Previous research on learning and memory has focused on how they are constructed through past experiences. Recent studies, however, have shed light on that such cognitive processes are in service of higher goals of maximizing future rewards. This review paper aims to introduce and discuss a related line of research. First, this paper introduces researches that show goal-directed model-based reinforcement learning, in which agents choose a behavior that does not necessarily bring immediate rewards but will allow future rewards, based on generalization and analogical extrapolation. It also reviews studies on neural substrates of goal-directed learning, and discusses that cognitive process implicated in striatal dopaminergic signals can also influence memory. Especially, memory is not a merely passive process of storing and retrieving past experiences homogeneously, but rather results of a decision-making process to serve higher goals. The body of research suggests that information on future rewards can have influence on current cognitive processing in a retrospective manner.