cumulative reward intuition

The single most important shift in thinking RL asks of you: care about the total reward over time, not the next reward. A move that scores zero now but sets up a big win later beats a flashy move that grabs a point and loses the game. The agent's true objective is the sum of all future rewards from here on, called the return.

Usually we use a discounted sum, multiplying each future reward by a factor between 0 and 1 raised to how far off it is. This gently favors sooner rewards, reflects uncertainty about the far future, and keeps the total finite in endless tasks. Tuning that discount factor sets the agent's horizon—near zero makes it short-sighted and greedy, near one makes it patient and far-seeing.