An agent completes a task, which consists of a sequence of states, actions, and rewards (s_1, a_1, r_1), (s_2, a_2, r_2), ..., (s_T, a_T, r_T). To improve the agent's performance, we need to adjust the likelihood of taking each action a_t at state s_t. Consider two different ways to calculate the 'quality score' used to update the action a_t:

• Method 1: The score is the sum of all rewards in the sequence: r_1 + r_2 + ... + r_T.
• Method 2: The score is the sum of rewards from time step t onward: r_t + r_{t+1} + ... + r_T.

Which of the following statements best explains why Method 2 is generally a more effective approach for training the agent than Method 1?

An agent completes an episode of 4 time steps, receiving the following sequence of rewards: r_1 = -10, r_2 = +2, r_3 = +5, r_4 = -1. When updating the agent's decision-making process, what is the 'reward-to-go' value that should be associated with the action taken at time step t=2?

Analyzing Credit Assignment for a Policy Update