Future intelligent environments will operate in dynamic and unpredictable situations. Thus, they will have to be able to dynamically learn how to act, interact, and adapt, with little or no a priori knowledge and without human intervention. That is, such systems should become able to self-develop causal models of themselves and of the environment in which they act (i.e., what their actions imply and what actions induce what effects on the environment), and of their social relationships (i.e., what interactions induce what impact on other systems). In this paper, we introduce key concepts of self-development in intelligent environments, both at the individual and collective level, by framing its key concepts and its relation with causal models. Then, we introduce two case studies, focus of our current (preliminary) experiments. Finally, we discuss related work and some key research challenges.