Methods for Exploring Simulation Models

International audienceSimulation models are an absolute necessity in the human and social sciences, which can only very exceptionally use experimental science methods to construct their knowledge. Models enable the simulation of social processes by replacing the complex interplay of individual and collective actions and reactions with simpler mathematical or computer mechanisms, making it easier to understand the relationships between the causes and the consequences of these interactions and to make predictions. As the formalism of mathematical models offering analytical solutions is often not suitable for representing social complexity (Jensen, 2018), more and more agent-based computer models are being used. For a long time, the limited computing capacities of computers have hampered programming models that would take into account the interactions between large numbers of geographically located entities (persons or territories). In principle, these models should inform the conditions for the emergence of certain patterns defined at a macro-geographic level from the interactions occurring at a micro-geographic level, in systems whose behaviors are too complex to be understood directly by a human brain. Moreover, it is also necessary to analyze the dynamic behavior of these models with nonlinear feedback effects and verify that they produce plausible results at all stages of their simulation. This essential work of exploring the dynamics of modeled systems remained in its infancy until the late 2010s. Since then, algorithms combining more sophisticated methods, including genetic algorithms and the use of distributed intensive computing, have made it possible to make a significant qualitative leap forward in the exploration and validation of models. The result is an epistemological turn for the human and social sciences, as indicated by the latest applications realized with the help of the OpenMOLE platform presented here