47

Agent-based systems have been in use since the 1990s to simulate the dynamics of social phenomena. They help to conceptualize interactions between an organized set of agents and to explore the structures that emerge through the formalisation of complex situations where heterogeneous agents operate at various levels. This article presents and discusses a conceptual framework which articulates the relevant fields (empirical, conceptual, model) in a general way, and in the case of multi-agent systems. It then develops an analytical framework to capture the various uses of these systems, exploring in particular the methodological opposition between parsimonious models (KISS) and descriptive models (KIDS), which could be viewed as different stages in scientific investigation. Finally, it uses this framework and criteria to analyse several examples from sociology, economics and geography (residential segregation, educational inequality, the emergence of norms, market operation, urban hierarchy) and concludes with methodological proposals based on an abductive approach to the use of multi-agent systems in social science.