Agent Behavior and Transitions in N-Person Social Dilemma Games

Abstract

One of the most common way of analyzing the evolution of networks of interacting agents is based on simulation. The interactions among the agents can be modeled by N-person social dilemma games, and the evolution of the network can be characterized by repeated transitions. The nature of the transitions as well as the long-term behavior of the state of the network depend on the structures of the particular games under consideration. The purpose of this paper is to present the transitions that occur between N-person social dilemma games, the agent behavior causing these transitions, and the impacts of various parameters. Formerly, N-person social dilemma games such as Prisoners’ Dilemma, Chicken, Stag Hunt and Battle of the Sexes have been separately researched and analyzed. In this paper the specific behavior of Pavlovian agents are explored in a two dimensional cellular automaton environment using parameters that extend over all of these games. It is found that there are three significantly different types of agent behavior: bipartisan, partisan, and unison. Bipartisan agents stochastically decide to cooperate or defect based on a cooperating probability greater than zero and less than one, partisan agents either cooperate with certainty or defect with certainty, and unison agents cooperate 100% of the time. Each agent behavior can be associated with a set of social dilemma games and be represented as plateaus on a three dimensional graph. These plateaus themselves and the transitions between them are investigated in terms of where and why they occur. Lastly, this paper reviews the impact of initial cooperating probability, neighborhood size, learning factors, and grid size on these plateaus and transitions.