Agent-based modelling: what matters is action

Computer-aided management tools or models of production systems in the manufacturing or agricultural domains generally rely implicitly on the theory of planned action. Every action is assumed to be part of an anticipated sequence leading from a current state to a predetermined goal. The main limits of this classical view are due to the difficulty to deal with unexpected changes and disturbances. To overcome these limits, we developed an agent model based on the theory of situated action. Whereas the classical approach puts the emphasis on actor’s decision (action being assumed to straightforwardly follow), situated action is viewed as a process endowed with a temporal thickness, spontaneously emerging from the situations created by the local interactions between the actor and its environment. This model, accounting for both the temporal and spatial dimensions of action as well as its contingent features, implements the concepts of affordance (capacity of objects to trigger actions) and stigmergy (self-organization mediated by marks left by individuals in their environment). Therefore we propose a multi-agents system where the perspective is reversed compared with the usual view: in our model it is the environment which is agentified and, so, endowed with the capacity of acting by handling the entities it contains. Unlike in classical multi-agents systems, these entities (standing for humans, animals…) are, actually, considered as non-autonomous and passive. After advocating our choice to put the emphasis on action rather than on decision to represent actual human activity, we explain the concepts of affordance and stigmergy and outline the non-classical multi-agents system we devised with the perspective to simulate agricultural production systems