In this paper, we discuss the importance of information systems in modeling interactive computations performed on (complex) granules and we propose a formal approach to interactive computations based on generalized information systems and rough sets, which can be combined with other soft computing paradigms such as fuzzy sets or evolutionary computing but also with machine learning and data mining techniques. We apply proposed ideas to perception based granular computing where basic granules are results of perception. Information systems are treated as dynamic granules used for representing results of interaction of attributes with the environment. Two kinds of attributes are distinguished, namely the perception attributes (including sensory attributes) and the action attributes. Sensory attributes are the basic perception attributes, other perception attributes are constructed on the basis of sensory ones. Actions are activated when their guards (being often complex and vague concepts) are satisfied to a satisfactory degree. The guards can be approximated on the basis of measurements performed by sensory attributes rather than defined exactly. Satisfiability degrees for guards are results of reasoning called as the adaptive judgment. The approximations are induced using hierarchical modeling. We show that information systems can be used for modeling more advanced forms of interactions in hierarchical modeling. The role of hierarchical interactions is emphasized in modeling of perception based interactive computations. Some illustrative examples of interactions used in the ACT-R 6.0 system are reported. ACT-R 6.0 is based on a cognitive architecture and can be treated as an example of a highly interactive complex granule which can be involved in hierarchical interactions.