Learning and stabilization of altruistic behaviors in multi-agent systems by reciprocity