A financial firm is developing a generative AI application to help with loan approval decisions and requires fair and responsible outputs. Which action helps achieve that requirement?

Examining the training data for biases and ensuring representativeness is crucial for fair and responsible AI. Biased or unrepresentative data can lead to discriminatory outcomes, especially in sensitive applications like loan approvals. By proactively addressing data quality, the model is less likely to perpetuate or amplify existing societal biases. Using a deep neural network (many hidden layers) doesn't inherently guarantee fairness; it relates to model complexity. Keeping the model's internal decision process secret (lack of interpretability) hinders the ability to identify and mitigate biases. Relying solely on a fixed test dataset without ongoing checks can miss emerging biases or data drift over time, compromising long-term fairness.