Plan and manage an Azure AI solution →mediumMultiple Select

AI-102 Plan and manage an Azure AI solution Practice Question

Which TWO actions should you take when designing an Azure AI solution that uses Microsoft Foundry to ensure responsible AI practices?

⚠ Common exam trap

The trap is selecting performance or data-retention options that sound operationally beneficial but violate responsible AI principles — candidates must distinguish responsible AI practices (fairness, human oversight, transparency) from general engineering optimizations.

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

✓

Implement a human-in-the-loop review for critical decisions

Option A is correct because implementing a human-in-the-loop review for critical decisions ensures that high-impact or sensitive outcomes are validated by a person before action is taken, which is a core responsible AI safeguard against harmful or erroneous automated decisions. Option C is correct because running an AI fairness assessment on the model systematically evaluates performance across demographic groups and helps detect and mitigate bias, directly supporting Microsoft's responsible AI principles of fairness and inclusiveness. Option B does not belong because optimizing for maximum throughput is a performance and cost concern, not a responsible AI practice, and can even conflict with safety if it bypasses safeguards. Option D does not belong because retaining all training data indefinitely raises privacy, data minimization, and compliance risks rather than promoting responsible AI. Option E does not belong because removing explainability metrics reduces transparency and accountability, which is the opposite of responsible AI design.

Answer analysis

Option-by-option breakdown

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

✓
Implement a human-in-the-loop review for critical decisions
Why this is correct
Human-in-the-loop review places a person in the decision path for high-impact outcomes, catching errors and harmful outputs before they affect users. This directly satisfies the responsible AI requirement by ensuring critical decisions receive human oversight rather than fully automated action.
✗
Optimize the model for maximum throughput
Why it's wrong here
Throughput optimisation targets performance, not fairness, transparency, or harm mitigation, so it does not satisfy responsible AI design. It would be the right focus when latency or cost is the binding constraint, but here it addresses the wrong requirement entirely.
✓
Run an AI fairness assessment on the model
Why this is correct
A fairness assessment measures model performance across sensitive groups, exposing disparate error rates or outcomes. Running it satisfies the responsible AI requirement by quantifying and mitigating bias before deployment, supporting equitable treatment across user populations.
✗
Store all training data indefinitely for auditability
Why it's wrong here
Responsible AI principles favour data minimisation and defined retention, not indefinite storage of training data, which increases privacy and compliance exposure. Indefinite retention would suit a deliberate regulatory archive with documented justification, not a default design choice.
✗
Remove all explainability metrics to simplify the model
Why it's wrong here
Removing explainability metrics directly contradicts responsible AI by eliminating the transparency needed to interpret model decisions and detect bias. Explainability tooling exists to surface feature importance and error patterns for auditing. It would only be defensible where a model's outputs carry no accountability or regulatory scrutiny — rarely true in Foundry deployments.

Go deeper

AI-102 Plan and manage an Azure AI solution Practice Question

Why each option matters

Option-by-option breakdown

Related to this question