DSE Ethical Considerations in AI 2 — Questions and Answers
Question 1: A hiring algorithm trained on historical data consistently rates female applicants lower for engineering roles. This is best described as an example of:
- Algorithmic transparency
- Representational bias (Correct answer)
- Overfitting
- Feature engineering
Correct answer: Representational bias
Representational bias occurs when training data reflects historical inequalities, causing the model to perpetuate discrimination.
Question 2: Which principle in the EU AI Act classifies AI systems used in employment decisions as 'high-risk'?
- The precautionary principle
- Proportionality of risk to impact on fundamental rights (Correct answer)
- Open-source mandate
- Minimum viable AI standard
Correct answer: Proportionality of risk to impact on fundamental rights
The EU AI Act classifies systems as high-risk based on their potential to adversely affect fundamental rights such as employment and livelihood.
Question 3: Differential privacy protects individual data by:
- Encrypting all records with AES-256
- Adding calibrated statistical noise to query outputs (Correct answer)
- Anonymizing names and addresses only
- Requiring opt-in consent for each query
Correct answer: Adding calibrated statistical noise to query outputs
Differential privacy adds carefully calibrated noise so aggregate query results cannot reveal individual-level information.
Question 4: A data scientist discovers a production model produces biased outcomes but faces pressure from management to delay fixes. The most ethical course of action is to:
- Delay reporting until the next sprint
- Document findings and escalate through formal channels (Correct answer)
- Remove the model silently
- Reduce the model's accuracy to mask the bias
Correct answer: Document findings and escalate through formal channels
Documenting and escalating ensures accountability, creates an audit trail, and upholds professional responsibility.
Question 5: What does 'explainability' mean in the context of AI ethics?
- The ability to rewrite the model code in plain English
- Providing understandable reasons for a model's specific predictions (Correct answer)
- Publishing full training datasets publicly
- Automating model documentation with LLMs
Correct answer: Providing understandable reasons for a model's specific predictions
Explainability means stakeholders can understand why a model produced a particular output, enabling scrutiny and trust.
Question 6: Which technique is specifically designed to identify which features most influenced a model prediction for a single instance?
- Cross-validation
- SHAP (SHapley Additive exPlanations) (Correct answer)
- Principal Component Analysis
- Gradient clipping
Correct answer: SHAP (SHapley Additive exPlanations)
SHAP assigns each feature a contribution value for a specific prediction based on game-theoretic Shapley values.
Question 7: A company uses a facial recognition model that has a 1% error rate for light-skinned males but a 35% error rate for dark-skinned females. This performance gap is primarily a concern under which ethical principle?
- Efficiency
- Fairness and non-discrimination (Correct answer)
- Scalability
- Intellectual property
Correct answer: Fairness and non-discrimination
Disparate error rates across demographic groups violate fairness and non-discrimination principles central to AI ethics.
DSE Ethical Considerations in AI 2
A hiring algorithm trained on historical data consistently rates female applicants lower for engineering roles.
This is best described as an example of: