AI Strategy, Ethics and Governance interview questions

28 questions from the AI Strategy, Ethics and Governance course, each with the correct answer and an explanation. Try to answer before you open the answer.

AI Strategy Basics

1. What is the best starting point for an AI strategy?

  1. A vendor demo
  2. The newest model
  3. A business problem with a measurable goal
  4. A press release
Show answer

Answer: A business problem with a measurable goal

Value comes from solving defined problems, so the problem and the metric come first.

Lesson: Why Strategy, Not Hype

2. Which task is a strong first AI use case?

  1. Drafting routine support replies that a human reviews
  2. Fully automatic final approval of large loans
  3. A one-time unrepeatable decision
  4. A task with no way to check quality
Show answer

Answer: Drafting routine support replies that a human reviews

Frequent, checkable, low-severity tasks with a human in the loop are the safest place to learn.

Lesson: Finding Good Use Cases

3. Why include a "low risk" criterion in the scoring model?

  1. Because it is required by Python
  2. Risk does not matter
  3. To lower all scores equally
  4. So high-value but dangerous ideas do not automatically win
Show answer

Answer: So high-value but dangerous ideas do not automatically win

Scoring risk alongside value makes safety part of the ranking instead of an afterthought.

Lesson: Prioritising with a Scoring Model

4. Why record a baseline before a pilot?

  1. Baselines are decorative
  2. Without it you cannot show real improvement
  3. It speeds up the model
  4. Regulators ban pilots otherwise
Show answer

Answer: Without it you cannot show real improvement

Comparing against a known starting point is what makes the result trustworthy.

Lesson: Data and Capability Readiness

Business Case and Operating Model

5. What does payback period measure?

  1. How fast the model replies
  2. How long until net benefits cover the upfront cost
  3. The number of users
  4. The size of the dataset
Show answer

Answer: How long until net benefits cover the upfront cost

It is the time to recover the initial investment from ongoing net savings.

Lesson: ROI, NPV and Payback

6. Why compare total cost of ownership over several years?

  1. It avoids all risk
  2. TCO is always the same for every option
  3. Years do not matter
  4. Upfront price alone hides ongoing costs and can reverse the ranking
Show answer

Answer: Upfront price alone hides ongoing costs and can reverse the ranking

A cheap start can be expensive over time, as the crossover calculation shows.

Lesson: Build, Buy or Partner

7. Why name exactly one accountable person per decision?

  1. So responsibility is clear and nothing falls between roles
  2. Because committees are illegal
  3. To reduce the number of meetings to zero
  4. Accountability slows decisions down by design
Show answer

Answer: So responsibility is clear and nothing falls between roles

A single owner can be asked, decide and be held to account.

Lesson: Teams, Roles and Operating Model

8. What helps adoption most?

  1. Hiding the limitations
  2. Announcing the tool by email only
  3. Involving users early and training them to review output
  4. Making usage secret
Show answer

Answer: Involving users early and training them to review output

People adopt tools they understand, trust appropriately and helped shape.

Lesson: Change Management and Skills

Ethics Foundations

9. How should a team use ethical principles?

  1. Ignore them if the law allows
  2. Print them on a poster and move on
  3. Turn them into concrete questions asked before launch
  4. Leave them to the vendor
Show answer

Answer: Turn them into concrete questions asked before launch

Questions in the process make principles actionable and checkable.

Lesson: Core Ethical Principles

10. Who might be affected by an AI hiring screener besides its users?

  1. Job applicants who are screened by it
  2. Nobody
  3. Only the developers
  4. Only the cloud provider
Show answer

Answer: Job applicants who are screened by it

The people the system makes decisions about are stakeholders even if they never use it.

Lesson: Stakeholders and Ethical Dilemmas

11. Group A is selected 60% of the time and group B 36%. What is the disparate impact ratio?

  1. 96
  2. 1.67
  3. 0.24
  4. 0.6
Show answer

Answer: 0.6

36 divided by 60 is 0.6, below the 0.8 rule of thumb, so it warrants investigation.

Lesson: Measuring Fairness

12. What makes an explanation of a decision useful?

  1. It says "the system decided"
  2. It names the main factors and how to contest the outcome
  3. It is as long as possible
  4. It hides the data used
Show answer

Answer: It names the main factors and how to contest the outcome

People need reasons they can understand and a path to challenge or correct them.

Lesson: Transparency and Explainability

Risk Management

13. How is a risk score computed in a simple register?

  1. Likelihood plus cost
  2. Likelihood multiplied by impact
  3. The number of users
  4. The model size
Show answer

Answer: Likelihood multiplied by impact

The product ranks risks so effort goes where expected harm is greatest.

Lesson: The AI Risk Register

14. Why use risk tiers?

  1. They only apply to hardware
  2. All use cases get identical rules
  3. Tiers remove the need for review
  4. Controls can match the level of risk without slowing every project equally
Show answer

Answer: Controls can match the level of risk without slowing every project equally

Proportionate governance keeps safe innovation quick and focuses scrutiny on serious cases.

Lesson: Risk Tiers for Use Cases

15. When should an AI impact assessment be done?

  1. Before launch, and updated when the system changes
  2. Only after a complaint
  3. Never
  4. Once, then forgotten
Show answer

Answer: Before launch, and updated when the system changes

Assessing risks up front lets you fix design problems before people are affected.

Lesson: Impact Assessments

Governance Framework

16. Why provide approved tools and a fast approval route?

  1. Approved tools are always free
  2. To make the policy longer
  3. Otherwise people use unapproved tools in secret ("shadow AI")
  4. It removes the need for any rules
Show answer

Answer: Otherwise people use unapproved tools in secret ("shadow AI")

Governance that is slow or unclear is bypassed, which creates the risks it was meant to prevent.

Lesson: AI Policy and Governance Roles

17. Why keep an inventory of AI systems?

  1. It speeds up the models
  2. It is only decoration
  3. You cannot govern, review or audit systems you do not know exist
  4. It replaces risk assessment
Show answer

Answer: You cannot govern, review or audit systems you do not know exist

An inventory is the base of every other governance activity.

Lesson: Inventory and Lifecycle Gates

18. Why ask whether you can pin a model version?

  1. A silent model update can change behaviour and break your tests
  2. Pinning makes the model cheaper
  3. Versions do not exist
  4. It is illegal to update
Show answer

Answer: A silent model update can change behaviour and break your tests

Version control lets you re-run your evaluation before accepting a change.

Lesson: Third-Party and Vendor Governance

19. What makes documentation useful for audits?

  1. Deleting old records
  2. A very long single file
  3. Verbal agreements only
  4. Linked, versioned evidence of what was tested, approved and changed
Show answer

Answer: Linked, versioned evidence of what was tested, approved and changed

Traceable records let others verify decisions and reconstruct what happened.

Lesson: Documentation and Auditability

Regulation and Standards

20. What does a risk-based regulation approach mean?

  1. The same rules for every system
  2. Stricter obligations for systems with greater potential harm
  3. No rules for any system
  4. Rules only for hardware
Show answer

Answer: Stricter obligations for systems with greater potential harm

Obligations scale with the level of risk, which mirrors internal risk tiering.

Lesson: The Regulatory Landscape

21. Which are the four functions of the NIST AI RMF?

  1. Buy, Sell, Hold, Trade
  2. Govern, Map, Measure, Manage
  3. Plan, Do, Skip, Stop
  4. Read, Write, Edit, Delete
Show answer

Answer: Govern, Map, Measure, Manage

Govern, Map, Measure and Manage structure how an organisation handles AI risk.

Lesson: NIST AI RMF, ISO/IEC 42001 and OECD

22. What turns a legal requirement into something enforceable inside a company?

  1. Waiting for an audit
  2. A poster in the lobby
  3. A promise in an email
  4. A control with an owner and evidence
Show answer

Answer: A control with an owner and evidence

Owners and evidence make a requirement checkable and sustainable.

Lesson: Compliance Operations

Operating and Monitoring

23. What is the point of a blameless post-incident review?

  1. To fix the system and process so it does not recur
  2. To punish the person involved
  3. To keep the incident secret
  4. To avoid learning anything
Show answer

Answer: To fix the system and process so it does not recur

Blame hides information; focusing on the system surfaces causes and improves safeguards.

Lesson: AI Incident Management

24. Why pair a value metric like containment with a quality metric like satisfaction?

  1. Containment is not measurable
  2. Satisfaction is always lower
  3. Raising one alone can hide damage to the thing that matters
  4. Pairs of metrics are required by law
Show answer

Answer: Raising one alone can hide damage to the thing that matters

A balanced scorecard stops teams from gaming one number at the expense of customers.

Lesson: KPIs and Reporting to Leadership

Putting It Together

25. Why was automatic claim rejection ranked low in the case study?

  1. It is too cheap
  2. Insurance cannot use AI
  3. It has high impact on customers and needs fairness and appeal safeguards first
  4. Scoring models ignore risk
Show answer

Answer: It has high impact on customers and needs fairness and appeal safeguards first

High-impact decisions about people need stronger safeguards before automation.

Lesson: Case Study: An Insurer's First 12 Months

26. Which is the best first AI use case for a cautious organisation?

  1. Human-reviewed drafting of routine documents
  2. Fully automatic decisions on loan approvals
  3. Automatic hiring rejection
  4. An unmonitored public chatbot with account access
Show answer

Answer: Human-reviewed drafting of routine documents

Assistive, reviewable, low-severity work is the safest place to build skills and trust.

Lesson: Revision: Cheat Sheet and Self-Check

27. A screening tool selects group A at 50% and group B at 30%. What does the disparate impact ratio suggest?

  1. Ratios cannot be computed
  2. 1.67, so all is well
  3. 0.2, so it is fair
  4. 0.6, below 0.8, so investigate
Show answer

Answer: 0.6, below 0.8, so investigate

30 divided by 50 is 0.6, under the 0.8 rule of thumb, which calls for investigation of data, features and process.

Lesson: Revision: Cheat Sheet and Self-Check

28. Why can a "silent" vendor model update be a governance problem?

  1. Updates are always improvements
  2. It can change behaviour and invalidate earlier evaluation and approvals
  3. It makes the policy shorter
  4. It has no effect on risk
Show answer

Answer: It can change behaviour and invalidate earlier evaluation and approvals

Approval rests on tested behaviour, so a change in the model means testing and approval must be repeated.

Lesson: Revision: Cheat Sheet and Self-Check

← Back to the course