This paper studies learning when agents evaluate outcomes in comparison to reference points, which may be adjusted in light of experience. It shows that certain models of reinforcement learning, motivated by those popular in machine learning and neuroscience, lead to classes of recursive preferences.
