”Superstition” in the Network: Deep Reinforcement Learning Plays Deceptive Games