Maps and their usage have widely evolved recently, to become more and more interactive, multi-scale and accessible. However, the design of maps did not change so much, leading to the following two problems: (1) in theory, it is not formalised how to create a good map in this context, (2) in practice, the most used maps are not good considering the quality criteria defined for the classical (static) maps. Therefore, it is necessary to question the usefulness of these principles in this new context. In this article, we focus on the role of cartographic generalisation in maps where one can easily zoom in and out to make information accessible. We draw up a list of hypotheses on the role of generalisation for pan-scalar maps, based on both a deductive approach (the role of map generalisation is deduced from a review of human-maps interactions), and an inductive approach (observation of maps with diverse qualities). Then, we discuss how these hypotheses might be experimentally verified.