Imagined We: Understanding And Bridging The Gap Between Human Cooperation And Multi-Agent Reinforcement Learning