AI-900 Question 49
Select 3A company is building an AI-powered recruitment platform to help screen job applications. During testing, the team notices that the system consistently scores candidates from certain demographic groups lower, even when their qualifications are comparable to others. What steps should the team take to address this issue and ensure fairness in their AI solution?
- A
Review the training data for potential biases and rebalance it if necessary.
- B
Re-train the model using only data from the underrepresented demographic groups.
- C
Implement techniques to evaluate the model's performance across different demographic groups.
- D
Add rules to the AI system to automatically boost scores for the underrepresented groups.
- E
Consult domain experts to better understand the potential sources of bias in the AI system.
Show answer and explanation
Correct answers: A, C, E
Explanation
To ensure fairness in an AI solution, it is crucial to identify and address potential biases in the training data, evaluate the system's performance across diverse groups, and incorporate insights from domain experts. These steps help create a more equitable AI system while avoiding actions like artificially boosting scores, which may lead to unintended consequences.
- A. Correct.
Reviewing the training data for biases and ensuring it is representative is a fundamental step in mitigating fairness issues in AI systems.
- B. Incorrect.
Re-training the model using only data from underrepresented groups can result in an imbalanced model that may introduce new biases rather than correcting existing ones.
- C. Correct.
Evaluating the model's performance across demographic groups helps identify disparities and ensures fairness metrics are being met.
- D. Incorrect.
Adding rules to boost scores for specific groups is not a recommended approach as it introduces artificial adjustments that could lead to other fairness issues.
- E. Correct.
Consulting domain experts can provide critical insights into the sources of bias and help design fairer systems.