AIF-C01 Guidelines for Responsible AI Practice Question

A financial institution uses a machine learning model to approve loan applications. The model is trained on historical data that includes biased lending practices. What is the most effective first step to address potential bias?

⚠ Common exam trap

AWS often tests the misconception that removing demographic features (Option C) is sufficient to eliminate bias, but the trap is that proxy features and systemic biases in the data remain undetected without a thorough audit.

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

✓

Audit the training data for bias and review feature selection

The most effective first step to address potential bias is to audit the training data for bias and review feature selection. This aligns with the AWS Responsible AI guidelines, which emphasize that bias mitigation must start with understanding the data and features before any model changes. Without auditing, you cannot identify the source of bias—whether it's in the labels, sampling, or feature correlations—making subsequent steps like retraining or deployment ineffective or harmful.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✗
Immediately deploy the model and monitor for biased outcomes
Why it's wrong here
Deploying a model trained on biased lending data exposes applicants to discriminatory decisions before any bias measurement occurs. It is tempting because monitoring can reveal disparate impact, but the first step is auditing the training data and evaluating fairness metrics offline, not releasing the model to production.
✗
Retrain the model with synthetic data generated from the original dataset
Why it's wrong here
Synthetic data generated from the biased dataset reproduces its historical lending patterns, so the model inherits the same discrimination. It is tempting because augmentation balances under-represented classes, but the first step is auditing training data and labels for proxy variables and sampling bias before any retraining.
✗
Remove all demographic features from the model
Why it's wrong here
Dropping demographic features leaves correlated proxies such as postcode or occupation, so the model still discriminates while appearing fair. It is tempting because it removes the obvious protected attributes, but the first step is measuring bias in the training data and outcomes before deciding on mitigation.
✓
Audit the training data for bias and review feature selection
Why this is correct
Historical lending bias is embedded in the training data itself, so auditing that data and reviewing feature selection is the only step that removes the bias at its source. Model tuning or post-hoc fixes cannot correct discriminatory patterns the data already encodes.

Go deeper

AIF-C01 Guidelines for Responsible AI Practice Question

Why each option matters

Option-by-option breakdown

Related to this question