Data Ethics
This domain tests how to use data and AI responsibly when legal compliance alone is not enough. CET questions cover ethical principles, consent and privacy, bias and fairness, transparency, accountability, and governance.
Core rules and concepts
- Principles: respect for persons (autonomy and consent), beneficence (benefit), non-maleficence (avoid harm), and justice (fair distribution of benefits and burdens).
- Privacy principles: lawfulness, purpose limitation, data minimization, accuracy, storage limitation, integrity/confidentiality, accountability. These mirror GDPR-style frameworks; the exam tests the ideas, not legal advice.
- Consent: must be informed, specific, freely given, and revocable. Blanket consent for unspecified future uses is weak.
- De-identification: removing names is not anonymization. Combinations of age, postcode, and date can re-identify people; pseudonymized data remains personal data.
- Bias: historical bias enters training data; measurement bias enters proxies; deployment bias enters use context. Fairness metrics conflict, so the choice must match the domain.
- Transparency and accountability: people should know when automated decisions affect them, and a named owner should answer for the system.
Worked example
A hospital wants to share patient records with an AI vendor to build a readmission model. The vendor proposes removing names and addresses.
Reasoning: removing direct identifiers is not enough. Dates, rare diagnoses, and postcodes can re-identify patients, and the model may encode bias against groups with less access to care. The hospital needs a lawful basis and specific consent or ethics approval, data minimization (share only needed fields), a data-processing agreement, and a DPIA-style risk review. It should also test subgroup performance and define who reviews model-driven decisions. Legality is a floor, not the whole ethical analysis.
Traps
Do not equate anonymization with deleting names. Do not treat consent as permanent or transferable. Do not assume a model is fair because it contains no protected attribute; proxies can carry bias. Do not confuse explainability with accuracy. 'It is legal' does not settle an ethical question.