The article presents a general model of the emergence of social order in multi-agent-systems (MAS). The agents consist of two types of neural networks that have the task to generate social actions as their output and to adjust these actions to the actions of other agents. The result is a form of social order, i.e., a set of rules of interaction. The agents can generalize these rules by applying them on new but similar agents. An example is given how this model could be applied to the interaction of humans and roboters for some tasks of NASA.