AB-731 practice question 17 of 35

Tailwind Traders deploys an AI-assisted screening tool for internal job applications. After an internal audit, they discover that candidates from an underrepresented group are recommended less often than others with similar qualifications and performance history. To meet responsible AI standards for fairness, inclusiveness, and accountability, the most appropriate response is to ______.

Full set: 389 questions
Answer options
A: Focus only on improving overall accuracy and ignore group-level di^erences
B: Accept the disparity as long as it improves average hiring speed
C: Investigate potential bias, improve data and features, adjust the model, and update the hiring process where needed Correct answer
D: Turn o^ the AI system for all roles permanently and rely only on manual screening
Correct Answer:
C. Investigate potential bias, improve data and features, adjust the model, and update the hiring process where needed
Explanation

The most appropriate response is to conduct a thorough investigation into the root cause of the bias. This aligns with the core principles of Microsoft's Responsible AI Standard, specifically fairness and accountability. A responsible approach involves analyzing the training data for representation gaps, assessing features for proxies of protected characteristics, and adjusting the model to mitigate the identified bias. Furthermore, accountability requires updating the end-to-end hiring process, which may include enhancing human oversight, to ensure the system is used fairly and equitably in practice.

Why Incorrect

A. Focusing only on overall accuracy while ignoring group-level disparities is a direct violation of the fairness principle.
B. Accepting a discriminatory outcome for the sake of efficiency is unethical and violates responsible AI standards.
D. Permanently disabling the system is an overreaction that avoids addressing the root problem and forgoes the potential benefits of a properly implemented AI tool.

References

1. Microsoft, "Microsoft Responsible AI Standard, v2". Microsoft AI Principles Documentation. June 2022. Section 2.2 (Fairness) and Section 2.5 (Accountability). The standard mandates identifying and mitigating unfairness and establishing oversight mechanisms.

2. H. Suresh and J. Guttag, "A Framework for Understanding Unintended Consequences of Machine Learning". MIT Computer Science and Artificial Intelligence Laboratory. February 4, 2020. https://doi.org/10.48550/arXiv.1901.10002. This paper outlines sources of bias (e.g., data, model) and emphasizes the need for systematic investigation, as described in option C.

3. Microsoft Learn, "Identify principles and practices for responsible AI". AI Transformation Leader Learning Path. Module 4, Unit 3: "Fairness". This unit details the necessity of assessing and mitigating fairness-related harms across different groups.

Scroll to Top