Reviewed by
Julia Frolova
Department of Cybernetics
Faculty of Computer Science
Omsk State University
Russia
This book presents selected tutorial lectures given at the summer school on Artificial Intelligence (ACAI 2001) with the subtitle "Multi-Agent Systems and Their Applications", held in Prague, Czech Republic, in July 2001. This was a joint event between ECCAI (the European Co-ordinating Committee for Artificial Intelligence) and AgentLink, the European Network of Excellence for Agent-Based Computing.
The 20 lectures by invited researchers presented in the book show the current state of the art in the theoretical foundations of multi-agent systems as well as demonstrating the applicability of these systems for practical tasks.
The papers are grouped into three sections: Foundations of Multi-agent Systems, Social Behaviour, Meta-reasoning and Learning; and Applications. The Foundations of Multi-agent Systems section includes eight papers addressing modelling technologies for multi-agent environments. The papers compiled in the Social Behaviour, Meta-reasoning and Learning section are intended to show that using biological and social analogies can be successful for designing complex multi-agent systems. The Applications section comprises six papers, all presenting applications of multi-agent systems in different areas. I will now describe the contents of these sections in turn.
The aim of the paper "Perspectives on Organisations in Multi-agent Systems" by Les Gasser is to illustrate and sensitise the reader to the variety of perspectives available for the simulation of organisations and to the fundamental nature of organisations as stable systems and as multi-perspective action systems.
The paper "Multi-agent Infrastructure, Agent Discovery and Middle Agents for Web Services and Interoperation" by Katia Sycara has two parts. In Part I, the author presents an overview of issues in modelling Multi-agent Systems (MAS), discusses the features and components required for MAS infrastructure and presents a model of generic infrastructure. The author then presents RETSINA as an example of an implemented MAS infrastructure. The RETSINA (Reusable Task Structure-based Intelligent Network Agents) Multi-agent infrastructure was developed at Carnegie Mellon University in Pittsburgh. In Part II, the author presents issues in agent and service discovery and interoperation through a set of domain-independent, active and intelligent registries called middle agents.
In "Logical Foundations of Agent-Based Computing", Wiebe van der Hoek shows that modal logic provides a nice tool for defining informational, motivational and dynamical aspects of agents. The author concludes by showing how an agent programming language can also benefit from this modal approach.
Yannis Labrou, in "Standardising Agent Communication", presents questions about standardising agent communication and the work that addresses them, alongside the historical evolution of Agent Communication Languages (ACL), their semantics and the results of their standardisation. ACL is a collection of speech-act-like message types, with agreed-upon semantics, with facilitate knowledge and information exchange between software agents. From Knowledge Query and Manipulation Language (KQML) to FIPA ACL, ACL have been a cornerstone for the development of systems with communicating agents, and simultaneously they have been the subject of intensive standardisation efforts. The Foundation for Intelligent Physical Agents (FIPA) is a non-profit association whose purpose is to promote the success of emerging agent-based applications and services. Formed in 1996, FIPA was formed to provide a forum for developing specifications for agent systems. Its goal is to make available specifications that maximise interoperability across agent-based systems.
In the article "Standardising Agent Interoperability: The FIPA Approach", Stefan Poslad and Patricia Charlton discuss both technical and scientific issues in defining standards for interoperability between agents in different Multi-agent Systems with a particular focus on the FIPA agent interoperability standards.
Edmund H. Durfee, in "Distributed Problem Solving and Planning", characterises the variations of distributed problem solving and distributed planning and summarises some of the basic techniques that have been developed. The author considers that in order for a problem to be a distributed planning problem each agent must formulate plans for action that (sufficiently well) take into account the plans of other agents.
The paper "Automated Negotiation and Decision Making in Multi-agent Environments" by Sarit Kraus presents some of the key techniques for reaching agreements in multi-agent environments. It discusses game theory and economics based techniques: strategic negotiation, auctions, coalition formation, market-oriented programming and contracting. It also presents logic-based mechanisms for negotiation. The focus of the survey is on negotiation by self-interested agents, but the paper also considers several mechanisms for co-operative agents who need to resolve conflicts that arise from different beliefs about the environment.
In "Agents' Advanced Features for Negotiation and Co-ordination", Eugenio Oliveira briefly summarises some proposals about agent negotiation capabilities including adaptation through reinforcement learning as well as qualitative multi-criteria negotiation and coalition formation protocols. Inspired by the robo-soccer domain, he also offers some basic hints about knowledge representation for agent team work.
Milind Tambe and David V. Pynadath in their paper "Towards Heterogeneous Agent Teams" illustrate how the Teamcore architecture successfully addressed the challenges of agent integration in two application domains: simulated rehearsal of a military evacuation mission and facilitation of human collaboration. A key novelty and strength of this framework is that powerful teamwork capabilities are built into its foundations by providing the proxies themselves with a teamwork model called STEAM.
The paper "Social Knowledge in Multi-agent Systems" by Vladimir Marik, Michal Pechoucek and Olga Stepankova addresses the problems of efficient representation, maintenance and exploration of social knowledge thus enabling task decomposition, organisation of negotiations, responsibility delegation and other forms of social reasoning for agents. Authors focus on multi-agent systems for integration of pre-existing software components. A specific tri-base acquaintance model (3bA) is formalised and discussed throughout the paper. This model helps to optimise communication traffic, to implement meta-reasoning processes and supports machine learning activities. Several practical applications of the 3bA acquaintance model are presented in different fields and the acquired experience is discussed.
In the paper "Machine Learning and Inductive Logic Programming for Multi-agent Systems", Dimitar Kazakov and Daniel Kudenko focus on important issues surrounding the application of machine learning (ML) techniques to agents and MAS. In this discussion, the authors move from disembodied ML over single agent learning to full multi-agent learning. In the second part of the paper, the authors focus on the application of Inductive Logic Programming (a knowledge-based ML technique) to MAS and present an implemented framework in which multi-agent learning experiments can be carried out.
The paper "Relational Reinforcement Learning" by Kurt Driessens presents an introduction to reinforcement learning and relational reinforcement learning at a level that can understood by non experts. The author gives an overview of the fundamental principles and techniques of reinforcement learning without any rigorous mathematical derivations through the use of an example application. Relational reinforcement learning is then presented as a combination of reinforcement learning and relational learning. The paper discusses the advantages of this method which include the possibility of using structural representations, making abstractions from specific goals pursued and exploiting the results of previous learning phases.
In "From Statistics to Emergence: Exercises in Systems Modularity", Jozef Kelemen sketches several different ways of considering complex systems in terms of their modularity. These range from considering whole systems without any regard to their modularisation, through systems composed from functionally specified modules and finally to post-modular systems consisting of relatively independent autonomous modules sharing a common environment and acting in more or less co-ordinated ways. He then presents, illustrates and discusses a relatively simple, uniform and productive theoretical framework (the theory of grammar systems) for the study of these aspects of systems behaviour and modularity.
Paolo Petta and Robert Trappl in their paper "Emotions and Agents" discuss the possible co-operation between emotion research and agent-based technology. On one hand, results from emotion research start to serve as a role model from nature, inspiring technical design criteria for individual agents at the micro level and agent groups and societies at the macro level as well as the sophisticated linkages between them. On the other hand, emotions are of immediate impact in important aspects of human-agent interaction and effective social co-operation between humans and conversational interfaces. In this broad survey, the authors offer an interesting selection of results from different areas of emotion research.
In their paper ("Multi-agent Co-ordination and Control Using Stigmery Applied to Manufacturing Control"), Paul Valckenaers, Hendrik Van Brussel, Martin Kollingbaum and Olaf Bochmann discuss multi-agent co-ordination and control using techniques inspired by the behaviour of social insects. They present a system design that enables desirable overall behaviour to emerge without exposing the individual agents to the complexity and dynamics of the overall system. The research focuses on manufacturing control.
The article "Virtual Enterprise Modelling and Support Infrastructures: Applying Multi-agent System Approaches" by Luis M. Camarinha-Matos and Hamideh Afsarmanesh summarises the main challenges in the field of virtual enterprises and describes several current Multi-agent System application approaches. A particular emphasis is given to the creation and operation phases of the virtual enterprise life cycle.
The article "Specialised Agent Applications" tries to give an overview of MAS applications in specific areas. Klaus Fischer, Petra Funk and Christian Russ concentrate on the application of MAS in the context of supply chain management in virtual enterprises. In this context, effective system behaviour requires both structure and organisation. To achieve this, the authors present the concept of "holonic" MAS and demonstrate how it can be used in the selected application domain.
In "Agent-Based Modelling of Ecosystems for Sustainable Resource Management", Jim Doran discusses agent-based modelling and social simulation with particular regard to ecosystem management. He describes and illustrates the steps involved in designing and building an agent-based model as well as the methodological problems typically encountered. The goals of integrated ecosystem management are examined and examples are given of agent-based modelling in this context. As a further illustration, consideration is given to a possible agent-based model of the Fraser River watershed in British Columbia and to the particular difficulties that such a model presents.
In the paper "Co-operating Physical Robots: A Lesson in Playing Robotic Soccer" Bernhard Nebel sketches some successes and pitfalls that arise in the development and testing of a system to play robotic soccer.
The purpose of the final paper "A Multi-agent Study of Interethnic Co-operation" by Vladimir Kvasnicka and Jiri Pospichal is to present a simulation study of co-operation between two ethnic groups. The approach chosen is a reformulation of the evolutionary Prisoner's Dilemma, where a population of strategies is evolved by applying a simple reproduction process with a Darwinian metaphor of natural selection (the probability of selection for reproduction is proportional to fitness). The computer simulations show that applying a principle of collective guilt does not lead to the emergence of interethnic co-operation.
As a whole, the book has three things to offer. Firstly, it serves as a tutorial, presenting an introduction to multi-agent system learning and its applications at a level that can be understood by students and researchers from different backgrounds. Secondly, the authors give pointers to related work and to the general literature for application-oriented research on MAS. Thirdly, nearly all of the papers supply the figures, diagrams or examples that facilitate understanding of the topic under discussion.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2005