Order this book
Department of Geography, University of Liverpool, UK.
This book presents the proceedings of the second International Conference on Computer Simulations and the Social Sciences, held in Paris in September 2000, and comprises 30 chapters (papers) summarising the work of over 50 contributors. As is the nature of such volumes, both the content and quality of individual chapters is highly variable, ranging from sketchy statements of research intent through to substantive findings of potential interest to many. Less excusable, perhaps, is the rather uneven editing. Although there is a clear house style, a significant number of chapters deviate from it in some way: non-standard citation and bibliographies, abstracts only in English or not at all, differing typeface and font size in two of the chapters. But these are minor gripes. In volumes of this sort, content is everything.
The book is divided into five, necessarily rather arbitrary, sections of varying size: Sociology, Geography and Urbanism, Politics and History, Economics: Individual Decisions and Games, Economics: Heterogeneity and Market Behaviour, Economics: Applied Market Models, Applications to Management. Below I present brief summaries of each section. At the end of the review I conclude with a few thoughts concerning the book's overall strengths and weaknesses.
This first section of the book contains six chapters. The first, by Rosaria Conte, argues against Axelrod's call for agents to be kept simple, calling instead for agents with greater mental complexity, in particular to allow better representation of agent interactions with their social environment. Mario Paolucci follows with a well-written account of a gossip model, in which the interactions between information spread and information accuracy are explored. False reputations for honesty and dishonesty are spread via two mechanisms: sharing of (possibly biased) personal experience in dealings with target of gossip and via mis-transmission of information. Paolucci correctly highlights the importance of model implementation in determining model outcomes and notes in particular the shortcomings of having so far omitted feedback effects (living up to type once tarred with a difficult to discard poor reputation). In an equally interesting paper, Francesco Billari advances a case for evaluating agent mental models by comparing their outcomes to empirical data on human behaviour. This argument is developed using the evaluation and refinement of a classic mating-search (The Next Best rule) proposed by Todd. Billari shows that, of itself, TNB produces skewed (monotonically decreasing) rather than demographically observed bell-curve shaped distributions of mating-times. A refinement in which heterogeneous rather than homogenous search times are allowed reproduces the required bell-curve. Andrzej Nowak et al. present an updated version of Sakoda's model of social interaction, in which the compatibility of two individuals is modelled as a dynamic state via a coupled logistic mapping. As the authors note, the work as presented is only at a preliminary stage, and has yet to be convincingly linked to actual human behaviours. Next, Guillaume Deffuant et al. present two variants of a majority model, in which opinions are not dichotomised, but rather represented as a continuous variable, and one in which individuals hold 'vectors' of opinions. In both cases the emergence of clusters of individuals with similar opinions is observed, at least for some parameter regimes. Last in the section, Alexander Laptev briefly presents a complex mathematical model of sociogenesis (the global evolution of society), producing a stable periodic solution. As Laptev notes, without any empirical grounding this model is currently of 'academic interest only'.
Four chapters in this section change the focus from the social to the spatial. Aschan-Leygonie et al. present a brief overview of a well-developed spatial microsimulation model of population dynamics in Southern France. Population growth is modelled at the level of the commune, with migration being mainly driven by the labour market. Initial qualitative validation led to the introduction of commune level constraints concerning local rates of job growth the impacts of overcrowding on commune attractiveness. Yet to be implemented, the authors intriguingly propose the implementation of person-based geographical location preferences that perhaps move the model more towards an agent-based approach. Also addressing the problem of migration, Diane Vanbergue et al. outline a multi-agent cellular automata based model of intra-urban migration flows in an 'artificial' Bogota. However, despite notable sophistication in some areas, the model as presented can be viewed at best as a proof of concept, rather arbitrarily disallowing individual (as opposed to household) migration and lacking any formal empirical validation. Turning to commuting flows, Matteo Bellomo and Sylvie Occelli have constructed a model, implemented in SWARM, of commuting flows that treats accessibility as a function of individual 'action spaces', the attributes of localities (e.g. congestion and the interaction between individuals and localities and individuals and other individuals.) As outlined, the model is highly preliminary, but offers considerable scope for future refinement. Finally, Jean-Luc Bonefoy et al. revisit (implicitly) Guy Hardin's 'Tragedy of the Commons', using a multi-agent Cellular Automata model of shepherding in an initially forested catchment area. Interactions between shepherds help to build up a communal picture of forest degradation, whilst individual compliance to communally agreed responses maintains complexity over time. The authors claim that this approach allows them to simulate a region as 'a dialectic ... between individuals, spaces and society'.
In this section of the book, six chapters shift the focus from spatial to power relations. First, Serge Galam applies statistical physics to political science to introduce an interesting hierarchical variant of the majority model. In essence local cells in one layer of a CA vote for a single representative in the layer above. The results of the model are tentatively offered as a potential explanation for the rapid collapse of eastern European communist parties at the end of the last century. Next, Dominique Lepelley et al. use Monte Carlo simulation to explore and quantify some of the potentially paradoxical outcomes (Condercet winners and losers) of alternative voting systems, whilst Alain Albert and Wolfgang Balzer examine production, predation and protection from a game-theoretic perspective, concluding that their initial model lacks empirical validity but identifying potential areas for future development. In contrast, Niels Lepperhoff explores aspects of rational choice theory in the multi-agent model Dreamscape. One conclusion is that rational choice theory fails when agents are allowed to kill each other. Lepperhoff suggests that the solution might lie in the role of institutions. Also addressing power relations, José Caldas and Helder Coelho model the evolution of a dependence network in a simple multi-agent model designed to start exploring the modelling of power relations. Their initial results suggestively replicate real world observations. Finally, and rather out of place in this section, Frank Beckenbach revisits the Epstein and Axtell Sugarscape trading model, introducing a number of modifications designed to overcome shortcomings introduced by constraints on agent behaviour in the original. The revised model itself is acknowledged to have theoretical shortcomings that need addressing.
The first of three economics-related sections presents three papers on aspects of individual decision and game theory. Pietro Terna presents a short but thought-provoking exploration of the debate addressed by Conte in the opening chapter of the book; the extent to which agents should be kept simple or endowed with a full suite of Beliefs, Desires and Intentions; a debate that Terna neatly encapsulates as 'mind or no mind?'. Terna's evidence, based on a review of a series of stock market type models, is mixed. Mindless agents can produce highly complex market behaviour, but complex agents can reproduce theoretical scenarios. As a caution, Terna points out that external constraints can have more impact than any degree of tinkering with the minds of agents will. Next, Robert Savit et al. revisit the minority game, examining what happens when pay-offs to the minority group depend upon group size. Perhaps surprisingly, they find that the main results are unaffected by this change in the payoff function. This gives greater confidence in applying findings from this model to the natural world. Finally in this section, Zimmerman et al. consider co-operation in an adaptive network by revisiting a weak version of the Prisoner's Dilemma.
The second economics-related section presents four papers on aspects of heterogeneity and market behaviour. The first, more mathematical, paper by Sorin Solomon finds a novel application for the Generalised Lotka-Volterra model, using it to model wealth distribution. As applied, the model appears to successfully capture the empirically observed power-law distribution in the upper tail. Solomon also usefully describes how the GLV can be applied in a multi-agent system. In contrast, Marco Janssen and Wander Jager report the latest results from a multi-agent model of consumer behaviour, in which consumption is driven by a combination of personal need and peer pressure (as reflected in market penetration of products within social networks). Their experiments highlight the important role that consumer psychology may have on product success, and the extent to which such psychology can lead to extreme market volatility. With the firm as agent, Gérard Ballot and Erol Taymaz use the MOSES multi-agent model of the Swedish economy to examine the potential impacts of firms poaching and training staff. They conclude that poaching and training can coexist in the long run but, perhaps surprisingly, also conclude that poaching can have a bigger negative impact on the poacher than the poached. They attribute this to the increased staff costs incurred by the poachers, who have to pay inflated wages to attract the staff they are trying to poach. Finally, drawing on graph theory to represent knowledge as a structure, Nicolas Carayol describes a multi-agent model of knowledge diffusion. The main finding from the model is that optimal rates of knowledge disclosure are variable, but in all cases greater than zero.
The final economics-related section of the book contains five papers on applied market models. In the first of these, Leslie Rosenthal uses an agent-based model to mimic the UK house buying process, leading to results that offer a simple explanation for the observed lag time between increases in purchase prices paid by first time and existing home owners. In the second paper, Takuya Iwamura and Yoshiyasu Takefuji describe their preliminary work on creating an agent-based stock market model, extended to include traders' attitudes to risk. This is followed by Jean-Michel Dalle and Nicolas Jullien's report on the application of an agent-based model to explore the dynamic interaction between two alternative marketing strategies. They conclude that there are plausible conditions under which an established proprietary market leader (Microsoft Windows) could be displaced by a competitor (Linux) distributed using a free software strategy. Staying with the theme of market share, Mariana Mazzucato investigates the impact of firm size and innovation on market share instability, highlighting the roles of negative feedback and one-off shocks. Deterministic and stochastic versions of a model are used to separate out the influence of structural and random factors. Finally, Georg Muller proposes and then exemplifies an interesting and novel approach to model validation, in which models are tested for robustness by seeking to produce counter-factual or counter-theoretical results through systematic changes in model parameters.
The final section of the book contains only one paper, which contrasts sharply with most of the papers that have gone before in two ways. First, the author, Eric Bonabeau, is based in commerce, not academia. Second, the paper provides brief examples of the successful commercial application of agent-based simulation to real-world problems: theme park labour scheduling; stock market regulation and operational risk management in financial institutions. In all three examples agent-based simulation is favoured over direct statistical/mathematical analysis due to non-linear and high-dimensional system behaviour.
The stated purpose of this volume is to "illustrate how computer simulations can help us to understand social phenomena". Given the highly preliminary and condensed nature of much of the work presented, this is perhaps over-ambitious. For such illustrations I would refer you instead to books such as that by Gilbert and Troitzsch (1999), and to journals such as JASSS. The real purpose of this volume is to provide an indication of current areas of research activity. That a web site of conference proceedings would more usefully, more cheaply and more speedily provide the same service to a far larger audience is perhaps a matter for consideration by future conference organisers.
GILBERT N. and K. G. Troitzsch 1999. Simulation for the Social Scientist. Open University Press, Buckingham. [JASSS review]
Return to Contents of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2002