Reading up on Richard Sutton

The standard reinforcement learning formalization errs by placing reward in the environment; instead, reward should be part of the agent, enabling goal-driven behavior.