Thinking the Social dimension of the Artificial World: Ontological status of Collective Beliefs Denis Phan 1 Jacques Ferber 2 1 Doctor in economics, Administrator P&T, CNRS senior researcher, GEMAS, UMR 8598 CNRS / University of Paris IV Sorbonne & CREM, UMR 6211 CNRS / University of Rennes I http://perso.univ-rennes1.fr/denis.phan/ Email: denis.phan@univ-rennes1.fr 2 Professor of Computer Sciences, University of Montpellier III LIRMM, UMR 5506 CNRS / University of Montpellier III http://www.lirmm.fr/~ferber/ Email: ferberlirmm.fr Abstract This paper deals with the modeling of the social dimension of artificial societies with cognitive agents. We provide a critical review of the ontological status of collective beliefs in some recent works in game theory, using a heuristic diagrammatic framework, useful for MAS design, called the “4Quadrant”. We introduce first a formal presentation of the problem of the “collective” versus “social” dimension of the beliefs, which states the problem more clearly, in particular from the point of view of software design ontology. This raises the problems of the existence of some “social entities” to be able to acquire some relative autonomy with respect to the agents. The “4Quadrant” allows us to discuss, for a given world / system of the ontological commitment and the articulation between both the individual and the collective on the one hand, and between the observable (reifiable) and the subjective on the other hand The last section provides an empirical counterpart, by reviewing and discussing some experiments from behavioral game theory. Keywords: artificial societies, behavioral game theory, cognitive agents, collective belief, coordination games, complex adaptive systems, intelligent agent’s ontology, emergence, experimental economics, learning in games, multi agent systems, population game theory, social ontology. 1. Introduction 1.1 Collective beliefs or social beliefs? In his recent work Orléan (2004) proposes a new definition of what a “collective belief” is. From our point of view, this initiative is of particular interest for two reasons. Firstly, it proposes a formal presentation (logical) of the “social” dimension of the beliefs, which states the problem more clearly, in particular from the point of view of software design ontology. Secondly, it raises the problem of the existence of some “social entities” (or “social objects”: here the “social beliefs”) to be able to acquire some relative autonomy with respect to the mind of individual agents of which they are the necessary support. This question is of great importance for the modeling of artificial society, in particular for the so-called “immergence” (Gilbert, 1995; Dessalles, Phan, 2005; Dessalles, Ferber, Phan, 2007) The existence of such abstract entities in game theory (and more largely in economics and social sciences) is far from being unanimously established. In particular, according to the methodological individualism approach, standard game theory generally does not use such “social” entities. The explicit objective of the quoted work by Orléan is to promote the interest of such social concepts within the framework of “cognitive economics” (Bourgine and Nadal, 2004, Walliser, 2004b). The ontological status, or mode of existence, of such entities is discussed among philosophers (among those who seek ontological bases with the social entities, one will quote: Gilbert 1987, 1989, 1990, 1996 1997, 2000; Pettit, 1993, 2000, 2004; Searle, 1995; Tuomela, 1990, 1992 without forgetting Descombes, 1996, on a related topic) and among sociologists as well. In this latter field, Boudon, (1998, 2003, 2006) contests the possibility of the existence of extra individual entities, but according to (Durkeim, 1895), there is an important tradition in which “social facts” must be considered objectively “like things”. See also Douglas, (1986) for a reinterpretation and (Livet, Ogien, 2000) for a synthesis. Our aim is to propose a new view of the modeling of social phenomena using multi-agent systems. This involves the reification of several concepts and thus imposes an ontological commitment, i.e. an analysis about the ontological status of the “social facts”. In order to examine the ontological status of these collective beliefs, we will