A lender is building a generative AI solution to offer discounts to new applicants and wants to minimize bias that could negatively affect some customers. Which actions should they take? (Choose two.)

To minimize bias, the lender should first detect imbalances or disparities in the data. Bias often originates from biased or unrepresentative training data. Identifying and addressing these issues, such as underrepresentation of certain demographic groups, is a crucial first step in building a fair model. Second, they should evaluate the model's behavior so the company can provide transparency to stakeholders. This involves rigorous testing for disparate impact across different groups, understanding how the model makes decisions, and being able to explain its outputs. This transparency builds trust and allows for continuous improvement. Ensuring the model runs frequently or has fast inference times are operational concerns, not directly related to mitigating bias. Using the ROUGE metric is for evaluating text generation quality, not bias, and guaranteeing 100% accuracy is unrealistic and doesn't address fairness.