Reviewed by
Cesáreo Hernández-Iglesias
E.T.S.I.Industriales. University of Valladolid
This book is the sixth volume of the "ABSS-Agent Based Social Systems" series. In the editor's own words, this series is intended to further the creation of the science of agent-based social systems. Each volume has been devoted to a collection of selected papers from workshops and conferences, mainly the annual International Workshop on Agent-based Approaches in Economic and Social Complex Systems (AESCS), organized by the Pacific Asian Association for Agent-Based Social Simulation (PAAA). It consists of a selection of 21 papers from those presented at the AESCS'07, held in Tokyo with two invited papers, one by Prof. Shu-Heng Chen from the National Chengchi University of Taiwan and the other one by M. Afzal and Dr. Richard Warren, from the Air Force Research Lab. Ohio. 29 papers were submitted to the Conference and 25 were accepted for presentation. The main pitfall of post-proceedings books is the lack of homogeneity of the papers. In many cases, they are just a collection of articles written by authors who met around a general topic, reflecting the individual contributions to the conference. Eventually, this problem is avoided by including feedback from the presentations discussions. Unfortunately, there is not such a feedback in this book. In such cases, a major challenge for the editors is to design a proper thread to make of the book a coherent conceptual framework. This is even more difficult when, as it happens with the AESCS'07, a book covers such a wide range of topics under the umbrella of Agent-Based Approaches to both Economic and Engineering complex systems.
These comments should not be taken as an assertion that the book does not achieve the goal of the ABSS Series, that is "to create a conceptual framework and design theory of socioeconomic systems of the twenty-first century in a cross-cultural and trans disciplinary context". What I consider is that a preface making explicit the thread of the Conference and its contribution towards these goals would have added substantial value to the book.
Apart from the plenary and invited talks, the book's chapters are classified according to the following broad topics: Organization and Management, Fundamentals of Agent Based and Evolutionary Approaches, Production, Services and Urban Systems, Agent-Based Approaches to Social systems, Market and Economy I, and Market and Economy II.
The plenary talk by Shu-Heng Chen - "Genetic Programming and Agent-Based Computational Economics: From Autonomous Agents to Product Innovation" - is a nice and original paper. The author thinks that ACE models have been generally weak in demonstrating discovery or a novel generating process to explain the core issues in economics. In this sense - he claims - they are not very different from their counterparts in neoclassical economics. To make progress, the author proposes a procedure to enable autonomous and heterogeneous agents to discover the modular structure of their neighbourhoods, and hence the agents can adapt by using these modules. He presents a genetic programming (GP) model complemented with what he calls ADTs: Automatic Defined Terminals. He brings back from the control and filtering theory of the seventies the Gram Schmidt orthogonalization process to define the ADTs hierarchical structures, and the modular knowledge generation. Apart from the interest of the model, Shu-Heng Chen makes a very valuable set of arguments to show that genetic programming helps capturing the three core elements for economics to catch up with the economy. Constant change that he relates with A. Marshall, adaptive populations of decision rules (Lucas critique) and modularity that he relates with Herbert Simon. In short, a very valuable contribution to economics.
The invited paper by Afzal and Warren - "Simulating the Emergence of Complex Cultural Beliefs" - should be of interest for JASSS readers. The authors define and construct a CCI Society, embedded into an AI domain called Multiagent Wumpus World (MWW). The acronym CCI, which stands for Communicate, Comprehend and Integrate, means that agents have a causal model of the environment, are goal-directed and can communicate and share information. The authors perform several experiments to show how CCI agents are capable of modelling other agents populating the environment and of reasoning so as to capture others' beliefs and intentions. A Pareto`s principle emerges in terms of information diffusion that has practical implications in marketing and information management.
Three papers are included in the "Organization and Management" section. The first one is "Synchronization in Mobile Agents AND Effects of Network Topology" by Aoyagi and Namatame. The authors study the relationship between flocking behaviour and network topologies and their simulations show that flocking behaviour can emerge and converge to stability using a few external links and a network with local information. The second one is "Evaluation of Mass User Support Strategies in the Theme Park Problem" by Yanagita and Suzuki This paper allows to conclude that, to handle the TPP, one should build small-world network and scale free networks so as to test for the effects of visitor's preferences, congestion and the moving cost on the coordination algorithm. "Agent-based simulation to analyze business office activities using reinforcing learning" by Kenjo et al. attempt to clarify team behaviour in cooperative organizations by agent-based simulations. For a hierarchical team, their simulations show that there is a relationship between the group form and their capacity to adapt to environmental changes. A high performer in the organization may fail to perform well in a different environment.
The "Fundamentals of Agent-Based and Evolutionary Approaches" section consists of three papers: "A Model of Mental Model Formation in a Social Context" by Gostoli, "Thought on the Continuity Hypothesis and the Origins" by Taniguchi, and "Modelling a small agent society" by Indo. In Taniguchi`s paper the issue of the "continuity hypothesis" can be truly qualified as fundamentals, although the discussion is based on rather rhetoric arguments without explicit connection to agent-based modelling. The other two papers do not seem to contribute to the fundamentals of "agent-based and evolutionary approaches".
The "Production, Services and Urban Systems" section includes four papers. Two of them are related to spatial models, one to production management and one related to the advertising and production strategic mix. Kaneda and He developed a pedestrian flow simulator with a route optimization function that they consider it could be applied to the analysis of crowd density in spaces with complicated shapes. Yoshida and Kaneda describe a model of shop around behaviour. Their simulations seem to be in agreement with survey data. Some policy making decisions about flow lines and parking lots facilities are examined. Femi Opadiji and Kaihara apply an agent based model to production scheduling following the market oriented programming by Wellman. Auctions are socially inspired methods for solving scarcity and choice problems and this papers is a good example. It seems that they are not aware of the stigmergic modelling approach to planning problems (e.g., Parunak). Wöckl describes a simulation approach that combines traditional artificial consumer markets (not ABM) with cellular automata to discover patterns of behaviour in terms of the product space, the life cycle product and advertising. The efficiency of advertising can be seriously influenced by word of mouth communication.
The "Agent-Based Approaches to Social Systems" section gathers four papers. Sato et al. propose a method to improve conjoint analysis by using information technology, but is not an ABM model. ABM in not even among the paper keywords. Ikeda et al. use a model with two layers: a learning simulator and a traffic simulator to investigate the mergence of social norms. Using GA for agents learning, they show that social norms emerge for different arrangements of the traffic signal system. Kurahashi investigates undiscovered historical facts in the dynamics of a successful Chinese family from data observed in the last 500 years. Data refer to the effect of the relationship between father, grandfather and great-grandfather in the successful civil service examinations. He uses inverse simulation based on a cohort fitness function of cultural capital degree distance. Yamamoto et al. analyze the effect of reputation management (RM) in promoting cooperative behaviour in a virtual C2C experiment run with 37 students at Tokyo university. The C2C was build up in the prisoners dilemma framework. The simulations show that RM induces cooperation when there are few new entrants in the experiment.
The "Market and Economy I" section includes 3 papers. Two of them deal with modelling investors sentiment (MIS). Tay creates a model to discover how network characteristics such as richness of the information environment, tendency of investors to extrapolate past data and social influence affect the evolution of a measure of the investor sentiment within the network. The emerging signatures are in agreement with conventionally accepted sentiment data. Yamada and Terano use an ABM with genetic learning to test if the model can capture investor sentiment. They compare the emerging time series with those of previous studies of MIS, based on statistical physics. The results are not fully conclusive, since price formation could result in different time series properties. Kunigami et al. investigate the emergence of money from a barter economy. Of course, being money a decreasing transaction cost instrument, the network model of the exchange definitively matters. The authors model the process of money emergence by mean field approximation and agent based simulation. They claim that, apart from the emerging results, their model allows to link social structure described by networks and dynamical analysis, by both differential equations and simulations.
In the "Market and Economy II" section there are four papers. Tseng et al. deal with a timely issue. Do agent-based prediction models generate hierarchical and scale free network structures? For a CDA with even zero intelligent agents, this happens to be the case. Even more, the scale free nature of the cash flows networks might rely on the institutional design and the structure of markets, rather than on trader's strategies. I completely agree with the first part of their conclusions: of course, institutions matter. But, the same can does not hold for the second part, since, as other colleagues and I have shown in published research, the strategic behaviour of the agents do matter as well, in terms of individual surplus. Yanagita and Onozaki use a computational model to show how monopoly or oligopoly can emerge from competition among the firms. They build a decentralized market populated by bounded rational adaptive agents and interacting with each other. The model is based on micro motives without the conventional artefact of a given demand, but the authors make strong assumptions about the consumers and firm behaviour. The results indicate that market dynamics, and in particular the industry life-cycle, can be explained by a β parameter that governs market share distribution. This is an interesting paper that deserves further investigations. Ohori and Takahashi study lead user innovation management in consumer product markets. They build an ABM supported by the CAMCaT framework (the acronym stands for Coevolutionary Agent-based Model for Consumers And Technologies) previously developed by the authors. They simulate the effects of lead users on conventional marketing and technology management strategies. The results are relevant for the marketing and technology mix. The authors are fully aware of the abstract level of the model, leaving for future research the analysis of specific markets with lead user innovation. The last paper is by Pospelov and Zhukova. They raise again the question of the role of monetary exchange versus direct barter. The question is of very low practical value for actual economic practice. At least, the paper by Kunigame mentioned above was intended to smooth the methodological gap between agent-based simulation and differential top down approaches. On the contrary, this last paper simply suggests a dynamic stochastic and top down model with no behavioural or micro motives bases. The result are almost self evident: the average utility of money depends on the proportion of money traders.
In conclusion, this book covers a wide range of ABM applications to the modelling of complex systems, both in economics and in management engineering. It is a good sample of the on-going ABM research that is being promoted by the Pacific-Asian Association for Agent-Based Approach in Social Systems Sciences (PAAA). Some of the papers are very relevant and timely. In spite of the heterogeneity of the contributions included, the book is undoubtedly worth reading. At the same time, the reader will notice the lack of attention paid to methodological issues, such as validation and model replication. On the contrary, my opinion is that these issues should be urgently put at the core of any ABM application to economics and social complex problems.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2009