On the necessity of integrating multiple levels of abstraction in a single computational framework

We argue that it is imperative that modelers select the right, and potentially differing levels of abstraction for different components of their computational models, as too global or too specific components will hinder scientific progress. We describe ACT-R, from the perspective that is a useful modeling architecture to support this process, and provide two examples in which mixing different levels of abstraction has provided us with new insights.