AML Professional Standards & Ethics 3 — Questions and Answers
Question 1: A recommendation system is found to create 'filter bubbles' that reinforce users' existing beliefs. Which ethical principle does this MOST directly violate?
- Non-maleficence through epistemic harm (Correct answer)
- Privacy by design
- Data minimization
- Algorithmic auditability
Correct answer: Non-maleficence through epistemic harm
Filter bubbles cause epistemic harm by limiting users' exposure to diverse viewpoints, which is a form of harm even without direct physical injury.
Question 2: What is the primary purpose of a 'model card' as introduced by Mitchell et al. (2019)?
- To store model weights in a portable format
- To document model performance, intended use, and limitations for transparency (Correct answer)
- To create a legal license for model deployment
- To summarize training data statistics for data scientists
Correct answer: To document model performance, intended use, and limitations for transparency
Model cards are standardized documentation intended to communicate a model's intended uses, performance metrics, and known limitations to stakeholders.
Question 3: Under GDPR's Article 22, individuals have the right to not be subject to solely automated decisions that produce significant effects. Which ML deployment scenario MOST likely triggers this requirement?
- A model that suggests product recommendations
- An automated loan approval system with no human review (Correct answer)
- A spam filter for internal email
- A model that forecasts inventory demand
Correct answer: An automated loan approval system with no human review
Automated loan decisions produce significant legal and financial effects on individuals, directly triggering GDPR Article 22 protections.
Question 4: A team member proposes falsifying benchmark results to win a competitive contract. Refusing to participate BEST aligns with which professional standard?
- Competence
- Integrity (Correct answer)
- Confidentiality
- Diligence
Correct answer: Integrity
Integrity requires honesty and refusal to participate in misrepresentation or fraud, even under competitive pressure.
Question 5: An organization uses synthetic data generated from real patient records to train a diagnostic AI. Which ethical consideration MOST requires attention?
- Whether the synthetic data degrades model accuracy
- Whether the generation process adequately protects underlying patient privacy (Correct answer)
- Whether the model can be deployed faster using synthetic data
- Whether synthetic data reduces the need for regulatory approval
Correct answer: Whether the generation process adequately protects underlying patient privacy
Synthetic data derived from sensitive records may still encode identifiable information, requiring rigorous privacy analysis before use.
Question 6: Which of the following is an example of 'Goodhart's Law' manifesting in an ML deployment?
- A model overfits to its training data
- A content moderation metric improves while actual harmful content increases (Correct answer)
- A model's inference speed degrades after retraining
- Feature drift causes a model's accuracy to decline over time
Correct answer: A content moderation metric improves while actual harmful content increases
Goodhart's Law states that when a measure becomes a target, it ceases to be a good measure — optimizing the metric no longer achieves the original goal.
Question 7: A practitioner is ethically obligated to report a colleague's serious misconduct even when it may harm the team's project timeline. This obligation is BEST supported by which ethical framework?
- Consequentialism only
- Deontological duty-based ethics (Correct answer)
- Virtue ethics alone
- Contractarianism
Correct answer: Deontological duty-based ethics
Deontological ethics holds that certain duties — such as honesty and preventing harm — are binding regardless of consequences.
AML Professional Standards & Ethics 3
A recommendation system is found to create 'filter bubbles' that reinforce users' existing beliefs.
Which ethical principle does this MOST directly violate?