Modeling human multitasking behavior in video games through modular reinforcement learning