This paper explores the problem of the goal of understanding, proposing that understanding may be linked to prediction. The approach taken is functional and evolutionary, and it integrates an epistemological perspective with the insights of the predictive processing models of cognition. I first consider the possible goals of understanding, taking as few epistemological assumptions as possible. I take into account the notions of ends in themselves and of explanations, and I distinguish between internally-oriented goals and externally-oriented goals. Then I present how prediction might be connected with the goal of cognition as a whole. To show the advantages of this view, I propose three thought experiments: the dark room, the imaginary room and the transparent room. These thought experiments demonstrate that understanding is not necessarily in contrast with prediction, and must be linked with the notions of trust and timescale. I address the potential objections and conclude that connecting understanding to prediction leads to a description of understanding as an epistemic success involving the management of corpora of information, ultimately connected to the precision-weighting of epistemic networks.