Artificial experiments in economics, with agent based models Pietro Terna Dipartimento di Scienze economiche e finanziarie G.Prato, Università di Torino terna@econ.unito.it Abstract We introduce here a small set of artificial experiments in social science and a simplified tool useful to develop them in the perspective of the analysis of complex systems, with different degrees of cognitive definition of agents’ behavior. The straightforward way to build this kind of structures requires the use of agent based simulation techniques, i.e. models in which small software routines behave representing artificial agents within artificial environments, and ecosystems are built that allow the emergence of institutions such as market, organizations, hierarchal structures. jESOF (java Enterprise Simulation Open Foundations, web.econ.unito.it/terna/jes ) is a tool created with the aim of simplifying for social scientists the access to agent based simulation, that does not require to be high level specialized computer scientists. This kind of research is in a middle point amid an applied work and a theoretical speculation because we can use this line of analysis both to model actual situations, analyzing also the effects of changes, and to build abstract structures of interacting units, firms or districts, to simulate emerging behavior. The application to the actual world is also useful to collect facts and situations to be used as stylized template into the theoretical framework. The two examples presented in this analysis show that it is relatively easy to develop models with jESOF and that we are virtually unconstrained in the definition of the related contents. The first model is a test implementation of the well known Preys Predators Model (PPM) with the capability of introducing non only two, but three or more interacting levels (hyper- predators etc.). The second model is based upon two coevolving population: Workers, with their Skills, and Firms (WSF): the two populations have quasi-independent behaviors, based on a closely bounded rationality paradigm. In perspective, the goal is that of discovering what happens if we introduce a more consistent behavior in a cognitive perspective, also evolving agents’ rules with soft computing techniques such as classifier systems or neural networks, using in this case the Cross Targets technique. General Hypothesis (GH) for agent based models To develop cognitive agent based experiments, we introduce the following general hypothesis (GH): an agent, acting in an economic environment, must develop and adapt her capability of evaluating, in a coherent way, (1) what she has to do in order to obtain a specific result and (2) how to foresee the consequences of her actions. The same is true if the agent is interacting with other agents. Beyond this kind of internal consistency (IC), agents can develop other characteristics, for example the capability of adopting actions (following external proposals,