Humans observe and interact with the world to acquire knowledge. However,
most existing machine reading comprehension (MRC) tasks miss the interactive,
information-seeking component of comprehension. Such tasks present models with
static documents that contain all necessary information