AI-900 Question 50
Select 3You are deploying an AI solution for a loan approval system. During testing, you notice that applicants from a specific demographic group are disproportionately denied loans compared to others. Which actions should you take to address fairness concerns in the AI solution?
- A
Analyze the training data to ensure it is representative of all demographic groups.
- B
Remove all demographic features from the dataset to eliminate bias.
- C
Implement bias detection and mitigation techniques during model training.
- D
Deploy the model as is, since fairness issues may resolve naturally over time.
- E
Engage stakeholders from diverse backgrounds to evaluate the model’s fairness.
Show answer and explanation
Correct answers: A, C, E
Explanation
Fairness in AI solutions requires addressing bias systematically at various stages of the AI lifecycle. Analyzing training data ensures representation, applying bias mitigation techniques during training actively reduces unfairness, and involving diverse stakeholders helps evaluate the solution from broader perspectives. Simply removing features or ignoring fairness issues does not adequately resolve the problem.
- A. Correct.
Analyzing the training data is critical to identify and correct any imbalances or underrepresentation of specific demographic groups, which can lead to bias in the AI model.
- B. Incorrect.
Removing demographic features entirely may not eliminate bias, as proxies for these features could still exist in the dataset, and this approach can also reduce the model's ability to handle fairness explicitly.
- C. Correct.
Bias detection and mitigation techniques, such as re-sampling or re-weighting the data, are essential steps to address fairness concerns during model training.
- D. Incorrect.
Deploying the model without addressing fairness concerns would perpetuate the existing bias and is not an acceptable action in AI solution development.
- E. Correct.
Engaging stakeholders from diverse backgrounds ensures that the AI solution is evaluated from multiple perspectives, helping to identify and address fairness issues that may not have been initially apparent.