AI0-001 AI Concepts and Foundations Practice Question

Answer analysis

Option-by-option breakdown

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

• ✓

  Document the model's intended use, limitations, and performance across demographic subgroups in a model card.

  Why this is correct

  A model card provides structured documentation of a model's development, intended use, and performance characteristics, including subgroup analysis. This practice directly supports transparency by making key information available to stakeholders. It helps clinicians and regulators understand when and how the model should be used, and where it may fail.

• ✓

  Implement a feature attribution method that shows which patient factors contributed most to each prediction.

  Why this is correct

  Feature attribution methods such as SHAP or LIME provide local explanations for individual predictions, showing which factors influenced the output. This supports explainability by allowing clinicians to understand the reasoning behind a specific risk score. It also helps identify potential biases or spurious correlations in the model.

• ✗

  Restrict access to the model's source code and training data to a small group of developers.

  Why it's wrong here

  Restricting access to code and data reduces transparency and hinders external review, which is contrary to ethical AI principles. Transparency often involves sharing appropriate documentation and enabling audits, not secrecy. This practice may erode trust and prevent identification of flaws.

• ✗

  Focus solely on maximizing predictive accuracy, as ethical concerns are secondary to clinical outcomes.

  Why it's wrong here

  Prioritizing accuracy above all else ignores ethical principles such as transparency, fairness, and accountability. A highly accurate model can still be biased or unexplainable, leading to harm. Ethical AI requires balancing performance with other values, not treating them as secondary.

• ✗

  Use a complex ensemble model and provide only the overall accuracy metric to clinicians.

  Why it's wrong here

  Providing only overall accuracy hides potential disparities and does not explain how the model arrives at predictions. Transparency requires more than a single aggregate metric; it demands insight into model behavior and limitations. This practice obscures important details that clinicians need to make informed decisions.