Can Simulation Techniques Contribute to Microsociological Theory? The Case of Learning Matrices

Anselm Eder and Walter J. Gutjahr

Abstract

Nearly all simulation techniques have one big disadvantage in common: oversimplification, leading to unrealistic results. And nearly all simulation techniques have one big advantage in common: oversimplification, which is the only way to express a theoretical model in clear terms. The oversimplification in our simulation model consists in assuming two interacting partners whose actions/reactions are determined by only two sets of parameters: 1. a matrix of reaction probabilities which is updated according to the subjective evaluation of each reaction of the partner to the actor’s behaviour at each interaction, and 2. a payoff matrix, reflecting each partner’s subjective evaluation for each pair of action/reaction, which remains stable over a longer sequence of interactions. Applications of this model to several problem areas, such as socialization agents, game theory approaches, and Ant Colony Optimization in previous publications of the authors, have shown some practical results. Here, we want to focus on three paradigms which we believe can be deducted from our work: 1. The law of sociodynamics: Social systems whose organization is similar to the model conditions formulated above, tend to a decrease in entropy as they get older, in contrast to physical systems, which seem to do the opposite. 2. Cultural values prevail over individual behavioural dispositions: We believe to have found an argument that social systems with properties similar to our model assumptions are likely to provide individuals with behavioural dispositions which depend a lot more on the (culturally more stable) values attributed to behaviour than on initial behavioural dispositions, reflected in reaction probabilities.
And 3. The slower the learning process, the better the results. Optimal solutions to typical dilemma situations such as the iterated prisoner's dilemma are found more frequently when individuals show 'silly' behaviour, i.e. the rate at which they change their behavioural dispositions is low, whereas a quick learning rate results more often in sub-optimal results for both partners.