IRIT - UMR 5505

SocLab, for Social Laboratory, is an environment dedicated to the study of social organizations. As a virtual laboratory, it allows its user to build and manipulate models of organizations in silico and to test hypotheses on objects for which the experiment in vivo is impossible. SocLab is based on the Sociology of the Organized Action (SOA) (M. Crozier, E. Friedberg), a well-established theory in the domain of the sociology of organizations, and it implements a formalization of the SOA. The SOA is a theory that attempts to discover the functioning of an organization and to explain the motives of the actual behaviors of the organization’s social actors, behaviors that, often, diverge from the ones specified by the formal rules of the organization: social actors seldom behave as they are expected to do. The SOA is widely used as the foundation of the methodologies applied by most consultants when they are called to analyze the very origin of dysfunctions within social organizations. It appears that, thanks to the generality of the SOA’s principles, their application is not limited to social organizations but can be extended to societies of software entities, especially Multi-Agents Systems. SocLab is a MAS that enables its users to build models of social organizations or of MAS’s, to make analytical analyzsises and to simulate the behavior of their actors. These models may describe existing systems; they may also describe envisioned systems, either in the context of organizational changesor at the design step of a MAS. The features provided by SocLab can be summarized as shown in this figure:

• An organization is modeled using the Editor module. This module allows creating, editing, copying, saving (as a xml database) and annotating an organization using an ergonomic interface.

• The static analysis module is used to calculate all organization’s remarkable states, to guide the modeling, and to analyze the structure of an organization using a graphical interface of tables, diagrams and networks.

• According to the SOA theory, the behavior of actors is strategic, framed by a bounded rationality, and geared towards cooperation. We modeled the actors’ rationality with a process of trial and error based on a self-learning rules system. According to this modelization, the simulator interface allows its user to initialize all simulation’s parameters (such as number of runs, actors’ parameters...), then to simulate the organization and know how it can be regularized.

• The sensitivity analysis module allows studying the correlation of an input (such as an actor’s parameter) with the simulation results. The synthesis modules present the results of simulations or sensitivity analysis in a graphical mode using curves and diagrams, and the development of facilities for statistical data analysis (PCA, multiple correlations...) is in progress.