Final Assessment

Responsible AI, Ethics, Privacy and Governance — Final Assessment

Thirty-five scenario-based questions across the orientation and twelve modules — institutional decisions on ethics, bias, privacy, safety, accuracy, copyright, classroom use, policy, procurement, risk and governance. Every question explains why answers are correct or incorrect. Pass mark 80%. Attempt again after focused review if needed.

1. What is the course's central premise about school AI?

Leaders build governance and keep human oversight.AI should govern the school.Vendors are accountable.Governance is optional.

2. Whose interests come first in an AI decision about a child?

The best interests of the child.The school's budget.The vendor's roadmap.Staff convenience.

3. What does the least-intrusive test ask?

Is there a safer, less data-hungry route?Is it the cheapest?Is it the newest?Does it save the most time?

4. Why is 'overall accuracy' insufficient evidence of fairness?

It hides higher error rates for a group.It proves fairness.All groups perform equally.Accuracy is irrelevant.

5. What is proxy discrimination?

A neutral variable proxying a protected trait.Direct use of race.An accessibility feature.Random error.

6. Why must every activity have an accessible, AI-optional path?

So lack of access is never a penalty.To require AI everywhere.Accessibility is optional.Only for exams.

7. What does data minimisation require?

Collect only what the purpose needs.Collect everything available.Collect for resale.Minimisation is optional.

8. Where do most AI privacy breaches in schools happen?

At the prompt, typing real student data.Only in the server room.Only during exams.Breaches never happen.

9. What does a privacy impact assessment end in?

A residual-risk decision after safeguards.A marketing summary.The vendor's price.Nothing actionable.

10. Why condition adoption on 'no student data for model training'?

It is high-risk secondary use beyond the purpose.It improves the school's tools.It is always harmless.Contracts cannot restrict it.

11. In an AI incident involving a child, what takes priority?

The child's immediate safety.The school's reputation.Preserving evidence first.The vendor's contract.

12. What is red-team thinking?

Anticipating misuse before deployment.Punishing after the fact.Ignoring risks.A marketing tactic.

13. Why must harmful content never be forwarded, even to colleagues?

It spreads harm and can be unlawful.Colleagues need to see it.It is always fine.Forwarding preserves evidence.

14. Why is a confident, cited AI claim not automatically true?

It can fabricate both the fact and the source.Citations always prove truth.Confident output is verified.AI cannot invent sources.

15. What do the six VERIFY steps stand for?

Verify, Examine, Review, Inspect, Fit, Yield.Verify, Edit, Repeat, Ignore, Finish, Yes.No fixed meaning.A vendor's product.

16. When should AI output never be the basis of a decision?

For high-stakes decisions needing human judgement.Never; AI can decide.Only low-stakes tasks.It is always fine.

17. Does an AI generating content mean the institution owns and can publish it freely?

No — ownership is uncertain and it may infringe.Yes, AI output is free.Yes, the school owns it.Copyright never applies to AI.

18. Is an AI-detector flag fair grounds to penalise a student?

No — a detector score is not evidence.Yes, detectors are proof.Yes, always penalise.Detectors are always accurate.

19. How does assessment redesign uphold integrity better than detection?

Authentic evidence resists undisclosed AI.It catches more cheaters.It replaces teaching.It needs no design.

20. Why should AI permissions differ by age?

Age changes what is safe and appropriate.All ages are the same.Younger need less supervision.Age never matters.

21. Where must AI never be the deciding factor in assessment?

In high-stakes, consequential grade decisions.In every task.Nowhere.Only in art.

22. How does the classroom use matrix protect student agency?

It never lets AI do the core learning task.It replaces work with AI.Agency is irrelevant.It bans thinking.

23. Why is a two-line 'we use AI responsibly' statement not a policy?

It answers no operational questions.It is a complete policy.Slogans are enough.Policies need no detail.

24. Why must exactly one person be Accountable for each AI decision?

Shared accountability is no accountability.Many owners is safer.No one need be accountable.Vendors are accountable.

25. What does RACI stand for?

Responsible, Accountable, Consulted, Informed.Review, Approve, Cite, Ignore.A vendor's product.A type of AI.

26. What should be established before evaluating an AI vendor's features?

A clear educational need, necessity, proportionality.The biggest discount.The longest feature list.The best testimonials.

27. Why is the contract, not the demo, the real procurement decision?

The contract governs data, rights and remedies.The demo is binding.Contracts never matter.The salesperson decides.

28. What is a red-line check in a vendor scorecard?

A non-negotiable that disqualifies on failure.A bonus point.A colour scheme.An optional extra.

29. What is residual risk?

The risk remaining after controls.The risk before controls.A vendor's price.Always zero.

30. What does 'patterns, not people' mean in monitoring?

Aggregate trends, not individual surveillance.Tracking every student.No monitoring.Surveilling staff.

31. What are the six stages of the governance cycle?

Govern, Map, Measure, Manage, Monitor, Improve.Buy, Use, Ignore, Repeat.Plan, Do, Stop.A vendor's product.

32. How does a tabletop exercise strengthen governance?

It exposes gaps before a real incident.It replaces monitoring.It is only paperwork.It wastes time.

33. In a multi-stakeholder case, whose interest anchors the decision?

The child's best interests.The vendor's reputation.The institution's convenience.The teacher's preference.

34. What should come first in a twelve-month governance roadmap?

Foundations that reduce the most risk fastest.The annual audit.Everything at once.The least important tasks.

35. What is the capstone of this course?

The sixteen-component governance portfolio.A single quiz.A vendor contract.Nothing.

36. How should course legal references be treated?

As guidance; confirm the law with a professional.As binding law everywhere.As the vendor states.Ignore jurisdiction.

37. Across the whole course, who is accountable for institutional AI?

The institution's leaders, always.The AI tool.The vendor.No one.

38. What distinguishes a voluntary standard from a statutory obligation?

Voluntary = optional best practice; statutory = legally required.They are identical.Voluntary is legally binding.Statutory is optional.

39. Why must AI never be presented as an autonomous decision-maker?

Human oversight must stay central to every decision.AI is more reliable than people.Autonomy saves time.Oversight is optional.

40. Why should the course carry a visible 'not legal advice' statement?

Content is guidance; law varies and changes.The course has no legal content.Law never changes.It is a marketing line.

41. What makes a bias-and-inclusion review a governance control?

It surfaces hidden harm before it scales.It replaces all testing.Bias is always obvious.It is only paperwork.

42. What must an AI contract's exit strategy ensure?

Clean exit with data export, no hostage data.That leaving is impossible.Secret deletion.Nothing.

43. Why require disclosure and provenance for AI-assisted content?

They build trust and traceability.They hide AI use.They are never needed.Only vendors need them.