AIF-C01 Guidelines for Responsible AI Practice Question

Which TWO practices help ensure transparency in AI systems? (Choose 2)

⚠ Common exam trap

The AIF-C01 exam often tests the misconception that transparency means simplifying the model (e.g., removing features) or hiding logic (e.g., using ensembles or black-box models), when in fact transparency is achieved through explainability tools and thorough documentation of limitations and data sources.

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

✓

Use model-agnostic explainability tools like SHAP

Option B is correct because model-agnostic explainability tools such as SHAP (SHapley Additive exPlanations) quantify each feature's contribution to a prediction using Shapley values from cooperative game theory, making the model's decision logic visible to stakeholders regardless of the underlying algorithm. Option D is correct because documenting model limitations, intended use, and data sources (as in model cards or datasheets for datasets) gives users the information needed to understand when and how a model's outputs can be trusted, which is a core requirement of transparency. Option A is not correct because combining multiple models to obscure decision logic deliberately hides how outputs are produced, which reduces rather than improves transparency. Option C is not correct because dropping all but the most predictive features is a feature-selection technique for performance or simplicity; it does not by itself explain decisions or disclose limitations and data provenance. Option E is not correct because black-box models intentionally conceal internal reasoning to protect proprietary algorithms, which is the opposite of transparency.

Answer analysis

Option-by-option breakdown

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

✗
Combine multiple models to obscure decision logic
Why it's wrong here
Ensembling several models deliberately hides how any individual prediction arises, which directly opposes transparency rather than supporting it. Combining models is used to raise accuracy or robustness, so it belongs in a performance-focused design, not one where explainability is the stated goal.
✓
Use model-agnostic explainability tools like SHAP
Why this is correct
SHAP quantifies each feature's contribution to individual predictions, exposing the reasoning behind model outputs rather than leaving them opaque. This directly satisfies the transparency requirement by making decision logic inspectable to stakeholders, auditors and affected users, enabling accountability and informed challenge of automated outcomes.
✗
Remove all features except the most predictive ones
Why it's wrong here
Dropping all but the most predictive features strips the context needed to explain outcomes and can encode bias through proxy variables. Feature selection is a performance and overfitting technique; transparency requires documenting which features are used and why each was included.
✓
Provide documentation on model limitations and data sources
Why this is correct
Documenting model limitations and data sources directly satisfies transparency by disclosing what a system cannot do and where its training data originated. This gives stakeholders the information needed to assess fitness for purpose, bias risk and appropriate use, rather than treating outputs as infallible.
✗
Use black-box models to protect proprietary algorithms
Why it's wrong here
Black-box models conceal their internal reasoning, so decisions cannot be explained to users, auditors or regulators. Proprietary algorithms are protected through licensing, encryption or access controls; transparency demands documented, interpretable model behaviour and disclosure of limitations instead.

Go deeper

AIF-C01 Guidelines for Responsible AI Practice Question

Why each option matters

Option-by-option breakdown

Related to this question