Study on norm emergence and its stability in conflict situations of heterogeneous agent network society

我们提出了智能体在社会中产生规范以有效解决冲突的学习方法，并研究了这种学习方法的特点(收敛、稳定性等)。这是常态。为此，我们使用马尔可夫博弈来表示冲突局势，该博弈具有各种支付矩阵，这些矩阵指示了代理对情况的态度和作为行为结果的回报。我们还表示除非其中一个代理放弃最佳操作(因此这可能导致时间惩罚)，否则永远不会解决的冲突情况。使用我们提出的方法，我们可以发现：(1)规范可以出现，但其分解和稳定的效率受到代理中的支付矩阵的强烈影响；(2)我们经常观察到已建立的规范被少数对冲突具有强烈态度的代理的异常所打破的现象；(3)在具有小世界性质的结构的代理网络中，经常存在不一致的规范。

We proposed the learning method in which agents produce norm in the society for effective conflict resolution and investigated the characteristics (convergence, stability, etc.) of the norm. For this purpose, we expressed conflict situations using a Markov game with a variety of payoff matrices that indicate agent's attitudes to situations and rewards as the result of behavior. We also represent the conflict situations that are never resolved unless one of agents gives up the best action (so this may result in the temporal penalty). Using our proposed method, we could find that (1) norms could emerge but its efficiency of resolution and stability is strongly affected by the payoff matrices in agents; (2) we often observed the phenomenon in which the established norms are broken by a few anomaly of agents that have strong attitude to conflicts; and finally (3) inconsistent norms often coexist in the agent network whose structure has the small-world property.