Exploration methods for simulation models
We first recall in this chapter to what extent simulation models are an absolute necessity in social sciences and humanities, which can only very exceptionally require to experimental sciences methods to construct their knowledge. The formalisation through mathematical models able to offer analytical solutions being most often not possible in order to provide satisfying representations of social complexity, computational models based on agents are more and more used. For long the limited computational capabilities of computer have forbidden to program models taking into account interactions between large numbers of entities geographically localized (individuals or territories). In principle these models should inform on the possibilities and conditions of the emergence of given configurations defined at a macro-geographical level from interactions occurring at a micro-geographical level, within systems with a too much complex behavior to be understood by a human brain. This however requires to study the dynamical behavior of these models including non-linear feedback effects and verify they produce plausible results at all stages of their simulation. This necessary stage of the exploration of the dynamics of algorithms remained rather rudimentary until the end of the last decade, when algorithms including more sophisticated methods such as evolutionary computation and the use of distributed high performance computing have allowed a significant qualitative leap forward in the validation of models, and even an epistemological turn for social sciences and humanities, as suggest the latest applications realized with the OpenMOLE platform described here.
READ FULL TEXT