NCA-GENM Question 211
Select 3A company is developing a multimodal generative AI model designed to assist hospitals in diagnosing medical conditions. While reviewing the deployment strategy, the team identifies potential biases in the training data that could lead to inaccurate diagnoses for certain demographic groups. Based on the ethical principles of trustworthy AI, what steps should the team prioritize to address these concerns?
- A
Conduct a fairness audit to evaluate and mitigate biases in the training data.
- B
Ensure that the AI model’s predictions are explainable to medical professionals.
- C
Proceed with deployment as planned and address bias issues after user feedback is collected.
- D
Implement diverse datasets that represent all demographic groups during retraining.
- E
Focus solely on improving model accuracy without considering explainability or fairness.
Show answer and explanation
Correct answers: A, B, D
Explanation
Trustworthy AI principles emphasize fairness, transparency, and accountability. To address bias concerns, the team must prioritize a fairness audit, ensure explainability, and retrain the model with diverse datasets. Ignoring these concerns or deferring them to a later stage would compromise the ethical deployment of the model.
- A. Correct.
Conducting a fairness audit is essential to identify and mitigate biases in the data, a critical step under the ethical principles of trustworthy AI to ensure fairness and inclusivity.
- B. Correct.
Ensuring explainability aligns with the principle of transparency, enabling medical professionals to understand and trust the model's decisions.
- C. Incorrect.
Proceeding with deployment without addressing known biases violates the principle of fairness and can lead to harm, making this an incorrect step to prioritize.
- D. Correct.
Using diverse datasets during retraining ensures fairness and inclusivity, which are key ethical principles of trustworthy AI.
- E. Incorrect.
Focusing solely on accuracy ignores the critical principles of fairness, transparency, and accountability, making this approach ethically insufficient.