Making Sense of Reinforcement Learning and Probabilistic Inference