This paper argues that computer systems can be classified according to the relative amounts of cognitive processing that are required from the human user and the system software, and that co-operativity takes different forms in these different classes of systems. It proposes a generic architecture for the class of systems in which cognitive processing is interactively shared between computer and user, and argues that in an important sub-class of systems, much of what is normally understood by ”user-modelling” is unnecessary to provide co-operative responses to questions.