A modelling language to represent and specify emerging structures in agent-based model.
Résumé
All modellers have come across, one day, one of these popular toy agent-based models (ABMs), like "Ants", for instance, which depicts the appearance of pheromone trails built by simulated ants. They are simple, but representative of the way "real", more complex, ABMs are designed: in addition to explicitly describe the individual entities used to represent the system, modellers make implicit references to abstractions corresponding to the emerging structures they are tracking in the simulations. Yet, these abstractions are not represented in the models themselves as first-class entities: they are either hidden in ex-post computations or only part of visualization tasks, as if an explicit representation could somehow damage the processes at work in their emergence. This clearly constitutes an obstacle to the development of multi-level models, where emergence is likely to occur at different levels of abstraction of the system: if some of these levels are not represented in the models, the emergence of higher-level structures is not likely to be observed. This paper describes a modelling language that allows a modeller to represent and specify emerging structures in agent-based models. Firstly, to ease the description, we present these structures and their properties in four toy ABMs: Schelling, Boids, Collective Sort and Ants. Then we define the operations that are needed to represent and specify them without sacrificing the properties of the original model. An implementation of these operations in the GAML modelling language (part of the GAMA agent-based platform) is then presented. Finally, two simulations of the Boids model are used to illustrate the expressivity of this language and the multiple advantages it brings in terms of analysis, visualization and modeling of multi-level ABMs.