World Models and Interpretability Are Two Sides of the Same Coin