What is the relationship between the score statistic $U(\theta) = \frac{\partial}{\partial \theta} \ln L(\theta)$ and the Fisher Information $I(\theta)$ in the context of large sample hypothesis testing?