Simulations vs. Human Playing in Repeated Prisoner’s Dilemma

Abstract

The repeated prisoner’s dilemma is an essential game model which is widely applied in real-world economic situations such as price competition between similar products. Studies of the game have focused on the equilibrium strategies for rational players in simplified settings, which do not necessarily reflect the complexity of real-world applications. Therefore, this paper proposes an advanced model that mimics the real-world dynamics of the game, and uses both simulations and human-playing to study the robustness and applicability of different strategies in the game. The result indeed discovers certain weaknesses of the classical strategies. It further shows that well-known dominant strategies such as tit-for-tat are rarely played by human players (less than $5\%$ of the participants played dominant strategies) and instead they tend to use more involved strategies, which again demonstrates their bounded rationality. Our model also plays a crucial role in analyzing real-world multi-agent systems involving human players.