Reviewed by
Magda Fontana
Department of Economics, University of Turin
As a new approach develops and spreads, the need for sound methodological grounds and for a common language becomes more stringent and, at last, unavoidable. The fil rouge of the papers gathered in this book is the awareness that agent-based modelling and simulations are "at the end of the beginning and therefore entering a new phase characterized by the conjunction of accumulated knowledge" (p. 2). Although the book is not divided in parts (a choice that would have helped the readers), it is possible to organise the papers along different threads. Each thread investigates an articulation of the state of the art and shows that convincing responses can be given to the criticisms raised to Multi Agents Systems and Modelling. The book provides evidence of the significant progress that multi-agent modellers have made in the last decade in "understanding their creations" - to paraphrase the title of the 1994's paper by Axtell and Epstein (Axtell and Epstein 1994)-, in the refinement of the concept of agent, in the definition of the slippery concept of emergent properties, in the assessment of the performance of models and in their comparability. All of these questions are treated in the book.
For the sake of simplicity, I will gather the papers along three intertwined lines:
i) methodological and epistemological issues ("Multi Agent Concepts and Methodologies", by Farber; "Introduction to Discrete Event Modelling and Simulation" by Ramat; "Computational Social Science: Agent-based Social Simulation" by Gilbert; "The End of the Beginning for Multi-Agent Systems Social Science" by Axtell; "Towards an Epistemology of Multi-Agent Simulation in Social Sciences" by Livet; "Emergence in Multi-agent Systems: Conceptual and methodological Issues" by Dessalles, Muller and Phan; "Epistemology in a Nutshell: Theory, Model, Simulation and Experiment" by Phan, Schmid and Varenne; "Philosophy of Social Science in a Nutshell: from discourse to Model and Experiment" by Dubois and Phan);
ii) assessment and validation of models ("Exploring Models by Simulation: Application to Sensitivity Analysis" by Ginot and Monod; "Assessment and Validation of Multi-agent Models" by Amblard, Bommel and Rouchier; "Comparison of Three Implementations of Schelling's Spatial Segregation Model" by Daude and Langlois; "An Introduction to UML for Modelling in the Human and Social Sciences" by Bommel and Müller);
iii) applications (to complexity theory) ("From Networks of Automata to Agent-based Models" by Phan); (to field analysis) ("Modelling with and For Stakeholders" by Ferrand); (to urban geography) ("Agent Models in Urban Geography" by Sanders), (to agent-based modelling ) ("Modelling, Implementing and Exploring Agent-based Models. An Example" by Daniel).
I would like to emphasize some (of the many) interesting statements. From the book it appears how far MAS modelling stands from its incipit in exploring the behaviour of models. In "The Complexity of Cooperation" (Axelrod 1997, 184), Axelrod claimed that without a process of comparison of models "computational modelling would have never provided the clear sense of domain of validity that typically can be obtained for mathematized theories" and set forth the lack of a literature on sensitivity analysis. In the same vein, Epstein and Axtell's "Growing Artificial Societies" (Epstein and Axtell 1996) showed that the very concept of validation was still unclear at that time. Ginot-Monod show the relevance of sensitivity analyses clearly defines them with respect to other methods of exploring models and give detailed examples. Amblard-Bommel-Rouchier afford the very critical topic of validation of multi-agents models by answering convincingly to the core question on what is validation and discuss what forms it may take. Validation is not solely similarity with observed data: since MASs pertain to the realm of models constructed to "grow" a phenomenon in silico, researchers must test for validity of their own model in a deeper and wider way. The authors stress that validation - and the adequate techniques to achieve it - are associated with the use of the model (i.e. to predict, to understand, to act). The concepts of internal validation (which subsumes sensitivity analyses and identification) and external validation are explained and discussed at length.
This rather technical set of papers is matched with other works that relate to methodology. Let me return to the beginnings of MAS modelling when Nigel Gilbert and a team of researchers at the Santa Fe Institute were independently developing the approach. A problematic discussion regarded those properties surprisingly emerging from seemingly simple models. In "Growing artificial societies" (Epstein and Axtell 1996, 52), for instance, the concept of emergence is admittedly fuzzy: it loosely relates unexpected behaviour and minimal requirements in terms of rules to the appearance the macro phenomenon of interest. The paper by Dessalles-Muller-Phan reviews some literature on the topic focusing on the methodological implications (e.g. emergence as encompassing both holism and the reductionist version of individualism) and then provides - through the analysis of two models - the notions of formal and quantitative emergence as tools to investigate the links between macro patterns and the cognitive dimension of the agents. The paper by Livet extensively discusses the epistemology of simulation in social science and the possibility of extending it to all the uses of simulation and to other sciences.
The book as a whole offers many remarkable insights that can be fruitful for both beginners and experienced practitioners. However, some promises made in the introduction are not entirely fulfilled. The idea of introducing two appendixes entirely devoted to epistemological issues and to the philosophy of science is praiseworthy; however, few references are made to MAS directly and most of the space is devoted to an effective (but a bit too long, especially in the case of appendix 1) reconstruction of the history of epistemology since the 14th century and to a survey of the main themes of the philosophy of science which may not capture the interest of the reader.
ReferencesAXTELL R and EPSTEIN JM (1994). Agent-based Modelling: Understanding Our Creations, The Bulletin of the Santa Fe Institute: 28-32.
EPSTEIN JM and AXTELL R (1996). Growing Artificial Societies - Social Science from the Bottom Up. Cambridge MA: MIT Press.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2008