What role does interpretability play in the development and deployment of Generative AI systems?