A Game Theoretic Perspective on Model-Based Reinforcement Learning