Hypotheses about typical general human strategic behavior in a concrete case

Abstract

This work is about typical human behavior in general strategic interactions called also games. This work is not about modelling best human performance in well-known games like chess or poker. It tries to answer the question 'How can we exactly describe what we typically do, if we interact strategically in games we have not much experience with?' in a concrete case. This concrete case is a very basic scenario - a repeated zero sum game with imperfect information. Rational behavior in games is best studied by game theory. But, numerous experiments with untrained subjects showed that human behavior in strategic interactions deviates from predictions of game theory. However, people do not behave in a fully indescribable way. Subjects deviate from game theoretic predictions in the concrete scenario presented here. As results, a couple of regularities in the data is presented first which are also reported in related work. Then, the way of using machine learning for describing human behavior is given. Machine learning algorithms provide automatically generated hypotheses about human behavior. Finally, designing a formalism for representing human behavior is discussed. © Springer-Verlag 2009.