AI Associate Ethical Considerations of AI Practice Question
A financial services company deploys an AI system to approve small business loans. The system uses a deep neural network trained on historical loan data. After deployment, an internal audit reveals that the approval rate for minority-owned businesses is 15% lower than for non-minority-owned businesses with similar financial profiles. The company's AI Ethics policy requires that AI systems be fair and transparent. The data science team has access to the training data, model architecture, and feature importance scores. The company wants to understand why the disparity exists and take corrective action. Which approach should the team take first?
⚠ Common exam trap
Salesforce often tests the principle that data bias is the most common root cause of AI fairness issues, tempting candidates to jump to model-level fixes (like threshold adjustment or fairness constraints) instead of first auditing the training data for sampling or labeling bias.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Analyze the training data to determine if there is sampling bias or labeling bias that caused the model to associate minority ownership with higher risk.
The first step in diagnosing an AI fairness issue is to audit the training data for biases such as sampling bias (e.g., underrepresentation of minority-owned businesses) or labeling bias (e.g., historical loan officers unfairly labeling minority applicants as higher risk). Since the team has access to the training data, analyzing it directly addresses the root cause of the disparity before making model-level changes. This aligns with the AI Ethics policy requirement for transparency, as data bias is a common source of unfair outcomes in deep neural networks trained on historical data.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Analyze the training data to determine if there is sampling bias or labeling bias that caused the model to associate minority ownership with higher risk.
Why this is correct
Bias often stems from training data; analyzing data for imbalances or incorrect labels is the first logical step.
- ✗
Apply a disparate impact analysis to quantify the adverse impact and then adjust the decision threshold.
Why it's wrong here
While useful, quantifying impact should come after understanding the data source of bias.
- ✗
Examine the model's weights and activations to identify which features contribute to the disparity.
Why it's wrong here
Deep neural network weights are not easily interpretable; this approach is impractical for understanding bias.
- ✗
Retrain the model with a fairness constraint that penalizes disparities in approval rates.
Why it's wrong here
Retraining without understanding the root cause may not address the underlying bias and could harm accuracy.
Go deeper
Related to this question
About these practice questions
One of 753 original AI Associate practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI Associate practice question is part of Courseiva's free Salesforce certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI Associate exam.