This paper argues that progress in our scientific understanding of the 'social mind' is hampered by a number of unfounded assumptions. It singles out the widely shared assumption that social behavior depends solely on the capacities of an individual agent. In contrast, both developmental and phenomenological studies suggest that the personal-level capacity for detached 'social cognition' (conceived as a process of theorizing about and/or simulating another mind) is a secondary achievement that is dependent on more immediate processes of embodied social interaction. The paper draws on the enactive approach to cognitive science to further clarify this strong notion of 'social interaction' in theoretical terms. In addition, it indicates how this interaction theory (IT) could eventually be formalized with the help of a dynamical systems perspective on the interaction process, especially by making use of evolutionary robotics modeling. It concludes that bringing together the methods and insights of developmental, phenomenological, enactive and dynamical approaches to social interaction can provide a promising framework for future research.