Categorizing Social Norms in a Simulated Resource Gathering Society

Our main interest research is focused on reaching a decentralized form of social order through the usage of social norms in virtual communities. In this paper, we analyze the effects of different sets of social norms within a society. The simulation scenario used for the experiments is a metaphor of a resource-gatherer prehistoric society. Finally, we obtain a qualitative ranking of all the possible sets of social norms in our scenario performing agent-based simulation. Introduction and Related Work Social norms are part of our everyday life. They help people self-organizing in many situations where having an authority representative is not feasible. On the contrary to institutional rules, the responsibility to enforce social norms is not the task of a central authority but a task of each member of the society. From the book of Bicchieri (Bicchieri 2006), the following definition of social norms is extracted: “The social norms I am talking about are not the formal, prescriptive or proscriptive rules designed, imposed, and enforced by an exogenous authority through the administration of selective incentives. I rather discuss informal norms that emerge through the decentralized interaction of agents within a collective and are not imposed or designed by an authority”. Social norms are used in human societies as a mechanism to improve the behaviour of the individuals in those societies without relying on a centralized and omnipresent authority. In recent years, the use of these kinds of norms has been considered also as a mechanism to regulate virtual societies and specifically societies formed by artificial agents ((Saam & Harrer 1999), (Shoham & Tenneholtz 1992), (Walker & Wooldridge 1995), (Grizard et al. 2006)). From another point of view, the possibility of performing agent based simulation on social norms helps us to understand better how they work in human societies. One of the main topics of research regarding the use of social norms in virtual societies is how they emerge, that is, how social norms are created at first instance. Copyright c © 2008, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. This has been studied by several authors ((Axelrod 1986), (Sen & Airiau 2007), (Gilbert 2002), (Kittock 1994), (Savarimuthu et al. 2007), (Excelente-Toledo & Jennings 2004)) who propose different factors that can influence this emergence. We divide the emergence of norms in two different stages: (a) how norms appear in the mind of one or several individuals and (b) how these new norms are spread over the society until they become accepted social norms. We are interested in studying the second stage, the spreading and acceptance of social norms, what Axelrod (Axelrod 1986) calls norm support. Our understanding of norm support deals with the problem of which norm is established as the dominant when more than one norm exists for the same situation. In the literature we can find several works ((Sen & Airiau 2007), (Kittock 1994)) that address with this problem, using a prisoner’s dilemma as evaluation function, converting the problem of norm support in a coordination problem, where the agents have to learn to cooperate with the rest of the society, otherwise any kind of social punishment will be applied to them. Our model, in contrast to those solving coordination problems, can deal with social norms that are not representable in a decision table and the rewards for following a certain norm are not known a priori. A similar approach can be found in the work of Cecconi and Parisi (Cecconi & Parisi 1998), where they also deal with a simulated resource consuming society. In their work, agents do not know beforehand how good the sets of social norms they follow are, even though the authors only consider two well differentiated sets of social norms (individual strategy or collective strategy of resource consumption). However, a society can have several (more than just two as we have already seen in the literature) sets of social norms abided by different members of the society. In the work of Sen (Sen, Biswas, & Debnath 2000), we observe that the authors present 6 different strategies (or sets of social norms), but they study the behaviour of mixed populations of these kinds of agents. Nevertheless, each of these sets of social norms, acting individually, can be of different quality with respect the society’s goal. Therefore, it is useful to know beforehand the quality of a set of norms in a society, assuming that all the agents share the same set of social norms. In this paper we present a deep analysis of simulation results and the statistical techniques used to establish a ranking of quality of all the possible sets of social norms that members of a well-defined society can abide by. The assumption addopted is that all the members share the same set of social norms, with the hypothesis that, when agents find themselves in a socially mixed society, they will tend to a common set of norms, and such set of norms should be optimal. The research contained herein follows that performed by (Villatoro & Sabater-Mir 2007) where a genetic algorithm was the mechanism in charge of finding the most efficient set of norms in a given society. The main motivation (and part of future work) of this research is, once the quality of each different set of social norms is defined, to create simulations of heterogeneous societies. In these simulations agents will be loaded with different sets of social norms, and agents will be provided with the ability of changing their set of social norms. Therefore, we plan to observe a convergence of all the agents into a set of social norms. Our final goal is to study the mechanisms that favour that the final dominant set of social norms is the best in the ranking we have previously established. The article is structured as follows: firstly, we present the motivation of the problem and the inspiration we are using for the simulation scenario. Secondly, it is described the problem we deal with in this article, as well as the hipothesis. Subsequently all the details of the simulation model are specified. Thirdly, the experimental setting is introduced and the results of the experiments are analyzed. Finally, we draw some conclusions from the results obtained.