Intelligent Agent Technology Research and Development pot

1
Proceedings of the
2nd Asia-Pacific
Conference on IAT
DzvdDpmzni
Editors
Ning Zhong
Jiming Liu
Setsuo Ohsuga
Jeffrey Bradshaw
World Scientific
Proceedings; trf the
2nd Asia-Pacific
Intelligent
Agent
2nd Asia-Pacific
rri 1 1
Conference
on
W A
eCnnOIOgy
Research
and
Development
Proceedings
erf the
2nd
Asia-Pacific
Conference
on IAT

Intelligent
Agent
Technology
Research
and
Development
Editors
Ning Zhong
Maebashi Institute
of
Technology,
Japan
Jiming
Liu
Hong Kong Baptist University
Setsuo Ohsuga
Waseda
University,
Japan
Jeffrey Bradshaw
University
of
West
Florida,
USA
^Sj
World Scientific
wB New
Jersey
*

London • Singapore
•
Hong Kong
Published by
World Scientific Publishing Co. Pte. Ltd.
P O Box 128, Farrer Road, Singapore 912805
USA office: Suite IB, 1060 Main Street, River Edge, NJ 07661
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library.
INTELLIGENT AGENT TECHNOLOGY
Research and Development
Copyright © 2001 by World Scientific Publishing Co. Pte. Ltd.
All rights
reserved.
This
book,
or parts
thereof,
may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be
invented,
without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.
ISBN 981-02-4706-0
Printed in Singapore by World Scientific Printers (S) Pte Ltd
PREFACE

Intelligent Agent Technology is concerned with the development of autonomous
computational or physical entities capable of perceiving, reasoning, adapting,
learning, cooperating, and delegating in a dynamic environment. It is one of the
most promising areas of research and development in information technology,
computer science, and engineering today.
This book is an attempt to capture the essence of the current state of the art in
intelligent agent technology and to identify the new challenges and opportunities
that it is or will be facing. It contains the papers accepted for presentation at The
Second Asia-Pacific Conference on Intelligent Agent Technology (IAT '01), held in
Maebashi, Japan, October 23-26, 2001. The second meeting in the IAT conference
series follows the success of IAT '99 held in Hong Kong in 1999. IAT '01 brought
together researchers and practitioners to share their original research results and
practical development experiences in intelligent agent technology. The most
important feature of this conference was that it emphasized a multi-facet, holistic
view of this emerging technology, from its computational foundations, in terms of
models, methodologies, and tools for developing a variety of embodiments of agent-
based systems, to its practical impact on tackling real-world problems.
Much work has gone into the preparation of the IAT '01 technical program:
Original, high-quality papers were solicited for various aspects of theories,
applications, and case studies related to agent technologies. 134 full papers were
submitted from 32 countries and regions of all continents. Each submitted paper
was reviewed by at least three experts on the basis of technical soundness, relevance,
originality, significance, and clarity. Based on the review reports, 25 regular papers
(19%) and 40 short papers were accepted for presentation and publication.
This book is structured into six chapters according to the main conference sessions:
Chapter 1. Formal Agent Theories
Chapter 2. Computational Architecture and Infrastructure
Chapter 3. Learning and Adaptation
Chapter 4. Knowledge Discovery and Data Mining Agents
Chapter 5. Distributed Intelligence

Chapter 6. Agent-Based Applications
In addition to the above chapters, this book also includes the abstract or papers for
the IAT '01 keynote/invited talks by Benjamin W. Wah, Toyoaki Nishida, Zbigniew
W. Ras, Andrzej Skowron, and Katia Sycara, which provide different perspectives
to Intelligent Agent Technology.
v
vi
We wish to express our gratitude to all members of the Conference Committee and
the International Advisory Board for their instrumental and unfailing support.
IAT '01 has a very exciting program with a number of features, ranging from
technical sessions, invited talks, agent demos, and social programs. All of this work
would not have been possible without the generous dedication of the Program
Committee members and the external reviewers in reviewing the papers submitted
to IAT '01, of our invited speakers, Benjamin W. Wah, Toyoaki Nishida, Zbigniew
W. Ras, Andrzej Skowron, and Katia Sycara, in preparing and presenting their very
stimulating talks, and of Jianchang Mao (Demos & Exhibits Chair) in soliciting
demo proposals and setting up the program. We thank them for their strong support.
The conference Web support team at the Knowledge Information Systems
Laboratory, Maebashi Institute of Technology did a terrific job of putting together
and maintaining the home page for the conference as well as building a software,
namely, cyber-chair, which is an intelligent agent and interface among organizers,
program committee members, and authors/attendees. We would like to thank Juzhen
Dong, Muneaki Ohsima, Norichika Hayazaki of the conference Web support team
for their dedication and hard work.
IAT '01 could not have taken place without the great team effort of the Local
Organizing Committee and the support of Maebashi Institute of Technology and
Maebashi Convention Bureau. Our special thanks go to Nobuo Otani (Local
Organizing Chair), Sean M. Reedy, Masaaki Sakurai, Kanehisa Sekine, and
Yoshitsugu Kakemoto (the Local Organizing Committee members) for their
enormous efforts in planning and arranging the logistics of the conference from

registration/payment handling, venue preparation, accommodation booking, to
banquet/social program organization. We are very grateful to the IAT '01 sponsors:
ACM SIGART, Maebashi Institute of Technology, Maebashi Convention Bureau,
Maebashi City Government, Gunma Prefecture Government, The Japan Research
Institute, Limited, United States Air Force Office of Scientific Research, Asian
Office of Aerospace Research and Development, and United States Army Research
Office in Far East, and Web Intelligence Laboratory, Inc. for their generous support.
We thank ACM SIGWEB, SIGCHI, Japanese Society for Artificial Intelligence,
JSAI SIGFAI, SIGKBS, and IEICE SIGKBSE for being in cooperation with
IAT '01. Last but not the least, we thank Ms. Lakshmi Narayanan of World
Scientific for her help in coordinating the publication of this book.
October 2001
Ning Zhong and Jiming Liu
Program Committee Chairs
Setsuo Ohsuga and Jeffrey Bradshaw
General Conference Chairs
CONFERENCE ORGANIZATION
General Chairs:
Program Chairs:
Demos and Exhibits Chair:
Local Organizing Chair:
Jeffrey M. Bradshaw (Inst. H&M Cognition, USA)
Michele L. D. Gaudreault (US AOARD)
Daniel T. Ling (Microsoft Corp., USA)
Jiming Liu (Hong Kong Baptist U.)
Jianchang Mao (Verity Inc., USA)
Hiroshi Motoda (Osaka U., Japan)
Masahiko Satori (Maebashi Inst. Tech., Japan)
Tadaomi Miyazaki (Maebashi Inst. Tech., Japan)
Nobuo Otani (Mabashi Inst. Technology, Japan)

Sean M. Reedy (Mabashi Inst. Technology, Japan)
Ning Zhong (Maebashi Inst. Technology, Japan)
Setsuo Ohsuga (Waseda U., Japan)
Jeffrey Bradshaw (Inst. H&M Cognition, USA)
Ning Zhong (Maebashi Inst. Technology, Japan)
Jiming Liu (Hong Kong Baptist U.)
Jianchang Mao (Verity Inc., USA)
Nobuo Otani (Mabashi Inst. Technology, Japan)
Setsuo Ohsuga (Waseda U., Japan)
Patrick S. P. Wang (Northeastern U., USA)
Yiyu Yao (U. Regina, Cadada)
Jie Yang (U. Science & Technology of China)
Ning Zhong (Maebashi Inst. Technology, Japan)
Jan Zytkow (U. North Carolina, USA)
Toshio Kawamura (Maebashi Convention B.)
Masaaki Sakurai (Maebashi Convention Bureau)
Kanehisa Sekine (Maebashi Convention Bureau)
Midori Asaka (IPA, Japan)
Yoshitsugu Kakemoto (JRI, Limited, Japan)
International Advisory Board
Local Organizing Committee
Program Committee
K. Suzanne Barber (U. Texas-Austin, USA)
Guy Boy (EURISCO, France)
Cristiano Castelfranchi (CNR, Italy)
Kerstin Dautenhahn (U. Hertfordshire, UK)
Edmund H. Durfee (U. Michigan, USA)
E. A. Edmonds (Loughborough U., UK)
Tim Finin (UMBC, USA)
Adam Maria Gadomski (ENEA, Italy)

Scott Goodwin (U. Regina, Canada)
Vladimir Gorodetsky (Russian Academy of Sci.)
Mark Greaves (The Boeing Company, USA)
Barbara Hayes-Roth (Stanford U., USA)
Michael Huhns (U. South Carolina, USA)
Keniti Ida (Maebashi Inst. Technology, Japan)
Tom Ishida (Kyoka oto U., Japan)
Lakhmi Jain (U. South Australia)
Stefan J. Johansson (U. Karlskrona, Sweden)
Qun Jin (U. Aizu, Jaoan)
Juntae Kim (Dongguk U., Korea)
David Kinny (U. Melbourne, Australia)
Matthias Klusch (German Research Center for AI)
Sarit Kraus (U. Maryland, USA)
Danny B. Lange (General Magic, INC., USA)
Jimmy Ho Man Lee (Chinese U. Hong Kong)
Jiming Liu (Hong Kong Baptist U.)
Mike Luck (U. Southampton, UK)
Helen Meng (Chinese U. Hong Kong)
Joerg Mueller (Siemens, Germany)
Hideyuki Nakashima (ETL, Japan)
Wee-Keong Ng (Nanyang Tech. U., Singapore)
Katsumi Nitta (Tokyo Inst. Technology, Japan)
Yoshikuni Onozato (Gunma U., Japan)
Tuncer Oren (Marmara Research Center, Turkey)
Ichiro Osawa (ETL, Japan)
Sun Park (Rutgers U., USA)
Van Parunak (ERIM, USA)
Zbigniew W. Ras (U. North Carolina, USA)
Eugene Santos (U. Connecticut, USA)

Zhongzhi Shi (Chinese Academy of Sciences)
Carles Sierra (Scientific Research Council, Spain)
Kwang M. Sim (Chinese U. Hong Kong)
Andrzej Skowron (Warsaw U., Poland)
Ron Sun (U. Misouri-Columbia, USA)
Niranjan Suri (U. West Florida, USA)
Takao Terano (U. Tsukuba, Japan)
Demetri Terzopoulos (U. Toronto, Canada)
Huaglory Tianfield (Glasgow Caledonian U., UK)
David Wolpert (NASA Ames Research Center)
Jinglong Wu (Kagawa U., Japan)
Takahira Yamaguchi (Shizuoka U., Japan)
Kazumasa Yokota (Okayama Prefectural U., Japan)
Eric Yu (U. Toronto, Canada)
P.
C. Yuen (Hong Kong Baptist U.)
Chengqi Zhang (Deakin U., Australia)
Ning Zhong (Maebashi Inst. Technology, Japan)
TABLE OF CONTENTS
Preface v
Conference Organization vii
Invited Talks
Intelligent Agents for Market-Trend Prediction 2
Benjamin W. Wah
Social Intelligence Design for Knowledge Creating Communities 3
Toyoaki Nishida
Query Answering Based on Distributed Knowledge Mining 17
Zbigniew
W.
Ras

Approximate Reasoning by Agents in Distributed Environments 28
Andrzej Skowron
Multi-Agent Infrastructure for Agent Interoperation in Open
Computational Environments 40
Katia Sycara
Chapter 1. Formal Agent Theories
SPY: A Multi-Agent Model Yielding Semantic Properties 44
F. Buccafurri, D. Rosaci, G. M. L. Same, L. Palopoli
ABT with Asynchronous Reordering 54
Marius-Calin Silaghi, Djamila
Sam-Haroud,
Boi Faltlngs
Social Rationality and Cooperation 64
Guido Boella
Belief Revision in Type Theory 69
Tijn Borghuis, Fairouz Kamareddine, Rob Nederpelt
Heterogeneous BDI Agents II: Circumspect Agents 74
Maria Fash
A Preference-Driven Approach to Designing Agent Systems 80
Stefan J. Johansson, Johan Kummeneje
Agent Consumer Reports: of
the
Agents, by the Agents,
and for the Agents 86
Xiaocheng Luan, Yun
Peng,
Timothy Finin
Logical Formalizations Built on Game-Theoretic Argument
about Commitments 92
Lamber Royakkers, Vincent Buskens

Asynchronous Consistency Maintenance 98
Marius-Calin Silaghi, Djamila
Sam-Haroud,
Boi Faltings
IX
Chapter 2. Computational Architecture and Infrastructure
Reasoning about Mutual-Belief among Multiple Cooperative Agents 104
Wenpin Jiao
Portable Resource Control for Mobile Multi-Agent Systems in JAVA 114
Walter Binder, Jarle G. Hulaas, Alex Villazon, Rory G. Vidal
An Agent-Based Mobile E-Commerce Service Platform for
Forestry and Agriculture 119
Matthias Klusch, Andreas Gerber
An Itinerary Scripting Language for Mobile Agents in Enterprise
Applications 124
Seng
Wai
Loke, Arkady Zaslavsky, Brian Yap, Joseph Fonseka
Intelligent Agents for Mobile Commerce Services 129
Mihhail Matskin
A New Concept of Agent Architecture in Agentspace 134
T.
Nowak,
S. Ambroszkiewicz
21
st
Century Systems, INC.'s Agent Enabled Decision Guide
Environment (AEDGE™) 139
Plamen V. Petrov, Alexander D. Stoyen, Jeffrey D. Hicks,
Gregory J. Myers

Proactiveness and Effective Observer Mechanisms in Intelligent Agents 144
Jon Plumley, Kuo-Ming
Chao,
Rachid
Anane,
Nick Godwin
Chapter 3. Learning and Adaptation
Parrondo Strategies for Artificial Traders 150
Magnus Boman, Stefan J. Johansson, David Lyback
BDI Multi-Agent Learning Based on First-Order Induction of
Logical Decision Trees 160
Alejandro Guerra Hernandez, Amal El-Fallah Seghrouchni,
Henry Soldano
Evolutionary Behaviors of Competitive Agents in Dilemma Situation 170
Tin
Tin
Naing, Lifeng
He,
Atsuko Mutoh, Tsuyoshi Nakamura,
Hidenori Itoh
A Strategy for Creating Initial Data on Active Learning of Multi-Layer
Perceptron 180
Kazunori Iwata, Naohiro Ishii
Equilibrium Selection in a Sequential Multi-Issue Bargaining Model
with Evolutionary Agents 190
Norberto Eiji Nawa, Katsunori Shimohara, Osamu Katai
Affect and Agent Control: Experiments with Simple Affective States 200
Matthias Scheutz, Aaron Sloman
Meta-Learning Processes in Multi-Agent Systems 210
Ron Sun

Scalability and the Evolution of Normative Behavior 220
Jorg Wellner, Sigmar Papendick,
Werner
Dilger
Thinking-Learning by Argument 230
Aladdin Ayesh
Evolution of a Foraging Model with Many Individuals by Kin-selection 235
Kazue Kinoshita, Atsuko Mutoh, Tsuyoshi Nakamura,
Hidenori Itoh
The Use of Emergent Behaviour in a Multi-Agent System to Drive
Self-Adaptation at the Interface 240
Peter Marshall, Sue Greenwood
A Biologically Inspired Four Legged Robot That Exhibits Some Natural
Walking Behaviours 245
5.
Peng, G. R. Cole, C. P. Lam
Chapter 4. Knowledge Discovery and Data Mining Agents
CM-RELVIEW: A Tool for Causal Reasoning in Multi-Agent
Environments 252
Brahim Chaib-Draa
User's Ontology-Based Autonomous Interface Agents 264
Tarek
Helmy,
Satoshi
Amamiya,
Makoto Amamiya
Integration and Reuse of Heterogeneous XML DTDs for
Information Agents 274
Euna Jeong, Chun-Nan Hsu
Virtual Museum's Assistant 284

Osvaldo Cairo, Ana Aldeco, M.E. Algorri
Index Based Document Classification with CC4 Neural Networks 289
Enhong Chen, Zhengya Zhang, Xufa
Wang,
Jie Yang
Price Watcher Agent for E-Commerce 294
Simon Fong, Aixin Sun, Kin Keong Wong
Automated Information Extraction from Web Pages Using
an Interactive Learning Agent 300
Jugal
K.
Kalita, Paritosh Rohilla
An Intelligent Agent with Structured Pattern Matching for
a Virtual Representative 305
Seung-ik
Lee,
Sung-Bae Cho
A Calendar Management Agent with Fuzzy Logic 310
Wayne Wobcke
XML Based Multi-Agent Collaboration for Active Digital Libraries 315
Yanyan
Yang,
Omer
F.
Rana, David
W.
Walker,
Roy Williams, Giovanni Aloisio
XII
Chapter 5. Distributed Intelligence

An Intelligent Channel Allocation Scheme for Mobile Networks:
An Application of Agent Technology 322
Eliane
L.
Bodanese, Laurie G. Cuthbert
An Atomic Approach to Agent-Based Imagery and Geospatial
Problem Solving 334
James
J.
Nolan, Robert Simon, Arun K, Sood
Model-Based Creation of Agents and Distribution of Problem Solving 344
Katsuaki Tanaka, Setsuo Ohsuga
A Distributed Algorithm for Coalition Formation Among
E-Commerce Agents 355
Guillaume
Vauvert,
Amal El Fallah-Seghrouchni
Optimal Reward Functions in Distributed Reinforcement Learning 365
David
H.
Wolpert, Kagan
Turner
Polygonal Approximation of Planar Digital Curves Using Ant System 375
Peng-Yeng Yin
A Biological View on Information Ecosystems 385
Bengt Carlsson, Paul Davidsson
The CoDAC Collaboration Framework 390
K
W.
Ng, T.

O.
Lee
A Multi-Agent Approach to Modelling Interaction in Human
Mathematical Reasoning 395
Alison Pease, Simon
Colton,
Alan Smaill, John Lee
Secure Asynchronous Search 400
Marius-Calin Silaghi, Djamila Sam-Haroud, Boi Faltings
Foundations of Market-Driven Agents: An Adaptation of Zeuthen's
Bargaining Model 405
Kwang Mong Sim, Chung Yu Choi
Chapter 6. Agent Based Applications
Kavanah: An Active User Interface Information Retrieval Application 412
Eugene Santos JR., Hien Nguyen, Scott
M.
Brown
iJADE WeatherMAN - A Multi-Agent Fuzzy-Neuro Network Based
Weather Prediction System 424
Raymond
Lee,
James Liu, Jane You
Acquaintance Models in Coalition Planning for Humanitarian
Relief Operation 434
Michal Pechoucek, Vladimir
Marik,
Jaroslav Barta
Agent Negotiation in a Virtual Marketplace 444
Walid S. Saba, Pratap R. Sathi
XIII

Modeling User Preferences to Facilitate More Automated and Accurate
Transaction Brokering within Heterogeneous Multi-Agent Electronic
Markets 454
G. Tewari, P. Maes, A. Berkovich, V. Gabovich
Attitude Based Agents in E-Commerce Applications 464
S. Au, N. Parameswaran
Organizing Internet Agents According to a Hierarchy of
Information Domains 469
Sylvie Cazalens, Philippe Lamarre
Introducing User Preference Modeling for Meeting Scheduling 474
Hon Wai
Chun,
Rebecca Y. M. Wong
Executive Attentional Control in Autonomous Robotic Agents 479
Jason
Garforth,
Anthony Meehan, Sue Mchale
Implementation and Analysis of Mobile Agents in a Simulation
Environment for Fieldbus Systems 484
R.
Hunstock,
U.
Ruckert,
T.
Hanna
Evaluating Believability in an Interactive Narrative 490
Jarmo Laaksolahti, Per
Persson,
Carolina Palo
iJADE Stock Predictor - An Intelligent Multi-Agent Based Time Series

Stock Prediction System 495
Raymond
S.
T.
Lee, James N. K. Liu
Approximate Sensor Fusion in a Navigation Agent 500
J. F. Peters, S. Ramanna, M. Borkowski, A. Skowron
Simulating Day-Ahead Trading in Electricity Markets with Agents 505
Max Scheldt, Hans-Jurgen Sebastian
Using Mobile Agents to Update and Maintain Course Materials on
Students' Computers in Internet-Based Distance Education 510
Hongxue Wang, Pete Holt
Author Index
515
INVITED TALKS
INTELLIGENT AGENTS FOR MARKET-TREND
PREDICTION
BENJAMIN W. WAH
Department of Electrical and Computer Engineering
and the Coordinated Science Laboratory
University of Illinois at Urbana- Champaign
Urbana, IL 61801, USA

(2001 IEEE Computer Society President)
In this presentation we discuss the role of intelligent agents in market-
trend predictions. Market-trend data, such as stock-market data, are charac-
terized by non-stationary time series that may depend on non-numeric and
non-quantifiable measures. The prediction of market trends, therefore, should
consist of prediction of non-stationary time series and the abstraction and in-
tegration of non-numeric information in prediction. In this talk, we survey

various prediction techniques for and mining of market-trend data. We pro-
pose to use intelligent agents in the abstraction of non-numeric information,
the decomposition of non-stationary time series into multiple stationary time
series,
and the prediction of trends using artificial neural networks. Finally,
we illustrate our techniques in predicting stock-market data.
2
SOCIAL INTELLIGENCE DESIGN
FOR KNOWLEDGE CREATING COMMUNITIES
TOYOAKI NISHIDA
Department of Information and Communication Engineering
Graduate School of Information Science and Technology
The University of Tokyo
7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan
nishida@kc.
t.
u-tokyo. ac.jp
Communities play an important role in knowledge creation by providing people with
opportunities to continually learn from others, find partners to collaborate with, and
demonstrate the significance of their disciplines. In education or business, it is relatively easy
to find typical examples of knowledge creating communities for sharing and exchanging
specialized knowledge among knowledge workers. In other domains such as NPO or local
communities, people are naturally practicing mutual learning and invaluable knowledge is
built as a result, even if knowledge creation is not deemed a primary goal of the community.
In this paper, 1 present an interdisciplinary approach to augmenting the community
knowledge creating process by integrating insights from social psychology, cognitive
psychology, and advanced information technology. I emphasize the role of conversations and
stories as a means of establishing a common background in a community.
I describe several systems that primarily use the conversational modality to mediate
community communication. Among others, EgoChat allows the user to make conversation

with virtualized egos responding on behalf of other users. It allows the user to take an
initiative by interrupting the conversation and changing its flow. VoiceCafe allows artifacts
to make conversation with people or other artifacts. It stimulates creative thinking by
bringing about utterances from the physical object's point of view, which might be strikingly
different from humans' view.
These engineering approaches should be tightly coupled with sociological and cognitive
approaches, to predict and assess the effects of community communication mediation
systems on the human society. 1 discuss issues on designing a constructive framework of
interaction for achieving practical goals without being caught by known pathological pitfalls
of group interactions.
1 Introduction
The agent technology plays a diverse role in the networked society. On the one
hand, agents may be intelligent surrogates that work on behalf of the user. This type
of
agents
includes intelligent brokers that seek best match between service providers
and consumers, intelligent traders that buy and sell goods on behalf of the user,
intelligent decision makers that negotiate contracts for the user, and so on.
Alternatively, agents may be embodied conversational interfaces that entertain the
3
4
user. This type of agents is becoming popular as agent portals on the Internet, or as
artificial pets in the entertainment and amusement domain.
In this paper, I discuss issues in applying the agent technology to the
development of a social information service for mediating communication among
people. From this perspective, the central issue is designing and understanding a
world where people and agents cohabit, rather than inventing a system of artifacts.
We will not be able to innovate a totally new kingdom of artificial agents apart from
the human society, but we have to carefully embed the agent system in the existing
human society. This means that we need to understand more about humans and the

human society to better design an embedded system. We need to pay much attention
on the effects the technology brings about the human society. We need to make
every effort to have the proposal accepted by the human community. In contrast, we
need not insist on the novelty of the technology or a pedagogical issue of whether
the artifact can be called an agent.
Let us call this field social intelligence design in general. Research on social
intelligence design involves such issues as how new technologies induce the
emergence of a new language and lifestyle. For example, interactive multimedia
websites are a new medium and maybe even a new language, with interesting new
conventions, and increasing adaptation to the support of communities. Japanese
teenagers have developed a new language for use originally with beepers and now
with mobile phones. These are both new mainstream real world developments that
should be studied further, and could probably give some valuable insights. The
theme of "social intelligence" is really an angle on the support of groups in pursuit
of their goals, whether that is medical knowledge, stock trading, or teenage gossip.
I focus on community support systems to shed light on key aspects of social
intelligence design. The goal of a community support system is to facilitate
formation and maintenance of human and knowledge networks to support activities
in a community. Examples of community support systems include socially
intelligent agents that mediate people in getting to know and communicate with each
other, a collaborative virtual environment for large-scale discussions, personalized
agents for helping cross-cultural communication, interactive community media for
augmenting community awareness and memory, to name just a few.
I emphasize the role of stories and conversations as a means of establishing a
common background in a community. Stories allow us to put pieces of information
into an intelligible structure. Conversations give us an opportunity to examine
information from various angles and search for a good story structure. In some
community support systems, story-telling agents play a central role. It should be
noted that their significance depends more on the contents of stories rather than
conversation mechanism.

I also emphasize the empirical aspects of social intelligence design.
Engineering approaches should be tightly coupled with sociological and cognitive
approaches, to predict and assess the effects of community communication
5
mediation systems on the human society. I show how psychological approaches are
applied to design and evaluation of community support systems.
2 Communities and Social Intelligence
Social intelligence design is distinguished from most of other conventional
engineering disciplines in that we have to be strongly aware of
the
human society as
a target. For this reason, I first take a look at the nature of my target, i.e.,
communities, in this section.
A community is a group of people loosely coupled by a shared interest or
environment. More formal discussion can be found in literature in sociology. For
example, Smith defines a community as follows:
Generically, a community can be understood as a set of on-going social
relations bound together by a common interest or shared circumstance. As a
result, communities may be intentional or unintentional, a community's
participants may purposely join together or be thrust into membership by
circumstance. Intentional communities are of particular interest because they
raise more questions about the reasons and causes for their emergence than do
unintentional ones [21].
Traditional communities were local communities that are characterized by
locality and shared living environment. The advent of a global information network
has not only considerably relaxed spatial constraints for communities to be built, but
also provided a new opportunities for existing communities. Typical networked
communities include:
• communities of interest, in which people are tied with a shared interest;
• communities of

practice,
in which a group of people work together and share
a common work practice; and
• enhanced local communities or smart communities, which result from
enhancing communication and information sharing facilities in existing local
communities.
Schlichter contrasts communities with groups and teams [23]. He characterizes
communities as sets of people who share something but who do not necessary know
each other or interact on personal basis. In contrast, groups are sets of people who
know each other but who do not necessarily cooperate, while teams are sets of
people who are cooperating to achieve a common goal. In educational
environments, the class of lecture may be regarded as a community, a discussion
group a group, and a learning group a team.
Recently, communities have become increasingly paid more attention in the
context of knowledge management and distance learning. A community provides its
members with opportunities to continually learn from others, find partners to
collaborate with, and demonstrate the significance of their disciplines. In education
6
or business, it is relatively easy to find examples of communities oriented towards
knowledge creation by sharing and exchanging specialized knowledge among
knowledge workers. In other domains such as NPO or local communities, people
are naturally practicing mutual learning and invaluable knowledge is built as a
result, even if knowledge creation is not deemed a primary goal of the community.
We consider that community knowledge creation is essentially a co-evolution of
human and knowledge networks [16, 17]. By human network, I mean a collection of
people connected by various social relations, such as acquaintance or partnership. A
human network is considered to embody tacit knowledge that may be shared in a
community but may not be explicitly spoken. In contrast, knowledge network is a
collection of documents or files connected explicitly by hyperlinks or implicitly by
references. Knowledge network explicitly describes shared knowledge and interest

in a community.
A knowledge network enables people with a common interest to know each
other, resulting in extension of human network. A human network, in turn, helps
new ideas grow through intimate discussions. It facilitates the extension of
knowledge network through publication of new knowledge. Thus, a synergetic cycle
of human and knowledge network will lead to a successful community.
A more elaborate characterization of human and knowledge networks is
proposed by Contractor [3]. He pointed out that observed knowledge networks are
different from cognitive networks that each individual possesses as a cognitive
perception of the network. He proposes to distinguish between: (i) knowledge
networks that represent the extent to which the same or disparate knowledge is
distributed among various members of the group, and (ii) cognitive knowledge
networks that represent individuals' cognitive perceptions of "who knows what"
within the group.
In order to understand the dynamics of community knowledge, Contractor
proposes to observe five types of network data: (i) a communication network of
actors based on existing tasks and project links between them, (ii) a knowledge
network based on actors providing an inventory of their skills and expertise, (iii) a
knowledge network of actors based on links between their web sites, (iv) a
knowledge network of actors based on common links from their web sites, and (v) a
knowledge network based on similarity in content between different actors' web
sites.
It should be noted that all kinds of interaction in a community may not bring
about fruitful results. In social psychology, various pathological pitfalls are known
about group interactions. A notorious example is flaming, an endless slander battle
on the net, which is rare in face-to-face communication. Flaming blocks discussions
among community members, possibly resulting in a destructive damage to a
community. False consensus is another undesirable phenomenon. It results from "a
spiral of silence", or "bandwagon effect", for instance, in which false cognition is
socially amplified.

7
3 Community Support Systems
The role of community support systems is to support community activities by
providing a communication channel for community members. Community support
systems are built on top of the communication and expected to help community
members (i) exchange awareness with other members, (ii) explore human and
knowledge networks, (iii) build community knowledge, (iv) organize public events,
(vi) form a group/team for collaborative work, (v) negotiate with others, and (vii)
discuss public issues and make decisions about the community. Community support
systems provide rather long-range, bottom-up communicative functions in the
background of daily life. This feature is contrastive with groupware that emphasizes
more task-driven, short-range collaboration, although awareness is equally
emphasized. In the rest of this section, I will discuss the first three functions.
3.1 Helping to Awareness with Other Members
Most of networked communities are based on intentional participation, based on a
common interest for instance. Compared with mission-oriented groups where
participants are incorporated in a certain work structure, the degree of necessity to
exchange awareness is relatively low in networked communities. Participants tend
to become silent unless a mechanism is provided for lowering the cost for
exchanging awareness with other members.
In order to support awareness, Schlichter uses spatial metaphors such as rooms
or hallways in "The Lecture 2000", a computational environment for supporting a
learning community. FaintPop supports a light-weight, acknowledge-only mode of
communications [19]. The major design goal of FaintPop is to communicate the
sense of connectedness, not to perform informative functions. FaintPop is a
communication device similar to a photo frame. Small photos or icons of the user's
colleagues are displayed in the frame, through which the user can communicate with
other users using a simple touch actions. Three types of touching are permitted: a
tap to communicate a neutral feeling, a pet a positive feeling, and a hit a negative
feeling. The user can communicate her/his feeling towards her/his colleagues by

using these three types of touching and other community members can observe it.
Sumi proposes to use interest-based information distribution system, which pushes
information to interested users, rather than passively waits for requests from users
[24].
Voice Cafe [8] allows artifacts to make conversation with people or other
artifacts (Figure 1). It stimulates creative thinking by bringing about utterances from
the physical object's point of
view,
which might be strikingly different from humans'
view. Each Voice Cafe artifact consists of a physical object and a conversational
agent. It can communicate with community members by exchanging gossips, or
small talks about members' conditions, schedules, thoughts and opinions, and so on.
8
(a) the conceptual framework of Voice Cafe
Figure 1. Virtualized egos as an interactive community medium.
By listening to the gossips, members can gain awareness of other people at the small
talk level.
3.2
Helping to Explore Human and Knowledge Networks
This facility helps the user
find
human and human resources in a community. Social
matchmaking is frequently used to locate people on the Internet who share some
similar interests and enable the automatic formation of interest group.
Social matchmaking calculates the distance between users by referring to their
user profiles. A major motivation
.
behind social matchmaking is to address
situations such that finding an expert is difficult and time consuming; people are
often working on similar projects without realizing it; or people feel socially isolated

9
Real World Inhabitant
Figure 2. Virtualized egos as an interactive community medium.
because nobody around s/he seems to share the same Interest. Yenta [4] is a multi-
agent matchmaking system that can automatically determine user interests and
operate in a completely decentralized, peer-to-peer fashion. Yenta is a persistent
agent that uses referrals to find each other, build clusters of like-minded agents, and
introduce users to each other. Special care is paid to protect user privacy.
Silhouettell [20] combines awareness support and social matchmaking to bridge
between informal and formal meetings. It projects the location of participants on the
screen as shadows, and facilitates conversation by presenting Web pages that are
inferred to common to the participants.
Referral Web [11] integrates recommendations and search through the concept
of
a
social network. It helps the user discover her/his relationship to the best human
experts for a given topic. It gathers all information from public sources, which
removes the cost of information posting and registration. It can also explain the user
why each link in the referral-chain appeared.
In order to provide an integrated method of exploring and building human and
knowledge networks, we use a talking-virtualized-egos metaphor in CoMeMo-
Community [14] and EgoChat [12] to enable an elaborate asynchronous
communication among community members. A virtualized ego mainly plays two
functions (Figure 2). First, it stores and maintains the user's personal memory.
Second, it presents the content of the personal memory on behalf of the user at
appropriate situations. By personal memory, we mean an aggregation of relevant
information represented in the context specific to a particular person. Personal
memory plays a crucial role not only in personal information management but also
in mutual understanding in a community.
A virtualized ego serves as a portal to the memory and knowledge of a person.

It accumulates information about a person and allows her/his colleague to access the
information by following an ordinary spoken-language conversation mode, not by
10
going up and down a complex directory in search for possibly existent information,
or by deliberately issuing commands for information retrieval. In addition,
virtualized ego embodies tacit and non-verbal knowledge about the person so that
more subtle messages such as attitude can be communicated.
As is also the case with Voice Cafe, we take a conversation-centered approach
in designing intelligent systems and capturing intelligence
itself.
Conversation plays
varieties of roles in human societies. It not only allows people to exchange
information, but it also helps them create new ideas or manage human relations. In
our approach, more emphasis is placed on creating, exchanging, reorganizing, and
utilizing conversational contents in knowledge creation, rather than implementing
intelligent agents or yet-another human interface.
3.3 Helping to Build Community Knowledge
The third function of a community support system is for helping community
members build a shared knowledge. Nonaka and Takeuchi pointed out that the
community knowledge is built by a spiral of interactions between explicit and tacit
knowledge [18]. They suggest that the process of knowledge creation is more
important than the body of knowledge, for people often find more value in
communities that evolve as a result of learning. This implies that more emphasis
should be placed on supporting interactions or the emergent aspect of community
knowledge [13] in community support systems.
The Public Opinion Channel (POC) [15, 16, 7] is a community-wide interactive
broadcasting system (Figure 3). A POC continuously collects messages from people
in a community and feeds edited messages back to them. POC is not intended to be
a system that broadcasts public opinions per se. Instead, it is intended to broadcast
miscellaneous information that can serve as a basis of public opinion formation.

A POC repeats a cycle consisting of call-for-opinion followed by one or more
repetition of responding by the community members and summarization by the POC
system. In the initial call-for-opinion message, the POC system specifies a focus of
discussion. Alternatively, people may also initiate discussion by submitting a topic.
Then, interested community members may respond with messages. In principle,
messages are not limited to pure opinions. Instead, they may include questions,
stories, findings, jokes, proposals, and all other message types. The POC system
may combine these messages, generate a story, and broadcast it to the community.
The POC system may issue a progress report based on responses from community
members. The process proceeds with altering subjects.
A POC brings about ecology of ever evolving stories. People can access to the
story pool at anytime by an on-demand-type access means. Another thing I would
like to emphasize here is that the POC broadcasting can be embedded in the ambient
environment, just like a radio broadcasting, so that people need not pay much
attention at all times.
11
Community
Broadcasting
Servers
(POC Servers)
POC
Clients
We have to
discuss of the
ethics of cyborg .^j
Advertising/or
Opinions
'
Community B
Broadcasting Opinions

. cyborg
nil be
Community C
Figure 3. The Conceptual framework of Public Opinion Channel (POC). The POC is an interactive
broadcasting system that continuously collects messages from community members and feeds edited
message streams back to the community.
Compared with existing mass media, a POC has various advantages.
Computational support and network connectivity enable a large amount of responses
to be analyzed on the fly, allowing real-time interactive stories to be generated. In
particular, a combination of statistical computation and semantic processing permits
minority opinions to be reflected in the structure of public opinion.
We believe that POC also contributes to community knowledge building and
public discussion.
4 Social Intelligence Design
Social Intelligence Design is a new discipline aimed at understanding and
supporting social intelligence, i.e., intelligence collectively exhibited by
(natural/artificial) agents to dynamically organize members' activities into a coherent
one by exploiting or innovating the social structure. Social intelligence models
intelligence as a phenomenon emerging from the way agents, either natural or
artificial, are interacting with each other. Research into community support systems
is concerned with engineering aspects of Social Intelligence Design. Meanwhile,
investigation into the sociological and cognitive aspects are equally or sometimes
more important. Engineering approach should be tightly coupled with sociology and
psychology and other disciplines closely related to the study of humans and human
society. Thus, Social Intelligence Design involves not only designing artifacts but