The Reward Hypothesis

Abstract: The reward hypothesis states "all of what we mean by goals and purposes can be well thought of as maximization of the expected value of the cumulative sum of a received scalar signal (reward)." In this talk, I present our recent line of work that aims to settle this hypothesis---we prove that, under a particular interpretation, the reward hypothesis is true if and only if a set of five conditions hold. I then explore some of the consequences of this finding for the design of learning agents and their goals.