Abstract : The human activity has consequences on the evolution of the society and of the environment. We are interested in the characteristics of the livestock farming as a human activity system in the North of Uruguay, in the area of Arapey. To carry out the study we use a systemic approach, where narrative analysis and computer simulations are combined by using multi-agent systems. The principal characteristics of the agricultural activity are described, by including the evolutions on the long term. We show thus how agriculture and the livestock breeding fostered the development of hierarchical societies and how this process accelerated with time. In particular, we show the interactions between society, science and agriculture, explaining how during the last century, an enormous increase in productivity was obtained by the industrialization of agriculture, thanks to the use of fossil fuels. The Uruguayan livestock production is in permanent interaction with what occurs abroad. The food globalization was born at the same time as the Uruguayan meat was distributed everywhere in the world, throughout the second half of the 19th century. The construction of the Arapey model includes three stages: i) identification of "significant" practices by means of a combination of, research and field observations ii) modeling of the system at the farm level, representing practices chosen to study, and by using the "Unified Modeling Language" (UML) and iii) evaluation of the consequences of these practices on the evolution of the exploitations and of the region, by using the Cormas platform to carry out simulations on a model called Arapey. Vis-a-vis with situations of great uncertainty, an improvement of the collective action can be obtained by the discussion of the actions to be taken. It must be based on a great participation of the stakeholders, and on the use of instruments making it possible at the same time to integrate knowledge of various origins and to explore different scenarios by representing various points of view, such as the multi-agent systems and/or a family of diagrams. A thorough knowledge of the situation is a necessary condition for this action can be salient, credible and legitimate. The MAS as "a virtual laboratory" can accelerate learning by incorporating qualitative variables, which makes the model more realistic and improving its capacity of being an effective support for decision making.