On State Variables, Bandit Problems and POMDPs