Reasons are given to support the claim that strong connections exist between linguistic theories on question semantics and work within AI and logic programming. Various ways of assigning denotations to questions are compared, and it is argued that the earlier, "naive" version is to be preferred to more recent developments. It is shown that it is possible to use the chosen denotation to give a model-theoretic semantics for the concept of "knowing what", from which a relationship between "knowing what" and "knowing" can provably be derived. An application to logic programming is described, which allows formal reasoning about what a logic database "knows", this being in a sense a generalization of the Closed World Assumption. Finally, a "knows-what" meta-interpreter in Prolog is demonstrated, and proved to be sound and complete for a certain class of database programs.