Mitigating Prisoner's Dilemma by Moral-Based Decision Making Among Agents

Takayuki Ito

Abstract: In this paper, we analyze the impact of morality to mitigate Prisoner's Dilemma in multiagent systems. Prisoner's dilemma is one of the examples of non-zero sum game. In a typical explanation of this game, because two agents make decision rationally (selfishly), the result will be the worst for both of the agents. Even if there is a socially better outcome, they can not reach that outcome. It is well known that agents' selfishness (complete rationality) is one of the main problems that happens this situation. This problem has been tackled by many different approaches. In this paper, we address this problem by introducing moral-based decision models for agents. Considerable focus has been placed on the axiomatic decision-making model, which emphasizes self-control in the presence of temptation. In our approach, self-control and shame are viewed as mental costs associated with morality, and are thus explicitly integrated into the utility function. This integration is supported by recent advances in behavioral economics, which have experimentally demonstrated that decision-making processes are influenced by multiple value criteria, not just a single one. Gul and Presendorfer initially proposed a decision-making model based on self-control and temptation (referred to as the GP model), which has since been further developed. Building on this framework, Dillenberger and Sadowski introduced a utility model that incorporates shame (or moral considerations), positing that an agent must navigate the trade-off between selfishness and moral behavior. Based on the above moral-based decision model, agents can efficiently solve the Prisooner's Dilemma problem. In this paper, we show the condition that enables agents to make better decision in the Prisoner's Dilemma problem.