The Curse of Possibilities

This paper argues that the distinction between what constitutes the parameters of an agent-based model and what constitute its structure is not as clear as we assume it to be. In some fundamental sense, the parameter space of any model is the space of all possible models. Modellers often rely on their experience and intuition to make structural modelling choices. Arbitrary choices are inevitable given the pragmatic constraints of the research process but their consequences are significant and we should be more explicit about them. Being formally explicit about which parts of the model structure could be changed makes it possible to automate the search for agent behaviours just like we automate the search for scalar parameter values. Work currently being done under the “inverse generative social science” label is pioneering this approach, but improvement in our modelling tools could open it up to the wider ABM community.