Creating an empirical basis for adaptation decisions

How can an adaptive intelligent interface decide what particular action to perform in a given situation, as a function of perceived properties of the user and the situation? Ideally, such decisions should be made on the basis of an empirically derived causal model. In this paper we show how such a model can be constructed given an appropriately limited system and domain: On the basis of data from a controlled experiment, an influence diagram for making adaptation decisions is learned automatically. We then discuss why this method will often be infeasible in practice, and how parts of the method can nonetheless be used to create a more solid basis for adaptation decisions.