To replicate ourselves in AI, we first have to embrace human error—and as sci-fi writer Hugh Howey argues, we probably shouldn’t.