6 .
1 http://grusma2.etse.urv.es/AgCitES/
2 http://www.agentcities.org
3 http://grusma2.etse.urv.es/AgCitES/subpagines/documents/f1.pdf
4 http://imaginatica.eii.us.es/2005/agentes/index.phppag=schedule
5 http://ma.ei.uvigo.es/aose/
6 http://grusma2.etse.urv.es/AgCitES/subpagines/AgentCitiES.pdf
A 1-year extension of the network, also coordinated
by Antonio Moreno, has already been aproved by
the new Education and Science Ministry, although
it has not yet officially started. The new network
has 20 research groups and involves around 120
researchers.
November 2005|AgentLink News 19
17

STANDARDS REPORT
Latest News from the Standardisation World
Monique Calisti
Whitestein Technologies AG
Switzerland
mca@whitestein.com
The “Latest News from the Standardisation
World” column reports on relevant agent-driven
or related standardisation activities and events.
The central idea is to promote awareness of the
important activities that several international
standardisation bodies are undertaking in the
perspective of consolidating the industrial uptake
of agent technology.
There has been a flurry of activity regarding
standardisation activities this summer. We
report from the FIPA IEEE, which held its First
Technical European Meeting in Budapest just
before the AL3-TF3, and report on the latest news
from the Global Grid Forum meeting, held in
Boston (October 3-6, 2005). As ever, a plethora
of recommendations and results from W3C
working groups have appeared in recent months;
we report on a selection that are of particular
relevance to the AgentLink community.
Finally, a Technical Forum Group for
Coordinating Agent Standardization
Activities, TFG-CASA, was formed, and met in
Budapest, Hungary, with three main objectives:
1) to assess the importance of ongoing agentdedicated
and agent-related standardization
activities; 2) to clarify how existing standardization
bodies and groups relate and possibly contribute
to establishing stronger liaisons; and 3) to
understand and position the role of AgentLink
for ongoing and future agent-dedicated and
agent-related standardization work. Full details
of this meeting are reported in the TFG-CASA
report on page 22.
As usual, don’t forget, if you are directly involved
in any standardisation activity, or you are aware
of important standardisation work and events
that you would like to publicise, you can directly
contribute by sending your material to:
standardisation-activity@agentlink.org
Many thanks to all contributors for their
support!
FIPA – IEEE
Contribution by:
James Odell
email: email@jamesodell.com
The IEEE FIPA Standards Committee held
its first meeting in Budapest, Hungary, 13-14
September 2005, with 27 attendees. Membership
signup will commence in October, with elections
announced by the end of 2005.
The following groups have been formed and
approved:
• Agents and Web Services Interoperability
Working Group - AWSI WG. The primary
objective of the AWSI WG is to fill the
interaction gap between agents and web services.
Agents should be able to locate and interact
with web services seamlessly and vice versa.
Chair: Hiroki Suguri (Comtec)
Email: suguri@comtec.co.jp
• Human-Agent Communications Working
Group - HAC WG. The primary goal of
the HAC WG is to produce one or more
IEEE standards that extend the current
FIPA performatives for human-agent
communications. Even though the current
FIPA agent communication language can be
applied to human-agent communications,
they were not designed for communications
in which agents assist human in various
contexts (e.g., decision makings). An
extended set of performatives for humanagent
communications can facilitate
the application of agent technologies
to domains that involve agents assisting
humans regarding decision-making.
Chair: John Yen (Pennsylvania State University)
Email: jyen@ist.psu.edu
• Mobile Agents Working Group - MA WG.
The primary goal of this working group is to
resume work on standards for mobile agents.
Existing specifications will be improved and
extended, and the latest research results and
experiences from existing implementations
will be incorporated. Along with specification
proposals, the working group will develop
reference implementations of protocols in the
form of software components for agent toolkits.
Chair: Peter Braun (Swinburne University of Technology)
Email: pbraun@it.swin.edu.au
• P2P Nomadic Agents Working Group
- P2PNA WG. The objective is to define a
specification for P2P Nomadic Agents, capable
of running on small or embedded devices,
and to support distributed implementations
of applications for consumer devices, cellular
communications and robots, etc. over a pure
P2P network. This specification will leverage
presence and search mechanisms of underlying
P2P infrastructures such as JXTA, Chord,
Bluetooth, etc. In addition, this working
group will propose the minimal required
modifications of existing FIPA specifications
to extend their reach to P2P Nomadic Agents.
Potential application fields for P2P Nomadic
Agents are healthcare, industry, offices,
home, entertainment, and transport/traffic.
Chair: Bernard Burg (Panasonic)
Email: burgb@research.panasonic.com
• FIPA Specification Review Study Group
- FSR SG. The primary goal is to review the
existing pool of 25 Standards specifications
and the related critiques of each of these
specifications. Within its remit, it will also
cover the specifications that did not make it
to the Standards level. This review, reflecting
also on the context of the development of
the FIPA specifications, should help FIPA
to decide whether or not to upgrade and
revise it’s published FIPA specifications.
Chair: Stefan Poslad (University of London)
Email: stefan.poslad@elec.qmul.ac.uk
The FIPA Board of Directors welcomes
comments, suggestions and expressions of interest
to join the IEEE FIPA Standards Committee. For
any further information, please contact the FIPA
Board of Directors at: board@fipa.org
GGF - Global Grid Forum
Contribution by:
Omer Rana
email: o.f.rana@cs.cardiff.ac.uk
Cardiff University and Welsh eScience Centre
The 15th GGF meeting was held in Boston
(October 3-6, 2005), and brought together
researchers and implementers involved in Grid
computing. The event featured a keynote address
by Tony Hey (Microsoft Corp) on the challenges
of building a secure and robust middleware
infrastructure to support Grid computing. Tony
highlighted key challenges in the security area
18
November 2005|AgentLink News 19

STANDARDS REPORT
that still need to be addressed, along with the
need for open standards to support interoperation
between different Grid systems.
Security issues also formed the basis of the
workshop on “Leveraging Site Infrastructute for
Multi-Site Grids” (organised by Von Welch from
NCSA). A key theme in this workshop was rolebased
access control support within Grids, which
extends current efforts based on X509 certificates
and MyProxy; systems discussed included
Shibboleth and “GridShib”. Security issues were
also explored in the HealthGrid session.
The International Grid Trust Federation (IGTF)
was established at GGF 15 to enable a single
sign-on facility for access to Grid resources across
the world. This would form the basis for a first
step authentication (identity-management) and
access control mechanism that could be shared
by scientific users. The IGTF is a federation
of certification authorities or Grid policy
management authorities (Grid PMAs), and
comprises three regional Grid policy management
bodies, the Asia Pacific Grid PMA (APGridPMA),
the European Policy Management Authority for
Grid Authentication in e-Science (EUGridPMA)
and the Americas GridPMA (TAGPMA). It
has 61 members and covers 50 countries and
regions. Many Grid projects, which require largescale
deployment (such as the EU EGEE and
DEISA projects), have already started to interact
with IGTF. The availability of such security
infrastructure can be of significant benefit when
deploying agents through multiple platforms.
The use of Web Service standards continues to
play an important role in Grid computing. The
GGF15 workshop on “Web Services Performance”
focused on addressing the performance issue that
still restrict the full adoption of Web Services
in Grid projects, such as advanced parsing
techniques, alternative binary encodings of XML,
and streaming XML to address memory usage.
GGF16 is scheduled to take place in Athens,
Greece (February 13-16, 2006).
Details about the IGTF can be found at: http://
www.gridpma.org/
W3C - Semantic Web Activity
Contribution by:
Terry Payne
email: trp@ecs.soton.ac.uk
University of Southampton
Mechanisms for accessing and indexing assertions
or documents are central to both the World Wide
Web and Semantic Web frameworks. URLs
became ubiquitous during the rise of the WWW
as a means of linking and addressing documents.
On September 9th, 2005, the World Wide Web
Consortium released “xml:id Version 1.0” as
a W3C Recommendation. The specification
defines an attribute name, xml:id, that can always
be treated as an identifier and hence can always be
recognized, without fetching external resources,
and without relying on an internal subset [1].
The problem of identifying and addressing services
is necessary when communicating between
services, and communication mechanisms such
as SOAP have become well established for Web
Service communication. In August, we saw
the release of “Web Services Addressing (1.0 -
Core)” as a W3C Candidate Recommendation.
Web Services Addressing provides transportneutral
mechanisms to address Web services
and messages. The candidate recommendation
document defines a set of abstract properties
and an XML Infoset representation thereof to
reference Web services and to facilitate endto-end
addressing of endpoints in messages.
This specification enables messaging systems to
support message transmission through networks
that include processing nodes such as endpoint
managers, firewalls, and gateways in a transportneutral
manner. More details are available at [2].
The advent of the Semantic Web has witnessed
a plethora of different knowledge-bases and
languages to access them. SPARQL has
recently emerged as a uniform query language
(pronounced “sparkle”), and is also designed
to convey queries defined in other RDF query
languages. The Data Access Working Group
(DAWG) has released a Last Call Working
Draft of the SPARQL Protocol for RDF [3],
which uses WSDL 2.0 to describe a means for
conveying SPARQL queries to an SPARQL
query processing service and returning the
query results to the entity that requested them.
Comments on this draft, and on the DAWG
activities can be found at [4].
[1] http://www.w3.org/TR/2005/REC-xml-id-
20050909/
[2] http://www.w3.org/TR/2005/CR-ws-addr-core-
20050817/
[3] http://www.w3.org/TR/2005/WD-rdf-sparqlprotocol-20050914/
[4] http://lists.w3.org/Archives/Public/public-rdf-dawgcomments/
The AgentLink Roadmap
has now been published
Michael Luck, Peter McBurney, Onn Shehory, Steve Willmott
and the AgentLink Community
Agent Technology: Computing as Interaction
Technologies
Trends and Drivers
Related Disciplines
Related Techniques
Grid
Computing
A Roadmap for Agent Based Computing
Economics
Agent Technology: Computing as Interaction.
A Roadmap for Agent Based Computing
AgentLink 2005
ISBN 085432
Mathematical
Modelling
Logic
Programming
Self
Systems
Ambient
Intelligence
Organisations
Trust and Reputation
Game Theory
Comple Systems
Logic
Philosophy
Biology
Coordination egotiation Communication
Hard copies of the roadmap will be distributed to all members in due course. If
you would like extra hard copies of the roadmap to distribute at conferences or, in
particular, to industrial or other contacts, mail admin@agenlink.org with details of the
distribution, and we will endeavour to provide them.
User
Interaction
Design
Uncertainty
in AI
Robotics
PeertoPeer
Computing
Anthropology
Sociology
Service Oriented
Computing
Reasoning
and Learning
The roadmap aims to provide an assessment of the current state-of-art of agentbased
computing, and to suggest important areas for the furure research and
development, both for academia and industry.
ormal
Methods
Programming
Languages
Artificial Life
Simulation
Softare
Engineering
Interoperability Infrastructure
Organisation Design
Political Science
Marketing
Decision
Theory
Semantic Web
The Roadmap may be downloaded at
http://www.agentlink.org/roadmap/
Compiled, written and edited by
Michael Luck, Peter McBurney, Onn Shehory, Steve Willmott and the AgentLink Community
November 2005|AgentLink News 19
19

AGENTLINK TECHNICAL FORUM 3
The Third AgentLink III Technical Forum
Budapest, Hungary,
September 15-17, 2005
Andrea Omicini, Università di Bologna, Italy, andrea.omicini@unibo.it
Paolo Petta, Medical University of Vienna and Austrian Research Institute for Artificial Intelligence, Austria, paolo.petta@ofai.at
Lazlo Z. Varga, MTA SZTAKI, Hungary, laszlo.varga@sztaki.hu
The third and final Technical Forum meeting
(AL3-TF3) to be organised by AgentLink III took
place in Budapest, Hungary, co-located with the
4th Central and Eastern European Conference
on Multi-Agent Systems (CEEMAS 2005),
which generated a positive synergy throughout
the week, resulting from several stimulating and
vibrant discussions that emerged during both
breaks and in the evenings.
Thanks to the deft and professional care of the
AgentLink III staff and the local organisers,
AL3-TF3 experienced a flawless unfolding of
the choreography of the meetings of seven TF
Groups, some of which have continued from
previous meetings:
• Agent-Oriented Software Engineering
(AOSE)
• Towards a Standard Agent-to-Agent
Argumentation Interchange Format (AIF)
• Coordinating Agent Standardisation
Activities (CASA)
• Environments for Multi-Agent Systems
(ENV)
• Multi-Agent Resource Allocation (MARA)
• Programming Multi-Agent Systems
(PROMAS)
• Self-Organisation in Multi-Agent Systems
(SELFORG)
As usual, each day concluded with a plenary
session where the TFG Chairs could report
back to the community on the emerging issues
from their meetings, as well as an update on the
recently released AgentLink Roadmap.
For this final issue of the AgentLink III Newsletter,
all the AL3-TF3 groups have summarised their
meeting activities for the AgentLink audience.
By reading the reports, one should get some
clues and appreciation of the vibrant and intense
discussions, the stimulating atmosphere, and the
achievements of the TFGs, and long reports will
appear shortly on the AgentLink III web site.
We therefore wish to thank Gusztav Hencsey
and Magdolna Zsivnovszki for their assistance
as the local liaisons, as well as Peter McBurney,
Catherine Atherton, Adele Maggs, Serena Raffin,
Rebecca Earl, Michael Luck, and Terry Payne for
their invaluable help in pro-actively addressing
issues well before they could be even perceived
as problems by the scientific organisation. It
is thanks to both the AgentLink team, and the
countless hours of effort provided by TF chairs
and participants over the last two years that the
AgentLink III Technical Forum Meetings have
been such a great success: indeed, all TFGs are
continuing their work, exploiting the substantial
impetus gained!
We thank you all.
Towards a Standard Agent
to Agent Argumentation
Interchange Format (AIF)
Steven Willmott 1 (Universitat Politecnica de Catalunya, Spain),
John Fox (Cancer Research UK, UK), Dan Grecu (Cancer
Research UK, UK), Simon Parsons (New York State University,
United States), Iyad Rahwan (The British University of Dubai,
United Arab Emirates), Chris Reed (Dundee University, UK),
Dave Robertson (University of Edinburgh, UK), Nicolas Maudet
(Universite Paris 9 Dauphine, France).
http://x-opennet.org/aif/
Argumentation theories provide a powerful
framework that facilitates decision making
by interacting agents, assessing the validity of
a fact, or otherwise resolving differences of
opinion. Argumentation, in general, focuses
on interactions where parties plead for and
against some conclusion [Argmas2005], and is
an essential ingredient of inter-agent dialogue,
negotiation, persuasion and collaborative
decision-making. The field of argumentation
draws on a wide range of areas including
philosophy, natural language, rhetoric, and logic,
and multi-agent system frameworks provide a
natural home for implementing, studying and
testing Argumentation models and scenarios.
In this context, the Argumentation Interchange
Format (AIF) TFG met in Budapest, Hungary,
1 Contact person: steve@lsi.upc.edu
as part of the TFIII meeting in order to open
discussion on the development of a first cut
notation or interchange format for argumentation
and arguments which could be used as a
convergence point for theoretical and practical
work on argumentation technology. Concretely
the TFG aimed to generate:
· A collection of inputs and opinions on the
nature of such an interchange format (now
on-line on the TFG homepage given below);
· A document capturing a draft first-cut format
proposal based on inputs and discussion (due
for release in October 2005 on the TFG web
site);
· A mini-roadmap for future development of
the format.
The event was attended by approximately 20
researchers and received 12 input documents
covering a broad range of perspectives from
philosophical foundations to low level encoding
issues; thus spanning an excellent cross section
of interested communities. Discussion at the
event yielded concrete decisions on the potential
structure of the AIF and made progress on the
definition of a top-level model for core concepts.
Amongst other things, it was decided that:
· Since operational concepts vary widely,
work would focus on the definition of a set
of argumentation concepts and subsequent
reifications to different concrete syntax
instantiations.
· That the set of concepts would be structured
as a core – i.e. common to many applications
– and extensions, which provide functionality
for more specific cases (and which could be
extended by users).
20
November 2005|AgentLink News 19

AGENTLINK TECHNICAL FORUM 3
Communication
(Locutions/
Protocols)
Refers-to / (Manipulates)
Drives
Argument
Networks
(Arguments/
Relations)
set of concepts which would be defined in a core
argumentation ontology, and the links between
the circles indicate top level relationships which
would subsequently be replaced by ontological
relationships between concepts in a complete
ontology.
Manipulates
Constrains
Influences
Constrains
The general principles, top-level ontology and
other discussion results are now being compiled to
form the first draft document (with an anticipated
completion date of the end of October, 2005).
Other future plans include ongoing discussion via
the associated AIF Google-Groups forum 2 , as well
as potential future events collocated with relevant
conferences/workshops.
· That no separation would initially be made
between the use of the format between static
tools (e.g. import / export of argumentation
schemas) and use of the format in agent-agent
communication; however, it may be necessary
to make this distinction in the future.
After discussion and analysis of the inputs, core
concepts in argumentation were grouped into three
top level types: concepts to do with arguments
themselves and relationships between arguments
Context
(Participants/
Theory
Fig 1: Grouping of top concepts for Argumentation Formats resulting from the TFIII Budapest Meeting.
(labelled argument networks); concepts to do
with communication of arguments – locutions
and protocols for argumentation (labelled
communication); and finally concepts belonging
to the context of argument such as participants,
rules/theory applying, groundings of terms and
so forth (labelled context). While other concepts
relevant to argumentation may fall outside of
this scope, these broad categories appeared to
capture the majority of relevant items. Figure 1
shows the concept groups: each circle represents a
The TFG organisers would like to greatly thank all
those that participated – both in providing input, in
attending the event and for post-event discussions.
The organisers and participants would also like to
heartily thank the European Commission-funded
project - AgentLink III, the Agentlink III board,
and the TFG organisers for providing this unique
opportunity for joint work. Lastly, the event would
not have been possible without the support of the
European Commission funded ASPIC project
(http://www.argumentation.org/ which provided
technical support/coordination and impetus for
the activity.
References
[ArgMAS05] Workshop on Argumentation for
Multi-Agent Systems, organised in conjunction
with the International Conference on Autonomous
Agents and Multi-Agent Systems (AAMAS) July,
2005, Utrecht, The Netherlands.
Agent-Oriented Software Engineering (AOSE)
Massimo Cossentino, ICAR - Consiglio Nazionale delle Ricerche, Italy, cossentino@pa.icar.cnr.it
Juan Pavón, Universidad Complutense Madrid, Spain, jpavon@sip.ucm.es
The third AOSE TFG meeting was aimed at
completing open activities from the previous two
meetings, and was enriched by the participation
of invited speakers from Australia and USA, who
introduced new arguments and perspectives to
the discussion. The main topics of this meeting
(apart from the talks) were the refinement of the
MAS meta-model defined in Ljubljana, at the
AL3-TF2 meeting, and the evaluation of several
AOSE methodologies through a questionnairebased
approach.
2 http://groups.google.com/group/Agent-Argumentation
The meeting was organised in three sessions:
1. Presentations regarding agent-related topics
from outstanding researchers in the area.
2. Refinement of the Ljubljana MAS metamodel.
3. Methodologies evaluation: results of the
questionnaires, comparison and debate.
The presentations were given during the first
session, commencing with a talk from G. Fortino
on an agent-based approach for the management
of distributed workflows. This was followed
by talks by L. Padgham on the Prometheus-
ROADMAP methodology, and B. Henderson-
Sellers, who described the FAME project that
deals with a method engineering framework
founded on a powertype-based metamodel.
In the second part of the session, J. Odell reported
about his most recent industrial experiences
on the use of AUML. Other talks included:
A. Garcia, who presented his work on the
introduction of aspects in MAS development and
followed this with a discussion on both lessons
learned and open issues; A. Molesini, who talked
about the importance and role of artefacts in
Agent-Oriented Software Engineering in general
and more specifically in the SODA methodology;
and M. Cossentino, who reported about the
experiences gained in the FIPA Methodology
Technical Committee, where the method
engineering paradigm has been adopted in order
to converge towards an agent-oriented design
methodology standardisation proposal.
The second day was divided into two sessions.
The Ljubljana MAS meta-model was considered
during the first session; where the introduction of
new elements to this model was discussed. The
November 2005|AgentLink News 19
21

AGENTLINK TECHNICAL FORUM 3
work started from a proposal by P. Giorgini and
M. Cossentino about the introduction of the Goal
Element, and the resulting outcomes included:
• A refinement of the definition of the Task
Element - initially introduced at the Ljubljana
meeting, and now defined as “the activity (set
of activities) motivated by some coherent
reason”;
• A definition of Goal as “a specification of a
state of the world that may be achieved or
maintained by the agent”.
• An initial definition was drafted for Plan as
follows: “A plan specifies how a goal can be
achieved”.
The session concluded with the agreement that
the MAS meta-model deserves further work,
which will proceed through the mailing list.
The second session included an active and
lively debate on Methodologies Evaluation,
with emphasis on a comparison of seven
different methodologies that emerged from
the results of a questionnaire presented by
Juan Pavón (the questionnaire was made
available through the AOSE TFG website). It
was adapted from a paper by Khanh Hoa Dam
and Michael Winikoff, which was presented
at the AOIS03 workshop held in Melbourne,
2003. The main comments regarded the
impact of the relationship between people
(who compiled the questionnaire) with the
methodology they evaluated, on the evaluation
results. Judgements were typically more critical
from users of methodologies (as would be
expected), and as a consequence, it was agreed
that the lack of a user-filled questionnaire
for some methodologies partially limited the
quality of the presented results. However, the
overall experience was useful in improving the
questionnaire for future editions. The type of
questionnaire used will be one of the main
tools used as part of the AOSE evaluation
framework, which would also include case
studies and specific metrics.
Coordinating
Agent Standardisation
Activities (CASA)
Monique Calisti, Giovanni Rimassa, Whitestein Technologies AG, Switzerland, {mca;gri@whitestein.com}
Omer Rana, Cardiff University and Welsh eScience Centre, United Kingdom, o.f.rana@cs.cardiff.ac.uk
Stefan Poslad, Queen Mary University, United Kingdom, stefan.poslad@elec.qmul.ac.uk
Terry Payne, University of Southampton, United Kingdom, trp@ecs.soton.ac.uk
James Odell, Agentis Software, Inc, France, omg@jamesodell.com
The main objectives of the AgentLink TFG-
CASA are:
1. To assess the importance of ongoing agentdedicated
and agent-related standardisation
activities.
2. To clarify how existing standardisation bodies
and groups relate and possibly contribute to
establishing stronger liaisons.
3. To understand and position the role of
AgentLink for ongoing and future agentdedicated
and agent-related standardisation
work.
The key aim is to clarify and strengthen the
importance of standardisation activities for
software agent technology, an issue of central
importance to many academic and industrial
organizations involved in the production of agentoriented
software and associated techniques.
Such standardisation activities are critical to the
commercialisation of agent-based research, and
towards strengthening interaction within the
agent community.
Towards this end, the TFG-CASA took a highlevel
perspective on standardisation activities at
the AL3-TF3 meeting in Budapest, by bringing
together researchers and organizations directly
involved in a number of standardisation bodies,
including IEEE FIPA, OMG, GGF, and W3C.
Participants met at the AL3-TF3 to exchange
information and discuss the importance of creating
synergies between the different standardisation
bodies. This included members already active
within standardisation bodies, and individuals
involved in identifying application requirements,
addressing possible infrastructure themes, and
promoting best practice and experience towards
agent-oriented computing. From this perspective,
the AgentLink TFG-CASA activity was an
essential component in helping to guarantee and
promote such meta-level and standardisationagnostic
coordination across various groups.
Additionally, the TFG-CASA hosted several
coordination sessions involving individuals from
other TF groups, including Agent Argumentation
Interchange Format (AIF), Agent-Oriented
Software Engineering (AOSE), and Programming
Multi-Agent Systems (PROMAS). The purpose
of these sessions was to gain an understanding
of where standards are required for the various
TFGs. The main issues that were identified are
listed below, and are expected to be addressed
in the context of the new IEEE FIPA Standards
Committee (http://www.fipa.org).
• Methodology: Method engineering enables
developers to customize system development
methodologies to fit the needs of an agent
project. The general idea is to achieve
modularity through the integration and
customization of multiple methodologies.
Possible areas for standardisation could
include a method fragments meta-model, a
method engineering repository meta-model,
and possibly a MAS infrastructure metamodel.
• Service-Oriented support for Agents: Web
Services do not provide an adequate model for
capturing the expressivity intrinsic to existing
agent models, such as ACL performatives.
A question that needs consideration is how
to structure the formal relationship between
agents and web services. Semantic Web
Service standards are clearly relevant and
were discussed, such as WSDL-S, OWL-S,
and WSMO.
• Auditing multi-agent systems: This would
require monitoring the interactions between
multi-agent systems at different levels, such
as the behavioural level, infrastructure level,
and so on. The notion of debugging agent
systems was considered, and distinguished
from the validation of agent behaviors. It
was generally felt that multi-agent system
debugging was of significant interest, and
required greater consideration.
• Formulation of application-driven
requirements on agent system
capabilities: This involves identifying
the functionalities necessary in an agent
system from the perspective of application
development. This implies proactive
engagement with the applications community
to elicit requirements and identify current
standards that may be impacted them.
• Interaction between agents and
environments: Agent systems rarely operate
in isolation from an environment and as
such, clearer statements concerning this
relationship need to be established. This
requires environment modeling, and defining
the nature of the agent-environment interface.
An important distinction was made between
the wrapping of legacy systems, and the
embedding of multi-agent systems within
larger applications.
Given that AgentLink III will officially end in
December 2005, several TFG-CASA promoters
and participants are considering alternate ways
in which it would be possible to continue this
meta-level coordination activity for agent-related
standardisation bodies. Of course, AgentLink IV
would represent the most ideal context!
Further details, and copies of the presentations
given at the TFG-CASA in Budapest can be found
at http://www.agentlink.org/casa. Feedback and ideas
are welcome. Please send your input to the TFG-
CASA mailing lists: tfg-casa@agentlink.org.
22
November 2005|AgentLink News 19

AGENTLINK TECHNICAL FORUM 3
Environments For Multiagent Systems (ENV)
Danny Weyns and Tom Holvoet, Katholieke Universiteit Leuven, Belgium,
{danny.weyns, tom.holvoet}@cs.kuleuven.be
Recent efforts, including past TFG meetings
(ENV) and the AAMAS workshops on
Environments for Multiagent Systems (E4MAS)
in New York and Utrecht have firmly established
the Environment on the research agenda of the
MAS community. As a result, the TF3-ENV in
Budapest had two main goals: to discuss actual
research issues on environments, and to bridge
ongoing research on environments with other
related research lines; in particular programming
multiagent systems (PROMAS) and selforganisation
(SELFORG).
Several core topics have been identified at
previous meetings, and these formed an
interesting starting point for discussions. The
first topic was: “Definition and Responsibilities
of the Environment”. Previous exploration had
yielded a basic understanding of the dual role of
the environment in MAS: on the one hand, the
environment is an essential part of any MAS; on
the other hand the environment is an exploitable
design abstraction for building MAS. During the
discussion, a number of essential aspects were
identified that should be part of a definition of
environment:
• the environment is a first-order abstraction in
MAS;
• the environment provides the surrounding
conditions for agents to exist;
• the environment enables and regulates both
interaction among agents and access to
resources.
Several environment responsibilities were
discussed: (1) the environment structures the MAS
(physical, communication, social structuring);
(2) the environment provides life-cycle support
and management (wrt. resources, agents, dynamics
in the environment); (3) the environment
provides observability and accessibility; (4) the
environment regulates the MAS (access control
to resources, interaction mediation, norms, laws,
etc.); and (5) the environment provides a domain
ontology. Rather than “being responsibilities of an
environment”, these aspects should be interpreted
as “responsibilities that can be assigned to the
environment”.
The first session concluded with a critical
self-reflection on whether or not traditional
middleware mechanisms already provide the
necessary components to support environments
in MAS The conclusion was no: (1) from a
middleware perspective, MAS require a very
specific set of middleware mechanisms of
which some have not yet been developed (e.g.,
support for societies). Moreover, the integration
of the required mechanisms is itself a research
challenge; (2) designing application-specific
environments (from software architecture down
to implementation) on top of middleware is a
huge research challenge.
The second discussion topic was: “Mechanisms and
Opportunities of the Environment”. Mechanisms
are building blocks of environments that
encapsulate reusable functionality for designing
MASs. The discussion mainly focused on defining
a taxonomy of mechanisms for environments,
which prove to be extremely difficult, and it is
unclear whether such a taxonomy would be of
any use at all at the moment, without a thorough
study of the literature.
During the third discussion session, the group
focussed on “Environment Engineering”.
Three aspects of environment engineering
were discussed: (1) modelling: a basis for
environment engineering is a reference model
that provides a generic functional decomposition
of environment; (2) software architecture: this
describes how the reference model is mapped
onto software elements; and (3) methodology:
should environment engineering be integrated
with existing agent-oriented software engineering
methodologies, or should a general software
engineering perspective be taken. With respect to
the latter, a number of architectural approaches
for environments were discussed, including
modelling the environment as a set of mediating
artefacts that agents can use, or modelling it as a
set of modules that represent different functional
concerns/responsibilities of the environment,
such as communication, perception, action and
interaction.
The topic of the last discussion session was
“Multiagent modelling and simulation: the
role of environment”, which focussed on two
activities regarding environment engineering in
the context of simulation: the first being that the
domain that is simulated must be modelled; and
the second, that the simulation infrastructure
must be engineered. However, several concerns
regarding the relationship between these two
activities were identified, including timing or
simultaneity of actions. An important challenge
in this domain is to develop a reference model
for environments specific to MAS simulation that
relates the concerns to one another.
Now the community has established a basic
understanding of the role of the environment
in MAS, these insights are being consolidated
to provide a basis for future research on
environments. An upcoming JAAMAS Special
Issue on Environments for MAS will aim to
communicate the acquired knowledge of research
on environments to the MAS research community
in general, and focus on environment engineering
in particular. Future meetings will continue, with
the next, E4MAS 2006 being held at AAMAS
2006.
ENV JOINT SESSIONS
The main goal of the joint session with
the PROMAS group was to discuss the
programming of environments in MAS.
A fruitful discussion on the very concept
of environment and its particular role
in MAS emerged from this, with one
interesting topic of discussion being the
role of the environment in organisational
issues of MASs. Alternate viewpoints
were argued: some believed that the
role of the environment should be
kept to a minimum, providing just the
medium for agents to interact; whereas
others argued that the environment
should be assigned an active role in the
organisation of a multi-agent system
(as in electronic institutions).
During the joint session with SELFORG,
the group had an open discussion
on the role of environment for selforganising
mechanisms. Whilst it was
accepted that the environment plays an
essential role in many self-organising
systems, from the perspective of the
designer of self-organising systems,
this role cannot be separated from the
role of the agents in self-organising
mechanisms. Such mechanisms
that involve the environment should
therefore be considered as an integral
concern.
November 2005|AgentLink News 19
23

AGENTLINK TECHNICAL FORUM 3
Multiagent Resource Allocation (MARA)
Ulle Endriss
University of Amsterdam, The Netherlands
ulle@illc.uva.nl
The allocation of resources is a significant issue
in both Computer Science and Economics
disciplines. To emphasise the fact that resources
are being distributed amongst several agents
and that these agents may influence the
choice of allocation, the field is also known as
Multiagent Resource Allocation. The questions
investigated by computer scientists are often
of a procedural nature (“how do we find an
allocation”), while economists are more likely
to concentrate on qualitative issues (“what makes
a good allocation”). A comprehensive analysis
of the problem at hand, however, requires an
interdisciplinary approach. It is here where the
Multi-Agent system paradigm offers an excellent
framework in which to study these issues.
The Technical Forum in Budapest has been the
second meeting of the AgentLink Technical
Forum Group on Multi-Agent System
(TFG-MARA). The group emphasises the
interdisciplinary aspect of the field and aims at
providing an environment for the exchange of
new ideas and also at fostering collaborations
between different European research groups.
The Budapest meeting hosted over 20 researchers
from a dozen different institution countries as
diverse as Hungary, France, Germany, Romania,
Spain, the Netherlands, and the UK.
After the first TFG-MARA meeting, held in
Ljubljana earlier in 2005, ten of the participants
came together to write an extensive survey paper
on Multi-Agent Resource Allocation, which
is both available on the Internet and due to be
published in the Informatica journal. This survey
was presented and discussed during most of the
first session of the meeting in Budapest.
The MARA Survey reviews various languages
used to represent the preferences of agents over
alternative allocations of resources, as well as
different measures of social welfare, to assess
the overall quality of an allocation. It also
discusses pertinent issues regarding allocation
procedures and presents important complexity
results. The presentation of theoretical issues was
complemented by a discussion of software packages
for the simulation of agent-based market places.
The survey also introduces four major application
areas for Multiagent Resource Allocation, namely:
Industrial Procurement; sharing of satellite
resources; manufacturing control; and Grid
computing.
At the previous TFG-MARA meeting, we had
identified Fair Division as a topic that is of
particular interest to the community, and that could
greatly benefit from a closer interaction between
computer scientists and economists. To address
this requirement, we decided to invite an expert
in the field to deliver an in-depth presentation
from the viewpoint of Economics at the Budapest
meeting. Thibault Gajdos, CNRS researcher at
the EUREQua research group at the University
of Paris 1, gave an invited tutorial on Fairness
and Uncertainty. The tutorial examined various
connections between decision theory and social
choice and demonstrated how decision theory
can provide a useful perspective in the context
of the analysis of resource allocation problems,
and more specifically, for the identification and
implementation of fair allocations.
The remainder of the meeting was filled with
several contributed talks reporting on recent
research activities within the European MARA
community. The topics covered included
complexity issues, applications in grid computing
and crisis response, combinatorial auctions,
heuristics and experimentation in negotiation,
fair division procedures, connections between
sociological theory and Multi-Agent systems, and
Multi-Agent simulation.
Full details of the presentations, as well as the
MARA Survey, are available at the MARA website
(http://www.illc.uva.nl/~ulle/MARA/).
Programming Multi-Agent Systems (PROMAS)
Mehdi Dastani, Utrecht University, The Netherlands, mehdi@cs.uu.nl
Jorge Gómez Sanz, Universidad Complutense de Madrid, Spain, jjgomez@sip.ucm.es
The aim of the PROMAS technical forum was
to promote collaboration between industry and
academy by providing an informal platform where
attendants can present their work and discuss
issues related to the implementation of multi-agent
systems. A little under 40 participants took place at
the meeting in Budapest, which was organised into
four sessions; each focussing on one of the topics
below:
• “Programming languages/issues for individual
agents”. Speakers: Viviana Mascardi (University
of Bologna), Birna van Riemsdijk (Utrecht
University), and Robert Ross (Universität
Bremen).
• “Programming languages/issues for multiagent
environments”. Speakers: Danny Weyns
(Katholieke Universiteit Leuven) and Mirko
Viroli (Università di Bologna a Cesena).
• “Programming languages/issues for multiagent
organisation/coordination structures”.
Speakers: Andrea Omicini (Università di
Bologna a Cesena), Mehdi Dastani (Utrecht
University), and Jorge Gómez (Universidad
Complutense de Madrid).
• “Debugging and IDEs for multi-agent
systems”. Speakers: Laszlo Gulyas (AITIA
company), Juan Botía (Universidad de
Murcia), and Alexander Pokahr (Universität
Hamburg).
The development of agent-oriented programming
languages is still a main issue in this community.
Many agent-oriented programming languages
have been proposed, e.g., IMPACT, SOCS,
3APL, Jason, Jadex, CLIMA, or Jack; however,
few have reached a level of maturity necessary
for widespread adoption, as there is still a need
for more powerful agent-dedicated programming
constructs, debugging facilities, formal
ontologies, primitives to deal with distributed
code, and development environments. Also, we
need to explore programming principles applied
to conventional programming languages, like
modularity, compositionality, and ease of use, and
introduce them in agent-oriented programming
languages. The importance of primitives for
invoking other languages, like Java and C++,
from an agent program were discussed, as was the
need for guidelines and manuals, and examples
that illustrate how these languages should be
applied.
A joint session was held with the ENV technical
forum. Here, the focus was the implementation
of the environment in multi-agent systems. The
environment of multi-agent systems is different
from multi-agent infrastructures, individual
agents, and their organisation structures. The
ENV-TFG presented two possible approaches
to the implementation of the environment for
multi-agent systems: by means of a set of artefacts
which can be exploited by agents to achieve their
24
November 2005|AgentLink News 19

AGENTLINK TECHNICAL FORUM 3
goals, and by a dedicated component like the
Delta framework. In either approach, laws can
be built in to codify, for example, the effect of
actions.
The third session was about coordination and
organisation of multi-agent systems. It was
argued that the organisation of multi-agent
systems is a specific kind of coordination, and
thus the implementation of such organisations
can be realised by means of various technologies;
including tuple centres, coordination artefacts,
and message oriented communications.
Finally, the meeting concluded with a debate
about the maturity of debugging techniques
applied to multi-agent systems. There are complex
systems such as agent-based social simulations,
which require specialised tools able to detect
bugs and errors through thousands of messages.
An example of such a tool is ACLAnalyzer, a
JADE third-party add-on. Moreover, inspection
and debugging of the internal states of individual
agents could be easier using concepts inspired
by agent meta-models and BDI models. Based
on conventional software engineering expertise,
a wish-list of desirable debugging features
was identified, including breakpoints, asserts,
automated testing, and tracing facilities. As
methods for distributed debugging were less well
understood, the community agreed to examine
them in preparation for future meetings.
More details, as well as presentation slides, can be
found at the PROMAS web page:
http://www.cs.uu.nl/~mehdi/al3promas.html.
Self-Organisation in Multi-Agent Systems (SELFORG)
Di Marzo Serugendo 1 , Giovanna, University of Geneva, Switzerland, dimarzo@cui.unige.ch
Gleizes, Marie-Pierre, University Paul Sabatier, France, gleizes@irit.fr
Karageorgos, Anthony, University of Thessaly, Greece, karageorgos@computer.org
Decentralised control, self-organisation and
emergent behaviour are issues of major interest
in distributed complex systems, such as MAS.
The aim of the Self-Organisation in Multi-
Agent Systems (SelfOrg) TFG was to discuss
and investigate these issues through examination
of representative case studies and organisation
of interdisciplinary collaborative interaction
activities.
The TFIII meeting in Budapest was the third
meeting of the TFG Self-Organisation in MAS,
and concentrated on three main subjects: 1)
finalising the definition of the emergence concept;
2) extending the understanding of existing selforganisation
mechanisms via presentations of
relevant work; and 3) finalising the case study
exercises, consolidating lessons learned, and
highlighting issues and directions requiring
further work.
The first part of the meeting focused on the case
studies that were developed during the previous
meetings. The decision to establish three case
studies was taken at the first meeting in Rome.
Subsequently, the second meeting in Ljubljana
focused on the description of the selected case
studies and on propositions of agent-based selforganisation
approaches for modelling and/or
for implementing solutions to the problems they
describe. In Budapest, Paul Verstraete presented
a solution to the Manufacturing Control Case
Study, where self-organisation is examined from
an environment perspective. After considering
this alternative view, the question of whether
or not the selected case studies could be useful
as reference points (benchmarks) for evaluating
self-organisation mechanisms was discussed. The
responses received indicated agreement on the
selection, but suggested that additional work
was required if the case studies were to better
serve the benchmark role. Such improvements
include extending and detailing the case study
descriptions (i.e. input data files) and clarifying
the measurements needed for a proper evaluation.
Furthermore, it was agreed that additional case
studies should be proposed, that new solutions to
the case studies should be contributed and added
to the TFG web site, and that an important final
objective of this work should be to compare
different mechanisms on the same case study.
The second part of the meeting consisted of
presentations of three self-organising applications.
Luca Gardelli from University of Bologna, Italy,
presented cognitive stigmergy; an extension of the
well-known stigmergy mechanism for cognitive
agents. Gerald Silverberg from Maastricht
University, MERIT, The Netherlands, presented
a model of innovation dynamics based on
invasive percolation by adding endogenous R&D
search by economically motivated firms. The last
presentation by Giuseppe Vizzari, University
Bicocca, Milan, Italy, focused on situated MAS
for the detection of emergent links in adaptive
Web sites.
A large part of the meeting was devoted to fruitful
discussions concerning topics relevant to selforganisation,
and a trial to define the concept of
“emergence” in the context of artificial systems,
such as MAS. To this end, the group had to
discuss the emergence concept from three points
of view: The first one was regarding engineering,
and focused on questions such as:
• Can we control systems with emergent
functionality
1 Giovanna Di Marzo Serugendo was not able to participate at this third meeting due to an important
event beyond her control.
• What is the key issue in the design of systems
with emergent functionality
• Is the environment fundamental in these
systems
The second view concerned analysis of the selforganisation
results with questions including:
• Do we need an observer
• If yes, where is the location of the observer
• What is its role
The last was the operational point of view, with
questions such as:
• Can we speak about emergence in artificial
systems
• Why and under which conditions can we
qualify that the phenomenon emerges in an
artificial system is emergent
• How can we define the notion of emergence
in an artificial system (what could be an
operational definition)
These questions were aimed at guiding
participants to form a definition of the emergence
concept. Although an initial definition was given,
additional discussion will be needed to refine and
assess this definition based on further insight
gained from additional elaboration on case
studies.
Beyond TFIII, the activities of the group will
continue remotely. Our aim is to improve the
case studies towards benchmarks which can be
used by the community to evaluate and compare
self-organisation mechanisms. Furthermore, we
would like to propose a detailed formal definition
of the concept of emergence in artificial systems.
Finally, the results of the TFG’s collective efforts
will be refined and comprehensively presented
in a text-book about self-organising artificial
systems, planned to appear in 2006.
November 2005|AgentLink News 19
25

EVENT REPORT
Semantic Web Days Munich
6-7 October, 2005
Andrea Kulas
webXcerpt Software GmbH
Germany
ak@webxcerpt.com
ALSO IN THIS SECTION...
The 1 st International Workshop on Context for Web Services
The 8 th Biennial Israeli Symposium on Foundations of
Artificial Intelligence
E4MAS 2005: Environments for Multiagent Systems
Int. Workshop on Agent-Based Models for Economic Policy Design
The 2005 IEEE/WIC/ACM Int. Joint Conference
on Web Intelligence and Intelligent Agent Technology
The 3 rd Workshop on Agents Applied in Health Care
Agents in Space
1 st Int. Workshop on Security and Trust Management
Alain Léger,
France Telecom R&D
France
Alain.leger@rd.francetelecom.com
Semantic Web Days 2005 provided a forum for
innovative companies and research institutions
with a strong desire to accelerate the uptake
of Semantic Web technologies to meet and
exchange experiences, challenges and solutions.
With attendees travelling from all over Europe
and countries as far a field as the US, and Korea,
the main goal of the two-day conference, held
in October in Munich, was to present the latest
Semantic Web technologies which are very
promising or already in use. This international
event was organised jointly by the EU Network
of Excellence REWERSE (REasoning on the
WEb with Rules and SEmantics) and by the
EU Network of Excellence Knowledge Web
(Realizing the Semantic Web).
One of the central questions addressed over
the two days focused on the degree of maturity
of Semantic web technologies i.e. Ontology,
Reasoning, Business rules, facing real needs
in industry i.e. data, services and business
automated integration. The message, which
was also discernable in the concluding panel
discussion, was that procedures to describe
semantic coherences presently mainly exist in
the form of pilot studies, but as yet there has
been little uptake of such procedures in major
industrial projects. Although only a limited
number of projects have so far been realized
with Semantic Web technologies, the power and
benefits of using Semantic Web approaches, as
well as the need for catch-up, became obvious as
Semantic Web Days 2005 proceeded. The fact
that there were more than 100 participants at the
Semantic Web Days illustrated that there can be
no doubt of the high potential for the application
of Semantic Web technologies; especially given
the already convincing pre-deployment of several
Semantic Web applications.
At present, established projects are focused
on the development of corporate Knowledge
Management, Telecommunication applications,
Business integration for agile enterprises and
Health care. Furthermore, what with more
than 12 companies and institutions featured
at the exhibition (as part of the Semantic Web
Days event), it was clear that a number of SMEs
already have products on the basis of Semantic
Web technologies.
The keynote speaker was Hermann Friedrich,
who is responsible at Siemens AG for the
development of the knowledge database used
internally in the company. He spoke about
initial experiences with the Semantic Web: “We
also have the possibility to classify documents
in content management, but with ontologies
it is possible to demonstrate more complex
coherences.”
Modeling complex causal coherences was also a
focus for Thomas Syldatke, one of the panelists at
the concluding panel discussion “Earning Money
with Semantic Web Technologies – examples of
best practice and outlook for promising projects
of the future”. He is responsible at Audi AG for
the development of software for testing engines,
and noted that: “…this is a very complex
procedure, there are a lot of rules which we are
integrating with ontologies which need to be
considered…”. Ten percent of the rules have
been integrated in the first pilot application,
which was finished in June. He then went on to
say“…we asked the system which should be the
next control step for the engine and compared
the result with the program which has been in
use already…”. Plans are currently underway
to integrate 100 percent of the rules with the
ontology by December this year.
An exciting field for demonstrating the power
of semantic technologies is the integration and
application of rules. Researchers at the University
of Munich demonstrated the possibilities with
a geographical information system based on
scalable vector graphics (SVG). Ontologies
model rules which form the basis for adapting
the graphics. Using this system, different views
AgentLink III ATC – 2005
Stockholm, 18 th October
The AgentLink III annual Agent Technology
Conference (ATC) facilitates the sharing of
experiences in deployment of agent-based
solutions among commercial organisations
and with discussion of the benefits this
technology brings their businesses. Several
talks were presented, covering a range of
issues from the technologies themselves to
commercial relevance and strategy
• Modeling the Aerospace Aftermarket with
Multi-agents Systems
Ken Woghiren (Lost Wax)
• Intelligent Agents for Networked Home
Environment
Götz Brasche (Microsoft Innovation
Center)
• Contribution of Agent Technologies to
Flexibility in Production and Product
Development
Kurt Fessl (Profactor)
• Administrable Services in a Mobile, Dynamic
World
David Marples (Telcordia)
• Intelligent Agents for Banking and Insurance
Industries
Chris van Aart (Y’all)
• How to Sell Agent Technology into
Commercial IT
James Odell
(James Odell Associates)
In the afternoon, demonstrations of existing
agent-based solutions were presented
by: CalicoJack, Certicon, Erxa, Lostwax,
Magenta, Profactor, Rockwell and Y’All.
http://www.agentlink.org/agent-Stockholm/
26
November 2005|AgentLink News 19

EVENT REPORT
can be modeled. On being asked if they were
just rebuilding existing tools, such as Google
Earth, the developer Hans Jürgen Ohlbach
replied “… when compared with conventional
navigation systems, we have the advantage of
being independent of certain tools”.
Four workshops on the topics “Industrial
Applications of Semantic Web technologies”,
“Vocabularies and Rules for Enterprise
Applications”, “Semantic Web Services
in Industry” and “Semantic Web for Life
Sciences”, as well as a presentation on
“Geospatial Information Processing for the
Web” demonstrated more in detail that there is
high potential for semantic technologies such as
ontologies and rules to provide better and faster
solutions than conventional methods. Pilot
projects presented at the Semantic Web Days
proved those potential advantages.
Overall, the organizers have been extremely happy
with the success of the event, and their impression
has been backed by the tenor of the participants,
who wished to establish this conference so that
developments and applications of Semantic Web
Technologies could be communicated between
companies and research institutions in a timely
manner. Last, but not least, the participants list
of the Semantic Web Days demonstrated that the
goal of exchange between business and research
has been achieved, as half of the participants
came from companies while the rest came from
research institutions. Qualified presentations,
the participants themselves, nice surroundings
and many opportunities for social networking
contributed to the success of the two days.
The First International Workshop on Context for Web Services (CWS’05)
Djamal Benslimane, Lyon 1 University, France, djamal.benslimane@liris.cnrs.fr
Chirine Ghedira, Université Claude Bernard, France, chirine.ghedira@iuta.univ-lyon1.fr
Zakaria Maamar, Zayed University, U.A.E, zakaria.maamar@zu.ac.ae
The First International Workshop on Context
for Web Services (CWS’2005) that was held
in conjunction with The Fifth International
and Interdisciplinary Conference on Modeling
and Using Context (CONTEXT’2005) took
place in Paris, France, on July 5th, 2005. The
workshop aimed at bringing together researchers
and practitioners who are actively engaged in the
fields of information systems, context and Web
services (provide more detail on the themes).
Nearly 20 people attended the workshop.
The workshop received 17 submissions, from
which 7 were selected for presentation and
inclusion in the workshop’s proceedings. A
keynote presentation by David L. Martin from
SRI International in California is also featured
in the workshop’s program. David L. Martin has
worked extensively in the fields of software design
environments, software agent frameworks, and
the Semantic Web. In the latter area, he is the
chair of the OWL-S coalition and a co-chair of
the Semantic Web services language committee
of the Semantic Web services initiative (SWSI).
Several topics have been discussed during the
one-day workshop including ontologies and
context for Web services, context-based semantic
matching for Web services composition, security
of context during Web services interactions, just
to cite a few.
Web services are nowadays emerging as a major
technology for deploying automated interactions
between distributed and heterogeneous
applications. In general, composing Web services
rather than accessing a single Web service is
essential and provides better benefits to users.
Composition primarily addresses the situation
of a user’s request that cannot be satisfied by any
available Web service, whereas a composite Web
service obtained by combining available Web
services might be used. Several questions have
been raised regarding Web services composition
and execution including which businesses have
the capacity to provision Web services, when and
where the provisioning of Web services occurs,
and how Web services from independent parties
coordinate their activities during execution so that
conflicts are avoided. To address some of these
questions, it is recommended to consider the
context in which the composition and execution
of Web services occur. Context is generally
perceived as the information that characterizes
the interaction between humans, applications,
and the surrounding environment. From a Web
services perspective, it is expected that context
should define a set of common data about the
current status of a Web service and its capability
of collaborating with other peers (i.e. webservices
or also non–web services), possibly enacted by
distinct providers.
The keynote speaker’s talk was geared towards
semantic Web services technology and the
mechanisms it provides to deal with context. He
considered what kinds of contextual knowledge
Web services-based systems need to handle, shown
how selected semantic Web services approaches
might be applied to these systems, and speculated
about what else might be needed. According
to David (Martin), a Web service is normally
conceived as a neatly encapsulated module of
functionality that can be easily reused, so long
as the inputs, outputs, and messaging protocol
are conformant with its description. However,
to support automated discovery and selection
of world-changing Web services, for example,
Web service descriptions must be unambiguous
about what situations will guarantee successful
uses of Web services, and what new situations
will result from those uses. In some scenarios,
a Web service’s behavior may vary with time,
location, user history, pre-existing contractual
commitments, and so on. Therefore, descriptions
of such distinctions can quickly become complex.
Moreover, many aspects of Web service use and
management may require knowledge that is not
normally captured in Web service descriptions.
When considering the full range of Web servicerelated
activities, it becomes clear that dealing
with context is a major challenge, requiring far
greater expressiveness and reasoning capabilities
than are supported by the current widely accepted
building blocks of the Web services stack.
Titles of some papers presented at the workshop
are as follows: “Generic framework for the remote
control of high technology instruments based on
composite Web service supporting user context”
(etc) by C. Gravier and J. Fayolle (name their
institution France), “Context management for
adaptive information systems” by C. Cappiello,
M. Comuzzi, E. Mussi, and B. Pernici ((name
their institution Italy) and “A system architecture
for context-aware service discovery” by C.
Doulkeridis and N. Loutas ((name their institution
Greece). The complete list of presented papers is
available at http://www710.univ-lyon1.fr/~dbenslim/
CWS05. The post-proceedings of the workshop
will appear as Electronic Notes in Theoretical
Computer Science (ENTCS) of Elsevier.
The workshop was supported by The European
Co-ordination action for Agent-based
Computing (AGENTLINK), The Object Web
Consortium, The Lyon Research Center for
Images and Intelligent Information Systems,
The ISTASE institute, and The Computer
Science Department of the IUT A Institute of
Lyon I University.
November 2005|AgentLink News 19
27

EVENT REPORT
The Eighth Biennial Israeli Symposium
on Foundations of Artificial Intelligence (BISFAI)
Claudia V. Goldman
University of Haifa
Israel
clag@cri.haifa.ac.il
The topic of the first satellite workshop was
recommender systems. Participants enjoyed a full
day of invited talks presenting the state-of-theart
in recommender systems and their prospects.
Invited speakers included Prof. Joseph A.
Konstan from the University of Minnesota, Prof.
Robin Burke from De Paul University, Prof. Luc
Moreau from the University of Southampton,
Prof. Derek Bridge from University College
Cork, Prof. Oded Shmueli from the Technion
and Prof. Daniel Lehmann from the Hebrew
University of Jerusalem. The workshop ended
with demonstrations of several systems including
a guided search assisting a user in an electronic
commerce scenario and a system that interacts
with a user with the purpose of recommending
music and refining its recommendation
mechanisms.
Prof. Martin C. Golumbic, director of the Caesarea
Rothschild Institute (CRI), and Dr. Gideon
Ariely, representing the Ministry of Science
and Technology, opened the main conference.
The first invited talk, from Prof. Yoav Shoham
of Stanford University, discussed the topic of
multi-agent learning and distinguished between
different types of problems and perspectives.
This was followed by two regular talks on game
theory and preferences. The second invited talk,
by Prof. Richard
Korf from UCLA,
enlightened us
about new search
algorithms for
successful queries
in very large data
domains. This was
followed by two
parallel tracks:
one on problem solving and the other on control
of agent behaviour and learning in uncertain
environments. The first day of the symposium
ended with two further parallel sessions: one on
E-commerce agents including talks by the leading
groups in Israel from the Hebrew University and
Bar Ilan University, and the other on information
retrieval.
Prof. Shlomo Zilberstein’s invited talk opened the
second day with a presentation on decentralized
decision-making under uncertainty. The
subsequent regular talks developed this theme,
one with a presentation of a heuristic algorithm
to solve infinite horizon decentralized POMDPs
and the other on learning how to communicate
in decentralized environments. The morning
session continued with two parallel tracks on
collaborative activity in multi-agent systems and
on information and machine learning. Prof. Bart
Selman from Cornell University gave the second
plenary talk on developing automated reasoning
methods for complex scenarios. The day ended
with two additional sessions on cooperative
activity and machine learning. The evening social
activity allowed participants to enjoy an Israeli
dinner, with music and informal discussions in a
great atmosphere.
We were honored to start the third day with a
plenary talk on simple deterministic free will by
Prof. John McCarthy of Stanford University. The
rest of the day included a variety of interesting talks
on logic, active databases, constraint satisfaction
problems, swarm intelligence, machine learning
and robotics-laws.
Prof. Helmut Prendinger of the Japanese National
Institute of Informatics gave the second invited
talk, discussing design for affective interaction
with embodied interfaces. This also began the
second satellite
w o r k s h o p ,
whose theme
was affective
computing, and
this continued
through the
rest of the day
closing the whole
symposium. It
included exciting invited talks by Gideon Dror on
learning facial attractiveness, Andrew Ortony on
emotions and intelligent agents, Fabio Pianesi on
a database of kinetic facial expressions and Moshe
Koppel on profiling of anonymous authors.
Prof. Helmut Prendinger
discussed design for
affective interaction with
embodied interfaces
We would like to thank the Caesarea Rothschild
Institute and the Center for Scientific and
Technological Research (ITC-irst) for hosting the
event at the University of Haifa, Israel, AgentLink,
the organizers of the symposium Prof. Martin C.
Golumbic, Prof. Uri Schild and Dr. Claudia V.
The 8 th Biennial
Israeli Symposium on
Foundations of AI (BISFAI)
took place from the 27 th to the 30 th
of June 2005 in Haifa Israel. The
symposium included two satellite
workshops: one on recommender
systems and another on affective
computing.Thirty nine reviewed
papers were presented in two
parallel tracks. Each day of the
main symposium started with a
plenary invited talk. A second
plenary talk opened the afternoon
session.
BISFAI is the Israeli main event
on artificial intelligence, which
gathers all the researchers and
students involved in AI research.
This year we also stressed
the participation of industry
at the symposium. Although
the conference is a national
gathering, it attracted international
speakers from Belgium, France,
Italy, Ireland, the UK and the
United States. Abstracts and
the technical program (including
photos!) are available from the
symposium web-site at
www.cri.haifa.ac.il/events/2005/
bisfai05/bisfai05.php.
Goldman, the program committee members and
all the participants. The next BISFAI is planned
to take place in Haifa in 2007.
28
November 2005|AgentLink News 19

EVENT REPORT
E4MAS 2005: Environments for Multiagent Systems
Danny Weyns
Katholieke Universiteit Leuven
Belgium
danny.weyns@cs.kuleuven.be
Eric Platon
Sokendai, National Institute of Informatics
Japan
platon@nii.ac.jp
Fabien Michel
Universite de Reims
France
fmichel@leri.univ-reims.fr
The second edition of the International
Workshop on Environments for Multiagent
Systems (E4MAS) was held with the Fourth Joint
Conference on Autonomous Agents and Multi-
Agent Systems (AAMAS 2005) in Utrecht. The
main goal of the E4MAS workshop series is to
provide a discussion forum on the emerging
research topic of Environments for Multiagent
Systems. E4MAS 2005 attracted 25 researchers.
Like the first edition, the workshop was organized
in a highly interactive and discussion oriented
style. The main topics of discussion were: (1)
Role of the environment, scope and models; (2)
Environment engineering; and (3) Multiagent
system applications with an explicit notion of
environment. 7 full and 4 short papers, selected
from 16 submissions, provided input for the
discussions.
Organization
In the first part of the morning session, authors
of papers got a strict time window (5 minutes for
short papers and 10 for full papers) to present their
work. Instead of traditional paper presentations,
speakers were asked to answer a number of
questions on open issues from the perspective
of their work, aiming to boost the subsequent
discussion sessions. During the remaining part of
the morning session, three parallel groups started
the discussion on the main topics mentioned
above. After lunch, discussions were resumed.
During the second part of the afternoon session,
all participants joined a plenary session to discuss
the results of each group and to outline challenges
for future research. The workshop concluded with
an open discussion on ongoing and new efforts
to further develop research on Environments for
Multiagent Systems.
Outcome
Role of the Environment, Scope, and Models
The main focus of the discussion session was
on regulating properties of the environment
in multiagent systems. On the one hand, the
environment enforces rules in multiagent systems,
e.g. laws in a physical domain, or communication
properties in a network; on the other hand
the environment can be exploited to mediate
interactions among agents and interactions
with resources, typical examples are norms and
organizational rules in virtual organizations.
During the discussion, regulating properties of the
environment were related to a deployment model
of multiagent systems that was presented at the
workshop. This model distinguishes three layers
in a multiagent system (from top to bottom):
Multiagent System Application layer; Execution
Platform, and Physical Infrastructure. Different
kinds of regulation pertain to each layer:
• Specific multiagent system application rules:
e.g. norms, interaction laws, etc.
• Execution platform rules: e.g. coordination
rules of middleware, scheduling policies, etc.
• Physical infrastructure rules: communication
bandwidth, topological constraints, physical
laws, etc.
Regulation appears as a feature of the
environment that spans the three layers in which
the rules of lower layers typically constrain rules
at higher layers. This observation is important for
environment engineering.
Environment Engineering
A main point of discussion in this discussion
session was the dependency between environment
engineering and agent engineering. One
viewpoint states that environment engineering
should be independent from agent engineering.
The general idea of this viewpoint is that it should
be possible to develop a library of environmental
mechanisms that designers can reuse when
building multiagent system applications. The
opposite viewpoint states that the environment
cannot be engineered independently from agents.
At the very least the interface between agents
and the environment cannot be uncoupled.
Moreover, if the environment is actively involved
in the mediation of agent interaction, concerns
appear that crosscut agents and the environment,
indicating the need for an integrated engineering
process.
Despite opposing arguments, the two points
of view are compatible. Some concerns of the
environment can be engineered separately, e.g.
internal dynamics of a pheromone infrastructure,
while others concerns must be considered in
relation to agent engineering, e.g. roles and social
structures. Further research is necessary to clarify
these essential engineering issues.
Applications
A number of interesting multiagent system
applications with an explicit notion of
environment were presented at the workshop.
One application showed how web sites can be
considered as the concrete environment for
web users (i.e. the agents) and demonstrated
how this environment can be used to provide
useful advice to users in their current context.
Another industrial application showed how
a physical environment was enhanced with a
virtual environment layer that allows flexible and
distributed coordination among a set of automatic
guided vehicles. These and other applications
highlighted several opportunities arising from an
explicit environment:
• Providing abstraction and hiding low-level
complexity of a physical entity (e.g. fusion of
sensory information)
Research on Environments for Multiagent Systems considers the environment
as a first-order abstraction in multiagent systems. The environment is an explicit part
of a multiagent system that encapsulates essential parts of a multiagent system that
conceptually do not belong to agents, such as infrastructure for communication, the
topology of a spatial domain, or laws of an e-institution. Basically, the environment provides
the surrounding conditions for agents to exist, it offers an abstraction of the external world
to agents in which they can act and interact. This abstraction bridges the conceptual gap
between the agent abstraction and low-level issues, such as details of communication, or
resources access. Moreover, the environment provides an exploitable design abstraction
for building multi-agent systems. The environment can also serve as a medium for agents
to share information and coordinate their behaviour.
November 2005|AgentLink News 19
29

EVENT REPORT
• Resolving coordination conflicts (e.g.
access control to resources, support for
transactions)
• Providing extended functionality for
coordination (e.g. exploiting marks in the
environment)
• Simulating future-world state space (e.g.
what-if analysis)
• Maintaining past-world state space (e.g.
enable causal predictions)
Follow-up
Selected and revised papers of the workshop,
complemented with several invited papers will be
collected in a second volume on Environments
for Multiagent Systems that will be published
early 2006 as volume 3830 in the Lecture Notes
in Computer Science series of Springer-Verlag.
In addition, a Special Issue on Environments
for Multiagent Systems for the Journal of
Autonomous Agent and Multiagent Systems
is in preparation. The aim of this volume is to
communicate the consolidated knowledge in
research on environments to the multiagent
system research community in general.
Finally, the organizing committee plans a third
edition of the E4MAS workshop at AAMAS 2006
to be held in Hakodate, Japan. General goals for
E4MAS 2006 are:
• To make a step forward in environment
engineering
• To broaden the perspective on environment
by investigating new uses of environments in
multiagent systems.
Detailed information about the E4MAS
workshop series is available on the web: http://
www.cs.kuleuven.ac.be/~distrinet/events/e4mas/
Bibliography
[1] Environments for Mulitagent Systems: Stateof-the-Art
and Research Challenges, Danny
Weyns, Van Parunak, Fabien Michel, Tom
Holvoet, Jacques Ferber, First International
Workshop on Environments for Multiagent
Systems, New York, 2004. Available at:
http://www.cs.kuleuven.ac.be/~danny/e4mas_
survey_2004.pdf
[2] Environments for Multiagent Systems I,
Danny Weyns, Van Parunak, Fabien Michel
(Eds.), Lecture Notes in Computer Science,
Volume 3374, Springer Verlag, Berlin,
Heidelberg, New York, 2005.
[3] Environments for Multiagent Systems, Danny
Weyns, Michael Schumacher, Mirko Viroli,
Alessandro Ricci, Tom Holvoet, Knowledge
Engineering Review, Cambridge University
Press (to appear). Available at: http://www.
cs.kuleuven.ac.be/~danny/ENV-KER-2005.pdf
International Workshop on Agent-Based Models for Economic Policy
Design (ACEPOL05), June 30 – July 2, 2005, Bielefeld, Germany.
Herbert Dawid, University of Bielefeld, Germany, hdawid@wiwi.uni-bielefeld.de
The International Workshop on Agent-Based
Models for Economic Policy Design (ACEPOL05)
took place from June 30 – July 2, 2005 at the
Center for Interdisciplinary Research (ZiF) at
Bielefeld University. The scientific coordinator
of the meeting was Herbert Dawid (Bielefeld
University). The goal of the workshop was to
assess and further advance the usefulness of agentbased
computational models for the evaluation
and design of economic policy measures.
In particular questions concerning the design and
validation of agent-based models in economic
policy applications as well evaluation of robustness
of simulation results were discussed.
There was strong interest in the meeting from
researchers in several fields. The workshop
programme consisted of 5 plenary and 25 regular
talks, selected from 55 submissions. More than
50 participants from 11 countries attended the
workshop where the background of participants
included economics, management science,
computer science, social science, engineering
and demography. This multi-disciplinary setting
gave rise to a fruitful exchange of ideas between
disciplines and lively discussions.
The five plenary talks addressed different
important issues concerning the potential of
agent-based models for policy evaluation and
policy design. Han LaPoutre (CWI Amsterdam)
discussed the importance of a careful design of
agent-based models in economics. He argued
that modelling choices and parameter values
have to be sound not only from an economic
modelling point of view but also should make
sure that computational pitfalls like premature
convergence are avoided in the simulations. Luigi
Orsenigo (University of Brescia) demonstrated
how industry simulation models which are able
to capture and reproduce past qualitative changes
in the structure of a certain industry (historyfriendly
models) can be used to gain insights
into the medium- and long-term effects of public
policy like antitrust measures, procurement or
investment in basic research. Leigh Tesfatsion
(Iowa State University) dealt with the application
of agent-based methods for market design.
In addition to a discussion of methodological
issues in this area she presented recent work in
a project where agent-based simulation studies
are used to test the reliability of a wholesale
power market platform proposed by the US
Federal Energy Regulatory Commission. The
complementary relationship between agent-based
economics and the currently flourishing field of
experimental economics was discussed by John
Duffy (University of Pittsburgh). He pointed
out that results from human subject experiments
can be very helpful in providing foundations for
behavioural rules used in agent-based models and
In recent years economists, computer
scientists and social scientists have started to
use an agent-based approach as a normative
tool to construct and evaluate market
designs as well as public policy measures
in simulated economic environments. The
agenda of the workshop was to demonstrate
in which economic policy areas agent-based
methods are already succesfully applied
and to deal with important methodological
challenges for agent-based models in this
area of application.
that carrying out agent-based and human subject
experiments in parallel are a promising way to
evaluate the predictive power of agent-based
models in (simple) economic environments. In
the final plenary talk Nigel Gilbert (University
of Surrey) used examples of his own previous
work and work presented at this workshop to
derive a list of guidelines a researcher in agentbased
economics should keep in mind in order
to fruitfully address policy issues with the help of
agent-based models.
Many of the regular talks in the workshop gave
an overview over the various different fields of
economic policy where agent-based models are
currently applied. The issues addressed included
30
November 2005|AgentLink News 19

EVENT REPORT
the design of electricity, labour and lottery
markets, the evaluation of the effects of tax
policy, industrial policy, agicultural policy, the
control of the spread of crime and flood control.
Several talks addressed issues of empirical and
experimental validation of agent-based models in
economics highlighting the fact that these issues,
which have been rather neglected for some time,
have attracted a lot of attention in recent years.
Selected papers from this workshop will be
published in a special issue of the Journal of
Economic Behavior and Organization on Agent-
Based Models for Economic Policy Design.
Financial support from Agentlink III and the
Center for Interdisciplinary Research (ZiF) met
the costs of accomodation of all speakers at this
event. The full programme of the workshop
including most papers and presentation files
is available at: http://www.wiwi.uni-bielefeld.de/
~dawid/acepol/index.htm .
The 2005 IEEE/WIC/ACM International Joint Conference
on Web Intelligence and Intelligent Agent Technology
(WI’05 and IAT’05)
Compiegne University of Technology, Compiegne,
France 19-22 Sep 2005
www.hds.utc.fr/WI05/
Pierre Morizet-Mahoudeaux
Universite de Technologie
France
pmorizet@hds.utc.fr
Autonomous entities capable of independent
actions, self-adaptation, communication and
cooperation that characterise living sentient
organisms are one of fundamental inspirations
of the agent computing paradigm. Research
progress on intelligent agents provides for modern
multi-agent systems aiming at solving intricate
problems involving highly complex processes
in heterogeneous environments, described in
different languages, often under uncertainty
and incomplete knowledge. One of such
environments is the World Wide Web, which has
enjoyed an unprecedented expansion in recent
years. The Web is no longer considered merely as
yet another channel of gathering information, but
that it effectively embodies a whole new paradigm
of computation, communication, interaction
and cooperation. Its size, structure, complexity
and behaviour is gradually resembling those
from the real world, in which many elaborate,
independent, and yet interacting processes occur.
A better understanding, and consequently, proper
modelling of such processes is vital to the design,
development and government of the Web and
Web-based computing. A principal approach
to this task, widely known as Web Intelligence
(WI), is based on the theory of autonomous,
distributed, interacting and cooperating
computing agents, which has been a central
subject of the Intelligent Agent Technology
research community. On the other hand, Agent
Technology researchers see the fast-expanding
Web as an exciting environment and a fresh, new
testbed for their theories and methods, as well
as a crucial complement to other models such as
grid or ubiquitous computing.
The IEEE/WIC/ACM WI-IAT Joint Conferences
attempt to provide a platform of exchange
for researchers on many of these problems.
Following WI-IAT’04 held in Beijing, China,
this year WI-IAT’05 was hosted by Compiegne
University of Technology, Compiegne, France.
Out of 328 submissions, 153 papers by 436
authors from 34 countries have been accepted
for WI’05. IAT’05 hosted 134 papers by 322
authors from 25 countries selected from a total
of 305 submissions. During four days, from
19 to 22 September 2005, 6 keynote speeches,
5 workshops, 5 tutorials, Industry and Demo
tracks along with 37 regular sessions have been
held.
The conference’s first keynote speaker, Peter
Schuster (University of Vienna), gave an
insightful lecture on the evolution of autonomous
simple systems and their optimisation, along
with a vivid illustration of described principles
on the Darwinian evolution of DNA and RNA
structures. He stressed the comparison of the
agent computing paradigm to behaviours of
living organisms. The speech was followed by
another keynote lecture on agent agreement
technologies by Nick Jennings from University
of Southampton, who thoroughly outlined
a number of crucial issues pertaining to
communicating networked software agents that
seek to reach agreements by way of negotiations
in a context-dependent environment.
Another key lecturer, Pat Langley from Stanford
University, presented in details the project
ICARUS, an adaptive architecture for physical
agents. It concentrates on exploring intelligent
behaviours in physical domains and encompasses
many profound aspects of cognition, perception,
inference, and planning actions. Wray Butine
(Helsinki Institute for Information Technology),
in turn, introduced to the audience many exciting
facets of the world of web search engines, with a
particular description of those developed with
open sources.
The last day of the conference has been marked
with two keynote speeches. Henry Lieberman
(MIT Media Laboratory), in picturesque detail,
presented a novel approach to providing online
assistance, expertise transfer and understanding
users’ language. Patrick Doherty from Linkšping
University gave a vivid and amusing presentation
of his unmanned aerial vehicle project, which
attempts to perform advanced tasks such as
automatic monitoring and surveillance using
approximate reasoning with complex concepts.
A broad range of issues, problems, approaches
and methods were presented and widely discussed
at the conference’s sessions. Most notably, a
paper describing myopic approaches to agent’s
communication by Raphen Becker, Victor Lesser,
and Shlomo Zilberstein and another dealing with
web mining using rough sets theory by Chi Lang
Ngo and Hung Son Nguyen were awarded best
papers for IAT’05 and WI’05 tracks, respectively.
The Best WI/IAT’05 Demo Award went to David
Sislak, Martin Rehak, and Michal Pechoucek
for their multi-agent platform with advanced
simulation and visualisation support (for more
information on this work, see the article on page 9
in AgentLink News 17). Robert Nolker and Lina
Zhou were awarded the WI’05 Best Student Paper
Award for their article on a method of identifying
member roles in online communities.
Apart from its extensive scientific program, the
conference’s organizers came up with pleasant
social activities, including visits to the historic
city hall and the Pierrefonds medieval castle.
The next year IEEE/WIC/ACM WI-IAT
conference will be held in Hong Kong, together
with the IEEE International Conference Data
Mining (ICDM), on 18-22 December 2006. For
more information, please visit http://www.comp.
hkbu.edu.hk/~wii06/
November 2005|AgentLink News 19
31

EVENT REPORT
Third Workshop on Agents Applied in Health Care,
at the 19th International Joint Conference on Artificial Intelligence
(IJCAI 2005)
Antonio Moreno, University Rovira i Virgili, Spain, antonio.moreno@urv.net
The 3rd workshop on Agents Applied in Health
Care was held on July 31st at Edinburgh,
Scotland, as part of the 19th International Joint
Conference on Artificial Intelligence. It was
organised by Antonio Moreno (Univ. Rovira
i Virgili, Tarragona, Spain), Ulises Cortes
(Technical University of Catalonia, Barcelona,
Spain),John Nealon (Oxford Brookes University,
Oxford, UK) and John Fox (Cancer Research
UK, London, UK). Previous editions of the
workshop had been held at ECAI 2002 (Lyon,
France) and ECAI 2004 (Valencia, Spain). The
main topic of these meetings is the use of agent
technology and multi-agent systems in all aspects
of health care. The workshop organisers are also
the main promotors of the AgentLink TFG on
Agents Applied in Health Care.
This, the third workshop in the series, was
attended by 20 delegates, with eight talks
coming from countries as diverse as Italy, Czech
Republic, The Netherlands, Germany, United
Kingdom and Spain. The main topics covered by
the papers were: the use of agents in diagnostics
and monitoring tasks, the management of health
care of old or disabled citizens, the simulation
and optimisation of medical processes, and the
retrieval of medical information from Internet.
There were three papers in the diagnostics and
monitoring field. Aoda Valls (Univ. Rovira i
Virgili, Spain) presented the PalliaSys project,
in which a multi-agent system that monitors the
evolution of palliative patients in a hospital is
currently being developed.
Lenka Lhotska (Czech Technical Univ., Czech
Republic) explained the first steps in the design of
ADAM (Agent architecture for Diagnostics And
Monitoring in medicine). Beatriz Lopez (Univ.
of Girona, Spain) gave a report on the Masictus
multi-agent system, which gives support to the
diagnosis of acute stroke and the coordination of
ambulance services.
Roberta Annicchiarico (Fondazione Santa Lucia,
Italy) presented the latest developments in the
on-going e-tools project, in which an agentcontrolled
wheel chair is being developed to
improve the quality of life of disabled people.
Richard Hill (Sheffield Hallam Univ., UK) argued
the need of having formal tools to represent
appropriately health care workflows within multiagent
systems.
Loes Braun (Inst. for Knowledge and Agent
Technology, The Netherlands) described an
agent-based system that allows doctors to retrieve
from Internet the medical information that may
be relevant to treat their particular patients.
Reiner Herrler (Univ. of Würzburg, Germany)
described a hospital simulation kit based on
the SeSAM simulation tool. Finally, Pancho
Tolchinsky (Technical Univ. of Catalonia, Spain)
explained the latest work in the Carrel project,
in which argumentative agents are being used to
increase the number of available human organs
for transplant.s
The workshop also featured an invited talk by
the IJCAI conference chair, Fausto Giunchiglia
(Univ. of Trento. Italy). In his talk he described
the concept of context ontologies, and their
representation in the language C-OWL.
The organisers would like to thank the financial
contribution of AgentLink III, that kindly
provided some funding to partially cover the
attendance costs of 3 PhD students that were the
first authors of accepted papers.
An special issue on agents applied in health care
of the IEEE Intelligent Systems journal, edited
by Antonio Moreno, is planned for next year.
It will include revised and extended versions
of papers presented at the workshop, as well as
papers contributed by other researchers in the
field. Papers will have to be submitted for review
around. May 2006.
Agents in Space
Michael Fisher
University of Liverpool
United Kingdom
M.Fisher@csc.liv.ac.uk
July 2005 saw a one-day symposium on
“Agents in Space” held in Liverpool, UK, and
jointly organized by the Research Institute for
Advanced Computer Science (RIACS) at the US
National Aeronautics and Space Administration
(NASA), space systems company SciSys Ltd,
and the Departments of Computer Science and
Engineering at the University of Liverpool. The
symposium co-chairs were Professor Michael
Fisher (Head of the Liverpool Verification
Laboratory) and Dr. Roger Ward (Head of
On-board Software at SciSys). The aims of the
symposium were to highlight the use of agentbased
software systems in space exploration and
to explore the use of formal methods to verify
that software systems in aerospace applications
would work as intended.
With the development of deep space and planetary
exploration, communications between spacecraft
and mission control on earth are subject to
extended delays. These delays could have serious
impacts on operations, and so there is increasing
need for spacecraft to act autonomously, that is,
without constant reference to mission control.
Such autonomy is often naturally captured by
the notion of an agent. However, with the use
of autonomous agents in safety and missioncritical
situations come several challenges. This
symposium concerned three of these, namely:
how to control autonomous agents; how to
organise effective collaboration between humans
and agents; and how to verify the behaviour of
autonomous agents. The symposium brought
together some of the leading experts in agents,
autonomy in space, and formal verification, with
the aim of stimulating further developments in
this exciting area.
At the symposium, Dr Maarten Sierhuis from
NASA spoke on teamwork between robots and
humans on Mars or the Moon, while Dr Willem
Visser spoke about techniques for verifying
aerospace software. In addition, scientists from
SciSys, Cranfield University, and the Universities
of Durham, Southampton and York in the UK
gave talks on related topics during the day. This
novel and exciting area is set for significant increase
through the strategic directions being taken by
various space agencies: -- NASA; the European
Space Agency, ESA; and the Japanese Aerospace
Exploration Agency, JAXA. Verification of space
software, particularly in mission critical or safety
critical situations, is vital.
The symposium programme is available from:
http://www.csc.liv.ac.uk/~michael/agents-in-space05.
html
32
November 2005|AgentLink News 19

EVENT REPORT
First International Workshop on Security and Trust Management
(STM 2005)
15 September 2005
Milan, Italy
Sjouke Mauw
Eindhoven University of Technology
The Netherlands
sjouke@win.tue.nl
In September 2005 the STM working group
of ERCIM organized a successful workshop on
Security and Trust Management in Milan, Italy.
The initiative for this workshop was taken at the
founding meeting of the STM working group
in January 2005 in order to fulfill some of the
goals of the working group to bring researchers
together and stimulate scientific discussion.
The organization of the workshop, co-located
with ESORICS’05 (European Symposium On
Research In Computer Security), was in the
hands of Fabio Martinelli, Pierangela Samarati
(general co-chairs), Valerie Issarny, Sjouke Mauw
(PC co-chairs) and Cas Cremers (vice-chair).
Due to the high number of submissions (36)
the PC could select nine papers of high quality,
which were presented at the workshop and will
appear in a special issue of ENTCS (Electronic
Notes in Theoretical Computer Science. Is there
a also series or theme). Sponsorship was received
from ERCIM (write out: the European Research
Consortium for Informatics and Mathematics)
and AgentLink.
The workshop had a broad scope, ranging from
cryptography and formal methods to physical
security. The link between the major topics,
security and trust, was made by keynote speaker
Prof. Dieter Gollmann (Technische Universität
Hamburg, Harburg), who challenged the
audience with his presentation entitled “Why
Trust is Bad for Security”. He argued that
the notion of trust has many different (often
conflicting) interpretations, while there is a need
for clarity and precision.
The first session of the workshop covered smart
dust security, a formal approach to multiparty
contract signing and the development of security
models for mobile agent security. In the second
session, the notions of credit and responsibility
were formalized, a new scheme for trapdoor
hash functions was presented, and access control
mechanisms based on trust were studied.
The third and final session was completely
dedicated to trust management. It covered
the extension of role based trust m anagement
languages with non-monotonicity, assigning
trust values to metadata, and a new authorization
strategy for distributed environments.
Given the success of this first STM workshop,
and the interest that it raised, we believe that
this will be the first of a series of successful
workshops.
The 8th edition of the
European Agent Systems Summer School
(EASSS’06)
Université de Savoie, Annecy, France, 17-21 July 2006
http://www.esia.univ-savoie.fr/easss06
Researchers and teachers in this field are invited to submit a proposal for a course. A typical course is
4 hours in total, but deviations (e.g. 2 or 6 hour) are possible. Programs of previous editions of EASSS
can be found at the web page.
Tutors are encouraged to submit a propsosal for a course to Mehdi Dastani (mehdi@cs.uu.nl), including
a 1 page description mentioning experience of the tutor, level of the course (beginners, advanced),
duration and needs for equipment.
Deadline for Proposals: December 17, 2005
Acceptance Notification: January 23, 2006
Submission of final Teaching Material (< 26 pp): April 23, 2006
November 2005|AgentLink News 19
33

Semantics
Semantics
PROJECT REPORT
ATHENA:
Advanced Technologies for interoperability of
Heterogeneous Enterprise Networks and their Applications
Klaus Fischer
Deduction and Multiagent
Systems, DFKI GmbH
Germany
Klaus.Fischer@dfki.de
ATHENA is an Integrated Project sponsored by the
European Commission in support of the Strategic
Objective “Networked businesses and government”
as set out in the IST 2003-2004 Work programme
of FP6 (Framework Programme). Building upon
an ambitious Vision Statement that “By 2010,
enterprises will be able to seamlessly interoperate
with others”, ATHENA aims to make a major
contribution to interoperability by identifying
and meeting a set of inter-related business,
scientific & technical, and strategic objectives. The
ATHENA programme of work is geared towards
producing results that span the full spectrum of
interoperability from technology components
to applications and services, from research &
development to demonstration & testing and from
training to evaluation of technologies for social
impact. In ATHENA, different Research and
Development projects (inside Action Line A) are
executed in close synergy and collaboration with
Community Building activities (inside Action
Line B), thus ensuring that solutions for multidisciplinary
research challenges are of optimal
industrial relevance leading to a broad uptake
by the end user. Figure 1 depicts the structure of
ATHENA and the relation between the projects
running in the ATHENA program.
ATHENA is committed to creating a long-term
impact for advancing interoperability which is
mainstream, inclusive and has critical mass. To this
end, ATHENA is initiating an open, neutral and
independent Enterprise Interoperability Centre
(EIC) in which all stakeholders, both in private
and public sectors, are invited to participate.
Fig 1: ATHENA’s Strcuture
The establishment of the EIC
is an action of the European
Union’s updated eEurope 2005
Action Plan for implementing
the Lisbon Strategy. The EIC
Executive Summary is available
as part of the public consultation
with interoperability stakeholders
in respect of the establishment of
the EIC.
The ATHENA consortium
currently comprises of 19 leading
organisations in research, academia, industry and
other stakeholder communities including SMEs,
working collaboratively in pursuit of a common
set of objectives in interoperability. The ATHENA
programme of work represents one of the largest
investments made in interoperability through
the co-financing arrangements of the European
Commission’s research programme (the total
committed project budget is 26.6 million Euro,
including 14.4 million Euro financing by the EC).
ATHENA started on the 1st of February 2004 and
is currently contracted for 3 years, with a 5 year
time horizon and planning.
ATHENA’s Research and Development
projects
Research and Development projects, under Action
Line A, will provide research results that fit into
an Interoperability Framework. The General
ATHENA Framework is a conceptual model
which looks at different aspects of interoperability,
combining and leveraging business knowledge
Enterprise A
Business
Knowledge
ICT Systems
Enterprise B
Business
Knowledge
ICT Systems
Fig 2: General Athena Framework
and modeling, Information and Communication
Technologies (ICT) platforms and Semantics
research.
This general framework is represented via three
main layers: business, knowledge and ICT
(Information and Communication Technology)
solutions, as illustrated in Figure 2.
In order for organizations to successfully
conduct business, both internally and with
other organisations, they must be aware of the
interdependency among the three layers as
mentioned in Figure 2. In addition, Athena has
added a fourth dimension regarding the semantics
that cut across the three layers. It was introduced
to promote an understanding of expression of
similarities and differences among businesses, the
way they operate, and their software solutions.
Research and Development has the primary
objectives of:
• Studying, analysing, enabling and managing
the business processes and the operations
of individual networked enterprises and
between collaborating enterprises;
• Studying, analysing, and managing
organizational roles, skills, competencies and
knowledge assets of individual networked
enterprises and between collaborating
enterprises;
• Designing and developing an Interoperability
Framework providing integration on a
conceptual, application and technical level;
• Studying, analysing and improving techniques
to increase interoperability between
applications, data and communication
components;
• Studying, analysing, and developing concepts,
tools, and methods to leverage and share
enterprise knowledge.
34
November 2005|AgentLink News 19

PROJECT REPORT
Figure 3 presents ATHENA’s operational vision
to interoperability. In this context three activities
of ATHENA are most interesting for agent
technologies:
1) Planned and Customisable Service-
Oriented Architectures (SOA): SOA are
being viewed as the next wave of technology
to impact the computing landscape. They do
this by enabling the utilisation of distributed
components by allowing software vendors
to provide not only applications to market,
but a suite of services. These services can be
utilised by a wider audience and charged
for through an access or usage business
model. This project is concerned with the
development of service-oriented solutions
that can be more easily planned and then later
customised when they are being deployed.
This is intended to provide better industry
focused solutions that can be adapted for
deployment into client environments. This
is to directly counter the problem of high
costs of customisation and integration of
highly generic industry solutions. The main
objectives of this project are to develop
modelling and specification systems that
accurately express services and serviceoriented
architectures and particularly
assist in the planning of solutions and the
marking of intended customisations for
the better deployment of a solution into
a wide range of client environments; to
develop technologies that enable the easier
composition of services and the brokering,
mediation and ultimately the negotiation
of pre-specified but customisable services;
to develop an execution framework for
Company
Company B
planned and customisable service-oriented
architectures utilising user scenarios from
the ATHENA community building
activities; and to evaluate the developed
technologies and execution framework in
proto-typical implementations.
2) Model-Driven and Adaptive Interoperability
Architectures: the main objective activity is
to provide new and innovative solutions for the
problem of sustaining interoperability through
change and evolution, by providing dynamic
and adaptive interoperability architecture
approaches. The objective is to advance state of
the art in this field, by applying the principles of
model-driven, platform independent architecture
specifications, and dynamic and autonomous
federated architecture approaches. This includes
the following sub-goals: to support requirements
and validate solutions for the involved sectors
specified in the ATHENA community
building activities; to provide meta models and
methodologies for interoperability architecture
solutions; to evaluate and extend multiple
adaptive autonomous and federated architecture
approaches, including agent and peer-to-peer
technologies and the model-driven architecture
approach; to provide support for non-functional
interoperability aspects, through a modeldriven
approach: to apply the use of ontologies
and semantics in model and service registry/
repositories for better semantic interoperability;
and to provide semantic mapping and mediation
technologies that provide executable frameworks
and support for active models.
3) Knowledge Support and Semantic
Mediation Solutions: This project activity
examines the development of methods and
Service Wrappers / Interoperability Management
Evaluation & Negotiation of Available
Functionality
Execution Environment
Service Interconnection Bus
Fig 3: ATHENA’s operational vision to interoperability
Repository
Existing Enterprise Applications
tools for enterprise ontology management,
with a focus on supporting enterprise
knowledge integration and interoperability
for enterprises and software applications.
The main objective is to build an ontology
environment aimed at supporting the
integration of the different sorts of knowledge
that can be found in an enterprise. Enterprise
knowledge can be divided in two main
categories. The first category comprises
the knowledge represented by all sorts of
documents: from technical reports to emails,
from scientific papers to circuit blueprints;
this category is mainly conceived to be
used by human users. A second category is
represented by formal symbolic knowledge,
e.g., modelled by using semantic nets or
description logics, stored in an ontology,
which is mainly organized to be exploited
by a computer. The proposed ontology
enterprise environment, based on ontology
knowledge, will be used in two main
enterprise areas: to support the integration
of documental knowledge and to achieve
enterprise software interoperability. Both of
them can be achieved by applying semantic
annotation techniques, although our focus
will be on the latter. On the technological
level, semantic annotation will be used to
associate meaning to computable services
(e-services, agents, software components,
etc.) selected to be used in advanced software
architectures, such as management domain
architecture, SOA, distributed application
integration. Semantic annotation is one of
the key enabling technologies to implement
a number of solutions for achieving
interoperability, e.g., from service discovery
to matchmaking, to the linking of Business
Processes to computable services.
In the first two activities, agent technologies
are directly investigated in how far they can
contribute to the problems the projects deal with.
Most important are the questions, how agent
technologies can be smoothly integrated with
SOA and how the MDA methodology fits with
an agent-oriented approach to system design.
Because tools for the design of BDI agents are
available that allow to describe the agents and
even teams of agents using UML style models,
BDI agents seem to be the most promising way
to integrate agent-oriented software design with
the model-driven development methodology that
MDA suggests. Still it remains an open question
in how far the meta-models ATHENA comes up
with relate to a meta-model of BDI agents and
in how far the idea of model transformations can
be used to improve the way in which multiagent
systems are designed.
More details on ATHENA can be found at
http://www.athena-ip.org and by sending e-mail
to info@athena-ip.org or, if agent technologies are
concerned, to Klaus Fischer (Klaus.Fischer@dfki.de).
November 2005|AgentLink News 19
35

SITE REPORT
Business Decision Making
using Multi-Agent Systems
Nadia Yakounina
Magenta Technology
United Kingdom
yakounina@magenta-technology.com
The complexity of modern business operations
has now increased to the point that without
moving to new dynamic, adaptive systems,
business people face a difficult challenge in
terms of enterprise decision making. Either they
can rely on software systems that ignore the
true complexity of their operations in order to
automate high volume business processes or they
can try to address complexity by employing more
and more smart, experienced people. Neither
alternative is scalable or robust, and certainly
will not provide a competitive advantage. The
smartest software people can reach out for now
are “optimizers” that use elaborate “business
rules”, neither of which can react or respond
effectively under volatile, real-time pressure.
Therefore, in many instances, people still need
to get involved to re-work the solution and make
the trade-offs necessary for a practical solution.
To enable a major step-change in enterprises,
Magenta’s approach is focused on combining
several key elements: 1) Continuous Realtime
Planning 2) Adaptation 3) Team-based
Decision-making.
1) Continuous Real-time Planning
The main objective in moving from a cascade/
batch planning environment is to compress
the time taken to complete the Plan-Commit-
Execute cycle. The longer the delay between
planning and starting a task, the higher the
risk that the plan will become stale and nolonger
effective, and hence require some form
of exception handling. Consequently Magenta
has focused on Event-Driven systems that
incrementally react to commit changes to a
complex Schedule. The Magenta approach can
understand Scheduling Events from a flow of
messages.
2) Adaptation
When running practical business systems, the
core knowledge of the business process needs
first to be tuned to fit actual use, and then it
needs to be adapted to respond to the evolving
business needs. By using semantic network
technology, Magenta can describe, capture and
compare information and knowledge on-the-fly
to grow and alter the software over time.
3) Team-base Decision-Making
As business people who are geographically
dispersed typically work asynchronously when
working across a single process, enabling them to
share resources is not straightforward. The endto-end
business process must first be described
in detail, and stored in an industry accepted
standard notation, such as the Business Process
Modelling Notation (BPMN). Each process
typically has a number of layers of decisionmaking
authority and consultation, and should
About Magenta
Magenta Technology is the leading
developer of enterprise-ready multiagent
and semantic web software,
to enable the adaptive enterprise,
utilising a unique combination
of core technologies - Multi-
Agent systems, Semantic Web
technologies and Java/ J2EE. The
UK-based company offers a range
of enterprise-ready products that
release significant value trapped
within supply chain operations;
maximising revenue, lowering
costs and bringing added value to
business activities.
be mapped to the software workflow. Users also
require access to strong visualization tools, alerts,
and access to reports to monitor and understand
how well a process is performing.
Inside the Event-Driven Architecture
Within the Magenta solution, the agents reach
decisions by negotiating with each other, to
achieve the best individual outcome whist
adhering to enterprise goals to maximise value
to the whole system. For example, given the task
of scheduling the delivery of cargo to shipping
vessels, agents representing each cargo should
only seek to minimise their individual cost of
delivery if this does not result in an increase in
cost of delivering the full set of cargo. Figure
2 illustrates the high-level architecture for a
Magenta Technology system to schedule loads
to vehicles for logistics transport planning.
Figure 1: Decision Support Technology Sweet spots. Tools and approaches in software market today force business
people to either simplify the description of their business or increase manual activity to address complexity. Magenta’s
approach resolves this dilemma.
The system utilises XML datasets that describe:
events (orders, cancellations, breakdown of
vehicles etc); the assets used to fulfil orders
(trucks, vans etc); the geography of the transport
network; and perhaps an initial schedule.
This information is used to construct a scene
that represents the real world problem to be
solved, and which is an instance of an ontology
that describes the Business Process situation.
The ontology describes the business process,
including physical attributes of vehicles,
industry regulations, and preferences used in
36
November 2005|AgentLink News 19

decision making such as preferred routes, costs
of vehicles etc. The scene-forming component
has both an API and a toolset for managing the
ontology (OMT). The scenes are stored in a
database which provide persistence and a history
of the whole decision making process.
Armed with the business process ontology and
a populated scene of that ontology, a swarm of
agents operating in a virtual market negotiate the
allocation of orders to vehicles. The virtual market
is powered by a multi-agent engine that provides
messaging protocols for the agents. In the agent
swarm, every order and vehicle is represented by an
agent that seeks an optimal plan, thus identifying
the best result for that order and vehicle. To
achieve this, the agents have access to a toolbox of
capabilities, such as distance calculators and key
performance indicator (KPI) calculators. A user
interface (UI) component provides the human
planner with access to the system and reporting
capabilities. The whole system is built on a J2EE
architecture, which gives platform operating
system and database independence, robustness,
reliability, integration with existing systems, and
scalability out of the box.
Figure 2 Magenta Multi Agent System Architecture for Transport Logistics Planning
Results from this system have been compared
with results from traditional optimisation tools.
The quality of the schedule produced by the
multi-agent system is comparable to these tools,
but exhibits the following benefits:
• New orders or cancellations can be scheduled
in near real-time without having to re-plan
the whole schedule. Real time, incremental
planning enables a business to plan new
orders just before loading the vehicle;
• Changes to the business process can be
quickly implemented in the ontology, which
is independent of the agent code;
• The ontology can be changed by a domain
expert, rather than a computer systems
expert;
• The system can be readily used to do scenario
planning because of its incremental planning
capability;
• The system does not impose hard rules on
the plan – all preferences can be stressed, just
like real life;
• The system can cope with very high levels of
complexity and contradictory preferences.
This makes multi-agent systems ideal for business
decision-making in fast moving, complex,
evolving, and uncertain environments.
EUMAS
Brussels, Belgium
7-8 December, 2005
In the last 15 years we have seen a significant increase of interest in agent-oriented technology. This
field is now set to become one of the key technologies in the 21st century and will underpin much of the
next generation of computing that seeks to address issues in Ambient Intelligence, Pervasive Computing,
Grid Computing, the Semantic Web, E-Commerce and many other areas. It is therefore crucial that
both academics and industrialists within Europe have access to a forum at which current research and
application issues are presented and discussed. The aim of this third European Workshop on Multi-Agent
Systems is to encourage and support activity in the research and development of multi-agent systems, in
academic and industrial European efforts.
For Details see http://como.vub.ac.be/eumas2005/
November 2005|AgentLink News 19
37

BOOKS
Agent Intelligence through
Data Mining
Andreas L. Symeonidis
Pericles A. Mitkas
ISBN: 0-387-24352-6
2005
http://www.springer.com/sgw/cda/frontpage/
0,,5-147-22-46687939-0,00.html
“Agent Intelligence
through Data
Mining” offers a selfcontained
overview
of a relatively young
but important
area of research:
the intersection of
agent technology
and data mining.
This intersection
is leading to
considerable advancements in the area of
information technologies, drawing the
increasing attention of both research and
industrial communities. It can take two
forms: a) the more mundane use of intelligent
agents for improved data mining and b) the
use of data mining for smarter, more efficient
agents. The book focuses on the second
approach.
Knowledge, hidden in voluminous data
repositories routinely created and maintained
by today’s applications, can be extracted by
data mining. The next step is to transform
this discovered knowledge into the inference
mechanisms or simply the behavior of agents
and multi-agent systems. Agent Intelligence
through Data Mining addresses this issue, as
well as the arguable challenge of generating
intelligence from data while transferring it
to a separate, possibly autonomous, software
entity. This book contains a methodology,
tools and techniques, and several examples
of agent-based applications developed with
this approach. This volume focuses mainly
on the use of data mining for smarter, more
efficient agents.
Agent Intelligence through Data Mining
is designed for a professional audience of
researchers and practitioners in industry.
This book is also suitable for graduate-level
students in computer science.
Multi-Agent Programming
Edited by R.H Bordini, M. Dastani,
J. Dix and A. El Fallah Seghrouchni
ISBN: 0-387-24568-5
http://www.springer.com/sgw/cda/
frontpage/0,11855,3-0-22-53996849-0,00.
htmlreferer=www.springer.com%2F0-387-
24568-5
M u l t i - A g e n t
Programming is an
essential reference
for anyone interested
in the most up-todate
developments in
MAS programming.
P r o g r a m m e r s ,
researchers, and
graduate students will
find this text unique
in its presentation of
the concepts and principles of this fast-growing
field. While previous research has focused
on the development of formal and informal
approaches to analyse and specify Multi-Agent
Systems, this book focuses on the development
of programming languages and tools which
not only support MAS programming, but also
implement key concepts of MAS in unified
frameworks.
Part I describes four approaches that are based on
computational logic or process algebra--Jason,
3APL, IMPACT, and CLAIM/SyMPA. These
programming languages have formal semantics
and use heavy machinery based on formal
methods, but also provide working platforms
for the development of multi-agent systems.
Part II presents agent languages and platforms
that extend or are based on Java--JADE, Jadex,
and JACK TM . Although these have no formal
semantics, the languages are well documented
and the platforms provide a variety of tools
that have been extensively used in practice. Part
III provides two significant industry specific
applications--The DEFACTO System for
coordinating human-agent teams for the future
of disaster response, and the ARTIMIS rational
dialogue agent technology. The book also features
seven appendices, summarising each of the agent
programming languages, hence facilitating
comparison of the approaches. In particular,
Appendix A describes the criteria used for
comparing the agent languages and platforms.
Combinatorial Auctions
Edited by Peter Cramton, Yoav
Shoham, and Richard Steinberg
ISBN 0-262-03342-9
January 2006
http://mitpress.mit.edu/promotions/books/
FL20050262033429
The study of
c o m b i n a t o r i a l
a u c t i o n s - -
auctions in which
bidders can bid
on combinations
of items or
packages--draws
on the disciplines
of economics,
o p e r a t i o n s
research, and computer science. This
landmark collection integrates these three
perspectives, offering a state-of-the art
survey of developments in combinatorial
auction theory and practice by leaders in the
field.
Combinatorial auctions (CAs), by allowing
bidders to express their preferences more
fully, can lead to improved economic
efficiency and greater auction revenues.
However, challenges arise in both design
and implementation. Combinatorial
Auctions addresses each of these challenges.
After describing and analyzing various CA
mechanisms, the book addresses bidding
languages and questions of efficiency. Possible
strategies for solving the computationally
intractable problem of how to compute the
objective-maximizing allocation (known
as the winner determination problem) are
considered, as are questions of how to test
alternative algorithms. The book discusses
five important applications of CAs: spectrum
auctions, airport takeoff and landing slots,
procurement of freight transportation
services, the London bus routes market,
and industrial procurement. This unique
collection makes recent work in CAs
available to a broad audience of researchers
and practitioners. The integration of work
from the three disciplines underlying CAs,
using a common language throughout,
serves to advance the field in theory and
practice.
38
November 2005|AgentLink News 19

AGENT EVENTS
CALENDAR
For further details visit:
http://www.agentlink.org
If you would like to announce an event
email: publications@agentlink.org
with the details
2005
ASW 2005 Nov 3-6 1st International Symposium on Agents and the Semantic Web Arlington, Virginia, USA
ExaCt 2005 Nov 3-6 International Symposium on Explanation-aware Computing Washington D.C, USA
MIPSA 2005 Nov 3-6 AAAI 2005 Fall Symposium on Mixed-Initiative Problem Solving Assistants Arlington, Virginia, USA
ROLES 2005 Nov 3-6 2005 AAAI Fall Symposium: Roles, an interdisciplinary perspective Arlington, Virginia, USA
ISWDS 2005 Nov 6-10 The 1st International Semantic Web Doctoral Symposium Galway, Ireland
MICAI 2005 Nov 14-18 The 4th Mexican International Conference on Artificial Intelligence Monterrey, Mexico
MA4CS 2005 Nov 14-18 Multi-Agents for modeling Complex Systems Paris, France
WISE 2005 Nov 20-22 The 6th International Conference on Web Information Systems Engineering New York, USA
TAAI 2005 Dec 2-3 10th Conference on Artificial Intelligence and Applications Kaohsiung, Taiwan
MASTA 2005 Dec 5-8 The 3rd Workshop on Multi-Agent Systems: Theory and Applications Covilha, Portugal
EUMAS 2005 Dec 7-8 The 3rd European Workshop on Multi-Agent Systems Brussels, Belgium
CISSE 2005 Dec 10-20 International Joint Conferences on Computer, Information, and Systems Sciences, and Engineering Online Virtual Event
SOAS 2005 Dec 11-13 The 2005 International Conference on Self-Organization and Adaptation of Multi-agent and Grid Systems Glasgow, Scotland
CIS 2005 Dec 15-19 The International Conference on Computational Intelligence and Security Xi’an, China
IASAA 2005 Dec 16-18 WSEAS Special Session: Intelligent Agents: Standards, Architectures and Applications Tenerife, Canary Islands
IICAI 2005 Dec 20-22 The 2nd Indian International Conference on Artificial Intelligence (IICAI-05) Pune, India
MAS 2005 Dec 20-22 IICIA-05 Special Session on Multi-agent Systems Pune, India
ICDCIT 2005 Dec 22-24 2nd International Conference on Distributed Computing & Internet Technology (ICDCIT 2005) Bhubaneswar, India
2006
ADS 2006 Apr 2-6 Workshop on Agent-Directed Simulation Huntsville, USA
MDAI 2006 Apr 3-5 Modeling Decisions for Artificial Intelligence Vienna, Austria
ACE 2006 Apr 18-20 Agent Construction and Emotions:Modeling the Cognitive Antecedents and Consequences of Emotion Vienna, Austria
EMCSR 2006 Apr 18-21 The biennial European Meeting on Cybernetics and Systems Research Vienna, Austria
AAMAS 2006 May 8-12 The 5th International Joint Conference on Autonomous and Mutli-Agent Systems Hakodate, Japan
FLAIRS 2006 May 13-11 The 19th International FLAIRS Conference FL, USA
AGC 2006 May 16-19 International Workshop on Agent based Grid Computing Singapore, Singapore
CSWWS 2005 June 6 Canadian Semantic Web Working Symposium Quebec, Canada
AMT 2006 June 7-9 The 4th International Conference on Active Media Technology Brisbane, Australia
ESWC 2006 June 11-14 3rd European Semantic Web Conference Budva, Montenegro
ICAC 2006 June 12-16 The 3rd IEEE International Conference on Autonomic Computing Dublin, Ireland
Coordination 2006 June 13-16 8th International Conference on Coordination Models and Languages Bologna, Italy
DEON 2006 Jul 12-14 The 8th International Workshop on Deontic Logic in Computer Science Utrecht, The Netherlands
AAAI 2006 Jul 16-20 21st National Conference on Artificial Intelligence (AAAI-06) Boston, MA, USA
ISWC 2006 Dec 5-9 5th International Semantic Web Conference Athens, GA, USA
November 2005|AgentLink News 19
39

www.agentlink.org
About AgentLink...
AgentLink is the European Coordination
Action for Agent-based Computing,
a network of researchers and developers with a
common interest in agent technology. It is funded
by the European Commission, so membership is
free, but you can only take advantage of AgentLink
activities if you are a member. If your organisation is
engaged in agent-related activities, you should join.
Contact the coordinator or administrator for details,
or visit www.agentlink.org.
The aim of AgentLink News is to provide an
informal way of communicating both what is
happening in AgentLink and the agent world generally.
AgentLink News offers a range of articles including
features, reports on conferences and workshops,
informal description of research results and new
software, book reviews, and website developments.
ISSN 1465-3842
AGENTLINK PUBLICATIONS AND
WEB COORDINATOR
Serena Raffin
School of Electronics and Computer Science
University of Southampton
Southampton SO17 1BJ
United Kingdom
[e] web@agentlink.org
AGENTLINK ADMINISTRATORS
Adele Maggs
Department of Computer Science
University of Liverpool
Liverpool L69 3BX
United Kingdom
[e] admin@agentlink.org
Rebecca Earl
School of Electronics and Computer Science
University of Southampton
Southampton SO17 1BJ
United Kingdom
[e] publications@agentlink.org
AGENTLINK EVENT COORDINATOR
Catherine Atherton
Department of Computer Science
University of Liverpool
Liverpool L69 3BX
United Kingdom
[e] events@agentlink.org
AGENTLINK TECHNOLOGY
SUPPORT COORDINATOR
Steve Munroe
School of Electronics and Computer Science
University of Southampton
Southampton SO17 1BJ
United Kingdom
[e] web@agentlink.org
AGENTLINK COORDINATORS
Peter McBurney
Department of Computer Science
University of Liverpool
Liverpool L69 3BX
United Kingdom
[e] Peter.McBurney@agentlink.org
Terry Payne
School of Electronics and Computer Science
University of Southampton
Highfield
Southampton SO17 1BJ
United Kingdom
[e] Terry.Payne@agentlink.org
AGENTLINK EXECUTIVE COORDINATORS
Michael Luck, University of Southampton, UK
Michael Wooldridge, University of Liverpool, UK
INDUSTRY ACTION
(Workpackage 1) COORDINATOR
Michal Pechoucek
Czech Technical University
Czech Republic
[e] Michal.Pechoucek@agentlink.org
STANDARDISATION ACTIVITY
(Workpackage 2) COORDINATOR
Monique Calisti
Whitestein Technologies AG
Switzerland
[e] Monique.Calisti@agentlink.org
RESEARCH ACTION
(Workpackage 3) COORDINATOR
Steven Willmott
LSI,Universitat Politècnica de Catalunya
Spain
[e] Steven.Willmott@agentlink.org
STUDENT INTEGRATION PROGRAMME
(Workpackage 4) COORDINATOR
Wiebe van der Hoek
University of Liverpool
United Kingdom
[e] Wiebe.van.der.Hoek@agentlink.org
TECHNICAL FORUM
(Workpackage 5) COORDINATOR
Andrea Omicini
DEIS, Università di Bologna
Italy
[e] Andrea.Omicini@agentlink.org
TECHNOLOGICAL ROADMAP
(Workpackage 6) COORDINATOR
Michael Luck
University of Southampton
United Kingdom
[e] Michael.Luck@agentlink.org
ADDITIONAL COMMITTEE MEMBERS
Onn Shehory
IBM Research
Israel
[e] Onn.Shehory@agentlink.org
Simon Thompson
BT Exact
United Kingdom
[e] Simon.Thompson@agentlink.org
Jörg Müller
Siemens AG
Munich
[e] Joerg.Mueller@agentlink.org