Artificial Societies , Theory

A growing body of researchers from the social and computer sciences are using computational experimentation in a highly exploratory way. The work done in this area is less "simulation" and more "construction". By analogy with artificial intelligence and artificial life, this synthetic approach has become known as artificial societies. It is argued that artificial societies can aid memetic theory building ultimately producing theories and hypotheses that can be tested in the real world. A set of methodologies loosely based on the Popperian (Popper 1968) conception of theory refutation through the falsification of hypotheses and consequent theory development is presented. This process occurs within a deductive system (the artificial society) which due to it’s complexity requires empirical investigations to refute hypotheses and generate theory. Current work (Hales 1998c) involving construction and experimentation with an artificial society in order to aid the building of meme theory around the processes of stereotyping and group formation is outlined. 1 Artificial Societies Computational modelling and simulation of social systems has a history of almost 40 years (Halpin 1998). Recently a speculative and exploratory form of social modelling has emerged. Termed "Artificial Societies" (ASoc), such models address "possible societies", their general processes, dynamics, and emergent properties (Gilbert & Conte 1995). In the same way that Artificial Intelligence is not limited to the accurate modelling of physiological brain processes so ASoc research does not start from some given scenario or particular social system. The aim is to model features and processes which characterise societies in general: co-operation, specialisation, group formation, hierarchy etc. Asoc work does not strive for superficial realism or direct correspondence with existing societies but for abstract logical relationships that characterise whole categories of phenomena. Generally ASoc’s consist of multiple interacting agents. Each agent minimally consists of: internal state; sets of possible actions; percepts (or perceptual inputs); a shared environment and some form of decision process informing action selection. This latter component of an agent’s "architecture" may vary considerably. It may consist of simple hardwired rules (e.g. the Sugerscape see below), deliberative, planning and goal directed AI systems (e.g. EOS Doran et al 1994); inductive learning (e.g. via connectionist models Hutchins & Hazlehurst 1992) or population level evolutionary methods (e.g. evolutionary game theory Hoffmann & Waring 1996). 2 Artificial Societies and Cultural Evolution One of the more recent and substantial contributions to ASoc work is the Sugarscape (Epstein & Axtell 1996). Cultural transmission in the Sugarscape is effectively an extension of Axelrod's Cultural Model (ACM) (Axelrod 1995) which can be compared to the Social Impact Theory (SIT) models of Nowak & Latané (1994). These two latter models are implemented within a Cellular Automata framework (a 2dimensional grid of fixed agents interacting within some small spatial