Reviewed by
Erol Taymaz
Middle East Technical University
Jean-Philippe Rennard has achieved a tremendous task of editing 58 papers written by 103 researchers in two volumes of the Handbook of Research on Nature-Inspired Computing for Economics and Management. Nature-inspired computing, that is almost synonymous with evolutionary computing, has become an influential research methodology in the last couple of decades thanks to enormous increase in computing power. The Handbook by Rennard provides a stocktaking of the state-of-the-art in the field. The papers edited in the Handbook are grouped in nine topical sections (the numbers of papers are in parentheses): Nature-Inspired Computing (6), Social Modeling (7), Economics (13), Design and Manufacturing (5), Operations and Supply Chain Management (12), Information Systems (4), Commerce and Negotiation (4), Marketing (3), and Finance (4). The Foreword by Eric Bonabeau offers the justification for the Handbook: "The foundations of economics have for so many years ignored the realities of human behavior and decision making that it has become a joke for business practitioners, managers, and consultants". However, a new breed of economists is now emerging. These new economists and social scientists are employing new techniques, such as "evolutionary algorithms, co-evolution, swarm intelligence, social networks, multi-objective decision making, and agent-base modeling", to understand, model, approximate, or even enhance human behaviour. The Handbook provides many examples of new approaches and artificial-life techniques that have been developed and used to analyze real-life problems.
Although the papers are grouped by their application areas in the Handbook, I would like to classify them by their content, for this review, into three categories: review essays, theory and methodology, and applications.
The review essay by Collet and Rennard (Ch. III) is focused on stochastic optimization algorithms, and they discuss very briefly the main characteristics, and pros and cons of various stochastic/evolutionary optimization algorithms (random search, hill-climbing, simulated annealing, tabu search, neural networks, evolutionary algorithms and genetic programming, data-level parallelism, particle swarm optimization, and ant colony optimization). The next chapter by Collet gives the flavour of evolutionary algorithms, and discusses the main aspects and operators of evolutionary algorithms, like representation of individuals, definition of fitness functions, variation operators (crossover and mutation), and selection and replacement operators. Verhagen (Ch. VIII) and Bruun (Ch. XIV) present a brief introduction to agent-based modelling in sociology and economics, respectively, whereas Di Marzo Serugendo (Ch. XXIX) describes the main characteristics of autonomous systems (decentralized control, self-organization, emergent behaviour, etc.) with a special reference to engineering applications.
Some of the chapters in the Handbook review specific methods or techniques frequently used by evolutionary researchers. These include multi-cellular techniques by Anderson (Ch. II), genetic programming by Collet (Ch. V), and multi-objective optimization by Coello Coello (Ch. VI). Although the title of the last chapter is "Evolutionary Multi-Objective Optimization in Finance", its content is much broader than the title implies, and it includes only a very short section on "some applications in finance".
The review essays in the Handbook provide an easy access for readers to the complex world of artificial societies and evolutionary computing. There are some unavoidable overlaps between these chapters, and the differences in terminology may sometimes confuse the novice readers. Thus, one may think that a meta-survey with a concise taxonomy of methods/techniques used in the field would have been very useful.
Specific methodological issues are discussed by Al-Dabass (Ch. XII on the use of hybrid nets for parameter knowledge mining), Abraham and Machwe (Ch. XXVII on human-centric evolutionary systems), Fidanova (Ch. XXXIII on a comparison between different algorithms used in ant colony optimization), and Gosling, Jin and Tsang (Ch. XXXVIII on strategy generation by evolutionary computation).
The dynamics of firms and industries has received considerable attention of evolutionary economists. The Handbook provides a rich set of papers on this crucial issue: For example, Pyka (Ch. XVI) criticizes the quantitative orientation of mainstream economics, and suggests that "agent-based models can cope with the challenges of an evolutionary setting". He then presents his model of industry dynamics to analyze qualitative change in economic systems. Kwasnicka and Kwasnicki (Ch. XX) have developed an evolutionary model of industrial dynamics to study how different market structures (monopoly, oligopoly, etc.) emerge endogenously. The focus of the paper by Saam and Kerber (Ch. XXIV) is the process of innovation and knowledge accumulation, as is the case in many evolutionary studies, but they also analyze the effects of limited imitability, that can hamper knowledge accumulation and lead to severe lock-in. Lavigne and Sanchez (Ch. LVIII) follow in the same direction, and study the effects of the availability of information on the dynamics of stock markets. Barr and Saraceno (Ch. XIX) use a general neural network model of information processing activities within the firm, and study the competition process in oligopolistic markets.
There are four papers on industrial district/clusters that need to be read together, because they explore different aspects of formation and development of industrial clusters, so that each paper complements others. Merlone and Terna (Ch. XXI) examine the effects of cooperation between social agents (workers and firms) on the formation of industrial districts, whereas Berro and Leroux (Ch. XXIII) are focused on the role of strategic interactions between firms and local authorities. Albino, Carbonara and Giannoccaro (Ch. XXII) investigate the determinants of the competitiveness of geographical clusters and, similarly, Dawid and Wersching (Ch. XXV) analyze the relations between technological specialization, profitability, and cluster formation.
A group of researchers has analyzed the individual behaviour itself. Nooteboom (Ch. X) studies the conditions under which trust and loyalty are viable in markets. Valée (Ch. XVIII) adopts genetic algorithms to model inflation learning. Brabazon, Silva, de Sousa, Matthews, O'Neil and Costa (Ch. XXVI) investigate in detail the effects of various heuristics on the process of product innovation, whereas Erez, Moldovan and Solomon (Ch. LIII) model the diffusion of new products in order to understand the role of "negative word of mouth" as a factor that inhibits the diffusion process.
Three papers specifically deal with the use of evolutionary computing in negotiations and auctions (Klein, Faratin, Sayama and Bar-Yam, Ch. XLVIII; Debenham, XLIX; Mochón, Sáez, Quintana and Isasi; Ch. L). However, the models presented in these papers do not address specific negotiations or actions, but provide a general framework in using evolutionary computing in negotiations and auctions. In that sense, these papers use evolutionary computing more for understanding than for solving real-life problems.
The papers by Rochet and Baron (Ch. XXX), Chakraborti (Ch. XXXI), Minis and Ampazis (Ch. XXXIX), Siebers (Ch. XLIII), and Cardosa, Rocha and Oliveira (Ch. LI) provide frameworks or guidelines in using evolutionary computing in various fields of management, manufacturing and production planning.
Finally, there are five papers that discuss in detail how researchers can use specific software tools in applying evolutionary computing. Medaglia and Gutiérrez (Ch. XL and Ch. XLI) describe the Java Genetic Algorithm (JGM) as a computational object-oriented software tool for solving complex optimization problems by evolutionary algorithms, and illustrate its use in specific cases. Medaglia, Gutiérrez and Villegas (Ch. XLII) then describe the Multi-Objective Java Genetic Algorithm (MO-JGM) as a tool to develop multi-objective evolutionary algorithms.
As this brief review indicates, this Handbook of Research on Nature-Inspired Computing for Economics and Management covers a wide range of issues and fields. Its coverage makes this Handbook a valuable addition to the existing literature. Although it is published in two volumes, there is a common subject index, published in both volumes, that makes it easier to search the Handbook, and also to find weak points! For example, there are only three entries for "learning", a central issue in evolutionary studies, and no entry for "classifier systems".
The extensive coverage and a large number of papers associated with it come with problems as well. There are some (unavoidable) overlaps and repetitions. But, more importantly, most of the papers are too short, presumably because of space constraint. This creates problems for those papers that could only present the overall structure of their simulations models without any detail.
This Handbook provides many answers, but it also leaves some key questions open. For example, when do we really need simulations? Should we adopt agent-based simulations even if a mathematical model can be analytically solved? How should we analyze or control stochastic disturbances and the effects of parameters whose values are guesstimated by model builders? The simulation papers in this Handbook usually present results for only one simulation run, with the notable exceptions of Chakrabarti (Ch. IX), Saam and Kerber (Ch. XXIV), and Brabazon et al. (Ch. XXVI) who simulate their models many times (30 or 80), and present either mean values, or estimate the effects of parameters by regression analysis. What is the best (or simply "good") practice regarding stochastic factors and parameter values? Similarly, the issue of calibration deserves more attention, especially given the thought-provoking cases presented by Naveh and Sun (Ch. XI) and Marks, Midgley and Cooper (Ch. LII) on comparing simulation results with the real data. This is also related to the problem of generalization in evolutionary studies. How could one show that the findings are generalizable? For example, Kaboudan (Ch. LV) finds that genetic programming performs better than weighted least squares in forecasting of neighbourhood median residential housing prices. However, it is not clear if this result is data-dependent or not. A better approach would be to employ Monte Carlo methods for a sample of models.
Upon reading the Handbook, one gets the feeling of striking richness (or diversity) in the concepts, methodologies, approaches, and presentation forms in evolutionary computing, and understands very well that the development of new concepts, theories and methodologies is also an evolutionary process.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2008