Social network analysis has become a wide-spread tool for the study of animal social organisation. However, despite this broad applicability, the study of social networks is currently limited by both a strong focus on pattern analysis as well as a lack of dynamic interaction models. Here I will show a dynamic modelling approach that can be used to capture the responses of fish social networks to changing environments. We found that fish maintain specific network characteristics across different types of habitat manipulations and that some of these characteristics (such as network positions) are unique to a given network. We also found that in contrast to habitat manipulations, social manipulations produced strong, albeit short-lived, changes in network dynamics. Furthermore, I will discuss the underlying evolutionary constraints that shaped these networks dynamics and identify some of the ecological consequences of such network dynamics. Finally, our model of the social behaviour of fish can be embodied by interactive robots (that are accepted by live fish as a conspecific) which allows experimental tests of transition dynamics across social networks.