Reward Maximization Through Discrete Active Inference