A Comparison of Human and Agent Reinforcement Learning in Partially Observable Domains