Research On Risk And Cooperation In The Iterated Prisoner’s Dilemma Game

With the globalization of the world economy and the rapid development of information technology, connections between individuals have become increasingly stronger in the 21 st century. Cooperation with others is a crucial strategy for individuals to achieve stable development in many areas. By introducing the concept of income stream risk into the iterated prisoner’s dilemma game(IPD), this thesis investigates the effect of consideration of stable income, i.e., sustainable development, on the emergence of mutual cooperation between individuals. The main contents and contributions include:(1) The background and the significance of the research on the emergence of cooperation are described. Methods based on the IPD game are introduced. A comprehensive review of the related work is given. The key scientific issues addressed in this thesis are presented with a specific research scheme.(2) By defining risk as the standard deviation of the one-move payoffs for players, a novel utility-based IPD model with incorporation of both payoff and risk is formulated. A two-population coevolutionary learning framework, which is characterized by a niching technique, is developed to search optimal strategies for two players, having either identical or different risk attitudes, to play the IPD game. The experimental results reflect that risk-aversion favors cooperation in the IPD game. In particular, in the IPD with short game encounters, in which cooperation has been demonstrated to be difficult to achieve in previous work, a high level of cooperation can be obtained if both players are risk averse.(3) Following prospect theory and other research on human risk behavior, which state that individuals become more risk averse after their performances exceed their aspirations, or more risk seeking after their performances fall short of their aspirations, we build a new multi-agent IPD model, in which agents’ risk attitudes and strategies are coevolved. In the coevolution, the aspiration levels of agents are determined either based on historical comparison or social comparison. Simulation results indicate that agents can sustain a highly cooperative equilibrium when each agent considers only its own historical payoff as the aspiration(i.e., historical comparison) in adjusting its risk attitude. However, agents tend to be more risk seeking when they rely on social comparison to adapt their risk attitudes, which leads to low levels of cooperation in the population.(4) A localized interaction topology and a discriminative interaction method are respectively introduced among agents in the above multi-agent IPD game. Simulation results reveal that, with these two interaction models, the agent population can still achieve a global cooperation at equilibrium under historical comparison. On the other hand, local cooperation emerges in the population in the case of social comparison. To further consider the effect of uncertain payoffs in the real world, a Gaussian variation is added into the game payoffs for both the local and the discriminative interaction models. It is found that the local cooperation of the population under social comparison is more robust to the variations of payoffs than the global cooperation under historical comparison.