In Multi-Agent Systems, agents sense their local information, and they estimate the global environment. Autonomous agents should have independent estimate functions according to their local information. The functions should adapt the global environment. We consider this estimate function to be exactly a "norms of behavior" in the human sense. In this study, we construct a competitive social system with selfish autonomous agents. Agents interact with their alternative actions each other repeatedly as like as IPD(Iterated Prisoner's Dilemma)games.Agents have a estimation table for their action independently. Agents'strategies of action are adapted to their own estimate function, and agents'norms of behavior satisfy. But global environment deprives their energy according to a global rule. In such systems, the agent which adapt not only strategy but also norms of behavior to the environment can survive.