Algorithms Regulatory Frameworks & Compliance 4 — Questions and Answers
Question 1: Which algorithmic bias type occurs when historical data used to train a model reflects past discriminatory practices?
- Measurement bias
- Historical bias (Correct answer)
- Aggregation bias
- Deployment bias
Correct answer: Historical bias
Historical bias occurs when training data encodes past societal inequalities or discrimination, causing the algorithm to perpetuate those patterns.
Question 2: A company's recommendation algorithm is found to charge women higher prices than men for identical products. Which U.S. law is most directly implicated?
- Sherman Antitrust Act
- Equal Pay Act
- FTC Act Section 5 (unfair or deceptive practices) (Correct answer)
- Sarbanes-Oxley Act
Correct answer: FTC Act Section 5 (unfair or deceptive practices)
FTC Act Section 5 prohibits unfair or deceptive acts in commerce, which can include discriminatory algorithmic pricing practices.
Question 3: What does 'explainability' specifically require of a high-risk algorithm under emerging AI regulations?
- That the algorithm can explain human emotions
- That stakeholders can understand why the algorithm produced a given output (Correct answer)
- That the algorithm self-documents its code
- That all training data is publicly listed
Correct answer: That stakeholders can understand why the algorithm produced a given output
Explainability requires that the reasoning behind an algorithm's output can be communicated in understandable terms to affected individuals or auditors.
Question 4: Under HIPAA, when is it permissible to use patient data to train a clinical prediction algorithm without individual consent?
- Never — all uses require explicit consent
- When the data is de-identified according to HIPAA standards (Correct answer)
- When the algorithm is only used internally
- When the hospital owns the data
Correct answer: When the data is de-identified according to HIPAA standards
HIPAA permits use of de-identified health information without authorization because it no longer constitutes protected health information (PHI).
Question 5: Which compliance concept requires organizations to assess an algorithm's potential privacy harms before deployment?
- Data inventory audit
- Privacy Impact Assessment (PIA) (Correct answer)
- Penetration testing
- Data lineage mapping
Correct answer: Privacy Impact Assessment (PIA)
A Privacy Impact Assessment systematically evaluates how a system collects, uses, and shares personal data and identifies risks before deployment.
Question 6: What is 'scope creep' risk in algorithmic compliance?
- When an algorithm becomes slower over time
- When an algorithm is applied to use cases beyond its original validated scope (Correct answer)
- When algorithm training data grows too large
- When regulatory requirements change faster than algorithms can adapt
Correct answer: When an algorithm is applied to use cases beyond its original validated scope
Scope creep occurs when an algorithm validated for one purpose is repurposed for decisions it was not designed or approved for, creating unvalidated risk.
Question 7: Which principle requires that the same algorithmic decision process be applied uniformly to individuals regardless of protected characteristics?
- Consistency
- Procedural fairness (Correct answer)
- Outcome parity
- Statistical parity
Correct answer: Procedural fairness
Procedural fairness requires that the process for making algorithmic decisions be consistent and not vary based on protected characteristics like race or gender.
Algorithms Regulatory Frameworks & Compliance 4
Which algorithmic bias type occurs when historical data used to train a model reflects past discriminatory practices?