Exploring Social Phenomena with Complex Systems Tools

This thesis explores the problems raised by the aggregation of entities into a global,collective level, an old problem encountered in many fields of science. We work on threeprojects related to the aggregation problem in social systems, using tools derived fromstatistical physics, and more generally quantitative tools. Great care is taken to tacklethe questions at the heart of these projects in a mutually beneficial way for both socialand natural sciences.The first project focus on a paradigmatic model of the emergence of puzzling macro-scopic behavior from simple individual rules, Schelling’s segregation model. This modelsimulates the evolution of the spatial repartition of two types of agents living in a virtualcity. It is widely known for this paradoxical effect: if the agents have a mild preferencefor one’s neighbors to be of the same kind, their move lead to segregative pattern at theglobal scale, even if total segregation does not maximize the collective utility. We firstuse simulations to show that introducing small amount of coordination in the agents’moving decision can significantly reduce segregation. We then propose an analytical resolution of Schelling’s model for a wide range of utility functions. Using evolutionarygame theory, we provide existence conditions for a potential function which character-izes the global configuration of the city and is maximized in the stationary states. Weuse this potential function to derive several analytical results. Switching on a physicistpoint of view, we generalize our potential function in a simplified version of the modelwhich interpolate between cooperative and individual dynamics.The second project is based on the exploration of huge databases on scientific liter-ature (mostly Web of Science) to investigate the existence and evolution of paradigmsor scientific institutions. We mostly use the old but quite unused bibliographic coupling(BC) approach, based on a normalized number of shared references, to measure relationsbetween articles. Using standard techniques to group similar articles, we can define ‘nat-ural’ communities characterized by their references. Thanks to a large database (141 098records) of relevant articles, we used this approach to empirically study the ‘complexsystems’ field. We show that the overall coherence of the field does not arise from auniversal theory but rather from computational techniques and fruitful adaptations ofthe idea of self organization to specific systems. We also investigate the idea of ‘trad-ing zones’, small communities creating an interface between disciplines around specifictools or concepts. We also apply our approach to develop a set of routines allowing todraw different maps of the research carried out in a scientific institution, specificallyco-occurrence (of authors, keywords, institutions) maps and BC communities maps. Wetel-00662484,version1-24Jan2012 use the example of the ENS de Lyon to discuss why these maps may become a valuabletool for institutions’ directors.Finally, the third project deals with the emergence of ‘institutions’ or ‘structures’ insocial systems. Our collaboration with a team of sociologists has lead us to question theassumption of a clear dichotomy between two ‘levels’, namely individuals and society.Building on the social theory developed by Gabriel Tarde at the end of the 19th century,we explore different possibility to visualize (and conceptualize) the evolution of socialphenomena without making a distinction between two levels. Bibliometric data are usedas an example. We also propose an attempt to formalize Tarde’s theory in the scope ofan algorithmic model. The point is to show how one can obtain ‘wholes’ through thesimplification of complex individuals. While our prototype model fulfil several of Tarde’sprecept, it raises many more questions. Finally, we focus on a single question raised byour colleagues sociologists: the existence of lasting structure from non lasting entities.We build on the physicist’s approach developed in opinion models. While most papersfocus on stationary properties, we choose to build a model to investigate the dynamical properties of social structures which are always changing. The key ingredients of ourmodel are the introduction of noise in the agents’ interactions, a turnover in the pop-ulation of agents and a generation effect, the agents taking into account their opinionand age difference in their interaction. The outcomes of our model display a rich phe-nomenology of group dynamics. Our point is not to produce realistic representations ofreality, but much more to help the sociologists enrich their conceptualizations of socialphenomena.