Skip to main content

Knowledge check

Quiz
Complete

Evaluate your understanding of this unit by completing the knowledge check​

These questions summarise core themes from ethical frameworks, privacy law, responsible data practice and governance.

Target illustration

Action item: Knowledge check​

Answer each question below. Correct answers and short feedback are included for self-review.

Knowledge check: 8 questions
1. Under UK GDPR, which principle requires personal data to be kept only as long as necessary for the purposes for which it was collected?
  • A. Accountability
  • B. Data minimisation
  • C. Storage limitation
  • D. Integrity and confidentiality
Correct Answer: C

Feedback: Storage limitation is about retention duration; minimisation focuses on collecting only what you need.

2. What is the main purpose of a Data Protection Impact Assessment (DPIA) for a high-risk AI project?
  • A. Replace the need for security controls
  • B. Identify and mitigate privacy risks before processing begins
  • C. Prove the model is unbiased
  • D. Eliminate the need for lawful bases
Correct Answer: B

Feedback: A DPIA is a structured privacy risk assessment; it complements fairness work but does not replace it.

3. Federated learning primarily helps organisations by:
  • A. Training models without centralising raw training data on one server
  • B. Removing the need for encryption
  • C. Guaranteeing that model updates cannot leak any information
  • D. Avoiding documentation of data lineage
Correct Answer: A

Feedback: Federated learning reduces central data aggregation; it still requires secure aggregation and threat modelling.

4. Deontological ethics in AI deployment emphasises:
  • A. Maximising average happiness regardless of rules
  • B. Letting the highest-paid stakeholder decide
  • C. Collecting as much data as possible
  • D. Following clear ethical rules and duties even when inconvenient
Correct Answer: D

Feedback: Deontology is rule- and duty-driven rather than purely outcome-maximising.

5. Temporal bias in training data means:
  • A. The model runs too slowly
  • B. A non-representative time window dominates the dataset
  • C. Timestamps were encrypted incorrectly
  • D. The dataset has too many categorical variables
Correct Answer: B

Feedback: Crisis periods or one-off events can distort “normal” behaviour if not handled carefully.

6. When relying on Legitimate Interests under UK GDPR, organisations must:
  • A. Skip transparency if the model is accurate
  • B. Avoid documenting the purpose of processing
  • C. Perform and record a balancing test against individuals' rights
  • D. Assume legitimate interests applies to any AI use case
Correct Answer: C

Feedback: Legitimate interests is not automatic; necessity and balancing are core.

7. An AI impact assessment (AIIA) is best described as:
  • A. A proactive review of benefits, harms and mitigations before deployment
  • B. A marketing review of model branding
  • C. A substitute for security patching
  • D. A one-line disclaimer in a privacy policy
Correct Answer: A

Feedback: AIIAs help teams anticipate fairness, privacy and societal risks early.

8. The EU AI Act is notable for introducing:
  • A. A ban on all neural networks
  • B. A single accuracy metric for every sector
  • C. Exempting healthcare AI from oversight
  • D. A risk-based framework with obligations scaled to AI system risk
Correct Answer: D

Feedback: Higher-risk systems face stronger governance, documentation and monitoring expectations.