COMP 3270 / Unit 3 / Unit Home / Lesson 3

Lesson 3.3 · Week 2 · Reading: Chapter 3

Human Authority, Oversight, Transparency, and Accountability

Establish who is responsible for what — design meaningful oversight, ensure transparency, and assign accountability to specific humans.

Suggested time: 130–160 minutes Continues Unit 3 Builds on Lesson 3.2

Table of Contents

1. Lesson Overview

In Lessons 3.1 and 3.2 you learned to define a human problem, assess suitability, and evaluate risk. Now you will address the crucial question: Who is responsible for what? Even if a use is suitable and the risks are managed, it remains irresponsible if human authority is absent, oversight is meaningless, transparency is lacking, or accountability is unclear.

This lesson teaches you to distinguish between different stages of an AI‑supported process (evidence, analysis, recommendation, decision, action), identify the forms of human authority required, design meaningful human oversight (not just nominal presence), establish appropriate transparency, assign accountability to specific people, and define clear escalation and stop conditions. By the end, you will be able to determine whether a proposed AI use has the necessary human governance to proceed responsibly.

Why this matters

Generative AI can produce recommendations, but it cannot take responsibility. Human authority, oversight, transparency, and accountability are the pillars that turn a technically capable system into a trustworthy tool. Without them, even a low‑risk use can become problematic because no one is truly in control.

Responsible‑use reminder: Do not enter personal, confidential, proprietary, privileged, or security‑sensitive information into any public or unapproved AI system. Use only synthetic, public, or generalised information for all activities.

2. Learning Outcomes

By the end of this lesson, you should be able to:

  1. Distinguish evidence, analysis, recommendation, decision, and action in an AI‑supported process.
  2. Identify decisions and activities that must remain under human authority.
  3. Explain what makes human oversight meaningful.
  4. Recognize nominal or rubber‑stamp oversight.
  5. Identify appropriate transparency and disclosure requirements.
  6. Assign accountability for outputs, decisions, actions, corrections, and stopping.
  7. Define preliminary escalation and stop conditions.
  8. Design a human authority and oversight plan for a given use case.

3. Before You Begin

4. Retrieval Practice

These questions activate knowledge from Lessons 3.1 and 3.2.

Q1. What is the difference between technical capability and suitability?

Show feedback

Sample answer: Technical capability is what the AI can do; suitability asks whether it is appropriate for the specific context, purpose, and stakeholders.

Q2. What does proportionality mean in the context of AI safeguards?

Show feedback

Sample answer: Safeguards should be commensurate with the level of risk. Low‑risk uses need lighter safeguards; high‑risk uses require stronger measures.

Q3. Why is it important to consider non‑AI alternatives?

Show feedback

Sample answer: To avoid the "solution in search of a problem" trap and to ensure that the chosen approach is genuinely the best fit for the human purpose.

5. Opening Scenario

📋 Scenario: The busy reviewer

A large university is piloting an AI system that generates initial feedback on student essays. The feedback includes comments on structure, argument, and grammar, along with a suggested grade. Instructors are asked to review and approve the AI's feedback and grade before sharing with students.

However, the university has set a target of reviewing 50 essays per hour per instructor. Many instructors report that they can only glance at the AI's output, and they feel pressured to approve most recommendations to meet the target. Some admit they do not have time to read the student's essay in full.

Questions to consider:

  1. Is the instructors' review meaningful? Why or why not?
  2. What evidence, expertise, time, and authority do the instructors have?
  3. Who is responsible if the AI's feedback or grade is wrong?
  4. What should change to make the oversight effective?
Show guided analysis

1. Is the review meaningful?
No, it is nominal or rubber‑stamp oversight. The volume of reviews (50/hour) is too high for careful reading. Instructors lack sufficient time to verify the AI's analysis or to consider the student's work independently.

2. Evidence, expertise, time, and authority:

  • Evidence: Instructors may not have the full essay or the rubric easily available.
  • Expertise: They are subject‑matter experts, but the volume prevents deep engagement.
  • Time: Insufficient.
  • Authority: They have authority to approve or reject, but the system design discourages rejection.

3. Who is responsible?
The instructors are accountable for the final grade, but if they are rubber‑stamping AI recommendations, the real accountability is ambiguous. The AI system cannot be held accountable; the institution and the instructors must bear responsibility.

4. What should change?

  • Reduce the review quota to allow meaningful engagement.
  • Provide instructors with the student's essay and the rubric alongside the AI feedback.
  • Allow instructors to reject or revise feedback easily.
  • Log review decisions and allow for appeals.
  • Ensure that the AI's reasoning is transparent and accessible.

6. Separating the Stages

A responsible AI‑supported process should distinguish between different stages of work. Collapsing these stages can lead to inappropriate delegation and loss of human control.

Table 1. Stages in an AI‑supported workflow
Stage Description Role of AI Human role
Information Raw data, facts, or inputs May retrieve or organise Authorises and curates
Evidence Verified, relevant information selected for use May suggest or extract Validates and selects
Analysis Interpretation or processing of evidence Generates analysis Reviews and interprets
Recommendation A proposed course of action or judgment Provides recommendations Evaluates and decides
Decision Choice among options Not appropriate Makes the decision
Action Execution of the decision May execute (if reversible and authorised) Authorises and monitors
Review Post‑action evaluation May assist Conducts review and learns
🔑 Key principle

AI may assist at any stage, but decisions that have significant consequences must remain human. The further the AI goes along the chain (from information to action), the stronger the human oversight must be.

7. Human Authority

Authority is the legitimate right to perform an action, make a decision, or control a process. Technical ability does not create authority. In an AI context, various forms of authority must be identified and assigned to specific individuals or roles.

Types of human authority to consider:

⚠️ Technical ability ≠ authority

Just because an AI system can generate a medical diagnosis, a legal opinion, or a financial recommendation does not mean it has the authority to do so. Authority comes from law, professional standards, institutional policies, and community consent.

8. Meaningful Human Oversight

Meaningful oversight requires more than a human being present in the loop. It requires that the human reviewer has the capacity and authority to genuinely influence the outcome.

Essential elements of meaningful oversight:

✅ Meaningful oversight in practice

Example: A radiologist reviews AI‑flagged suspicious areas on a medical image. The radiologist sees the original image, has expertise, takes time, can disagree, and is accountable for the final diagnosis. That is meaningful oversight.

9. Nominal Oversight

Nominal oversight (rubber‑stamping) occurs when human review exists in name only, but does not actually provide a check on the AI's output. It is a common failure mode in AI governance.

Indicators of nominal oversight:

⚠️ Nominal oversight is worse than no oversight

It creates a false sense of security, may legitimise bad outputs, and can shield the system from accountability. It is a serious responsibility failure.

10. Transparency

Transparency means making relevant information about the AI use available to stakeholders in a way that is understandable and actionable. Transparency is not just about disclosure; it is about enabling understanding and accountability.

What to disclose (as appropriate):

Transparency should be relevant, understandable, timely, accessible, and actionable. It must be tailored to the audience (e.g., end‑users, affected parties, regulators).

⛔ Disclosure ≠ permission

Disclosing that AI is used does not make an unsuitable use acceptable. If the use is illegitimate or high‑risk without adequate safeguards, transparency does not fix it.

11. Accountability

Accountability identifies who must answer for the outcomes of an AI‑supported process. Accountability must be assigned to specific individuals or roles, not to the AI system.

Accountability questions to answer:

🔑 AI is not accountable

Do not state that an AI system is accountable. Accountability requires human agency, moral responsibility, and the ability to answer for actions and their consequences. AI systems are tools, not agents.

12. Escalation and Stop Conditions

A responsible AI governance framework includes clear triggers for escalation (when to involve more senior or specialist humans) and stop conditions (when the process should be halted).

Escalation may be required when:

Stop conditions may include:

🛑 Stop is a valid decision

Stopping a process is a responsible action when the conditions are not right. It is better to stop than to proceed with unmanaged risk.

13. Decision‑Process Walkthrough

The diagram below illustrates the stages of human authority, oversight, and accountability in the responsible‑use decision process.

[INSERT VERIFIED HUMAN AUTHORITY AND OVERSIGHT DIAGRAM – Stage 3]
Figure 1. Stage 3 of the introductory responsible‑use decision process: human authority, oversight, transparency, and accountability.

Stages in this lesson:

  1. Define the AI role – which stages will AI assist with? (information, evidence, analysis, recommendation?)
  2. Identify the necessary human authority – who has the authority to define purpose, use information, make decisions, etc.
  3. Design meaningful oversight – ensure reviewers have expertise, evidence, time, independence, and authority.
  4. Establish transparency requirements – what should be disclosed, to whom, and how.
  5. Assign accountability – who is responsible for each stage and the final outcome.
  6. Define escalation and stop conditions – when to escalate and when to stop.
  7. Make a preliminary recommendation – proceed, proceed with safeguards, or do not proceed.
✅ Key insight

The goal is to ensure that humans are in control of the high‑value, high‑impact decisions and that there is a clear line of responsibility for every outcome.

14. Worked Example

Continuing with the meeting‑minutes use case, we now add human authority, oversight, transparency, and accountability.

📝 Use case: Drafting meeting minutes (continued)

The project manager uses AI to generate a draft of meeting minutes, which she reviews and corrects before distribution.

1. AI role

The AI generates a structured draft from the meeting transcript. It provides a summary of decisions, discussion points, and action items. The AI's output is a recommendation for the minutes.

2. Human authority

3. Meaningful oversight

4. Transparency

5. Accountability

6. Escalation and stop conditions

7. Preliminary recommendation

Proceed with safeguards. The human authority, oversight, transparency, and accountability are clear and adequate for the moderate‑risk use. The manager should document the process and review the safeguards regularly.

Show discussion

This example shows that even a relatively simple use benefits from explicit assignment of authority and accountability. The manager retains full control, and the transparency measures build trust with the team. The escalation and stop conditions provide a safety net.

If the manager were pressured to approve drafts without proper review, oversight would become nominal, and the accountability would be undermined.

15. Guided Learning Activities

🧩 Activity 1: Separate the Stages
  • Purpose: Practise distinguishing the stages of an AI‑supported process.
  • Learning outcome: Distinguish evidence, analysis, recommendation, decision, and action.
  • Estimated time: 15 minutes.

Instructions: For the scenario below, identify which stages the AI assists with and which stages remain human.

Scenario: A hiring manager uses an AI system to screen resumes. The AI ranks candidates based on keyword matches and provides a shortlist. The hiring manager then interviews the shortlisted candidates and makes the final hiring decision.

Show model analysis
  • Information: Resumes (human provides).
  • Evidence: AI extracts keywords and patterns.
  • Analysis: AI ranks and shortlists.
  • Recommendation: AI provides a shortlist.
  • Decision: Human (hiring manager) decides who to interview and hire.
  • Action: Human schedules interviews and extends offers.
  • Review: Human evaluates the process and outcomes.

Key point: The AI assists up to the recommendation stage; the human retains decision and action authority.

🧩 Activity 2: Meaningful Oversight Test
  • Purpose: Evaluate whether oversight is meaningful or nominal.
  • Learning outcome: Explain what makes human oversight meaningful.
  • Estimated time: 15 minutes.

Instructions: For each scenario, assess whether the oversight is meaningful or nominal, and why.

  1. A teacher reviews AI‑generated feedback on student assignments, but the review is optional and the teacher often approves without reading.
  2. A doctor reviews AI‑suggested diagnoses, has access to patient records, and can reject or accept the suggestion.
  3. A financial analyst uses AI to generate risk scores, and the analyst must sign off on each score, but the analyst is given only 30 seconds per score.
Show model analysis
  • Scenario 1: Nominal – optional review, often skipped; lacks accountability.
  • Scenario 2: Meaningful – doctor has expertise, evidence, time (implied), and authority.
  • Scenario 3: Nominal – insufficient time; the analyst cannot meaningfully evaluate each score.
🧩 Activity 3: Authority Map
  • Purpose: Identify who holds different types of authority in a given use.
  • Learning outcome: Identify legitimate human authority.
  • Estimated time: 15 minutes.

Instructions: For the scenario below, map each type of authority to a specific role.

Scenario: A university uses AI to generate draft course syllabi, which are then reviewed and approved by the department chair.

Show model analysis
  • Purpose authority: Department chair / faculty.
  • Information authority: Department chair (uses course descriptions).
  • Professional authority: Department chair (has academic authority).
  • Review authority: Department chair.
  • Decision authority: Department chair.
  • Action authority: Department chair (publishes syllabus).
  • Stop authority: Department chair.

Note: In some cases, the instructor who teaches the course might also have authority, but the scenario specifies the chair.

🧩 Activity 4: Transparency and Accountability Statement
  • Purpose: Draft a transparency and accountability statement for an AI use.
  • Learning outcome: Design a transparency and accountability statement.
  • Estimated time: 20 minutes.

Instructions: Draft a short statement for the following use:

Use: A non‑profit uses AI to draft responses to routine public inquiries. The responses are reviewed and sent by a staff member.

Show model statement

Transparency and accountability statement:

"We use generative AI to assist in drafting responses to routine public inquiries. All AI‑drafted responses are reviewed and approved by a staff member before being sent. The staff member is accountable for the accuracy and appropriateness of the response. If you have concerns about a response, please contact [staff name/email] for review or correction. We do not use AI for decisions that affect your rights or benefits."

Key elements: AI role disclosed, human review, accountability assigned, and a correction channel provided.

🧩 Activity 5: Escalate, Stop, or Proceed?
  • Purpose: Practise deciding when to escalate, stop, or proceed.
  • Learning outcome: Select proceed, escalate, revise, or stop.
  • Estimated time: 15 minutes.

Instructions: For each situation, decide whether to escalate, stop, or proceed with the AI use.

  1. The AI's output contains a factual error that can be easily corrected.
  2. The AI is processing personal data without explicit consent.
  3. The AI's recommendation conflicts with a professional standard, and the reviewer is unsure.
  4. The AI system has started producing gibberish for unknown reasons.
Show model analysis
  • Factual error: Proceed with correction (safeguard in place).
  • No consent: Stop immediately; this is an authority violation.
  • Conflict with professional standard: Escalate to a senior expert for review.
  • Gibberish: Stop the process; investigate the system.

16. Exercises

Exercise 3.3.1 – Separate analysis, recommendation, decision, and action
  • Learning outcome: Separate analysis, recommendation, decision, and action.
  • Estimated time: 10 minutes.
  • Scenario: A public health agency uses AI to analyse disease outbreak data and generate a list of potential interventions. Public health officials review the list and decide which interventions to implement.
  • Task: Identify which stage is the AI's role, which is the human's role, and where the transition occurs.
Show sample solution
  • AI role: Analysis (identify patterns) and Recommendation (list interventions).
  • Human role: Decision (choose interventions) and Action (implement).
  • Transition: AI stops at recommendation; human makes decision and action.
Exercise 3.3.2 – Identify legitimate human authority
  • Learning outcome: Identify legitimate human authority.
  • Estimated time: 10 minutes.
  • Scenario: A financial planner uses AI to generate investment recommendations. The planner has a fiduciary duty to act in the client's best interest.
  • Task: Who holds the authority to make the final investment decisions? Why can't the AI make them?
Show sample answer

Final decision authority: The financial planner (and the client).

Why not AI: The planner has a professional and legal obligation that cannot be delegated. The AI lacks the fiduciary responsibility, understanding of the client's unique circumstances, and accountability.

Exercise 3.3.3 – Distinguish meaningful from nominal oversight
  • Learning outcome: Distinguish meaningful from nominal oversight.
  • Estimated time: 10 minutes.
  • Scenario: A bank uses AI to assess credit risk. A loan officer reviews the AI's score and must approve or reject each application. The officer is evaluated on how quickly they process applications, and rejecting an application requires a lengthy justification.
  • Task: Is the oversight meaningful or nominal? Justify.
Show sample answer

Nominal. The pressure to process quickly and the disincentive to reject (lengthy justification) means the officer is unlikely to meaningfully challenge the AI. The officer lacks independence and time.

Exercise 3.3.4 – Design a transparency and accountability statement
  • Learning outcome: Design a transparency and accountability statement.
  • Estimated time: 15 minutes.
  • Scenario: An online education platform uses AI to generate personalised study plans for students. The plans are based on student performance data and are reviewed by an academic advisor before being shared.
  • Task: Draft a short transparency statement for students and an accountability statement for the institution.
Show sample statement

Transparency to students:

"Your personalised study plan was generated with the assistance of artificial intelligence, using your performance data. It has been reviewed by your academic advisor to ensure it is appropriate for your learning goals."

Accountability statement:

"Academic advisors are responsible for the final content of your study plan. If you have concerns, please contact your advisor to discuss revisions."

Exercise 3.3.5 – Select proceed, escalate, revise, or stop
  • Learning outcome: Select proceed, escalate, revise, or stop.
  • Estimated time: 10 minutes.
  • Scenario: A teacher uses AI to generate quiz questions. The first batch contains several questions with ambiguous wording and one with a factual error. The teacher can edit the questions.
  • Task: What should the teacher do? Proceed, escalate, revise, or stop?
Show sample answer

Revise. The teacher should correct the ambiguous and erroneous questions. Since the teacher can edit, this is a manageable issue. If the errors were widespread or the teacher lacked expertise, escalation or stopping might be appropriate.

17. Homework Problems

Formative · Ungraded · Recommended preparation for later units and Assignment 1 reflection.

Problem 1 – Process‑stage separation
  • Learning outcome: Separate analysis, recommendation, decision, and action.
  • Scenario: A city uses AI to analyse traffic patterns and suggest traffic light timing adjustments. City engineers review the suggestions and implement them.
  • Task: Identify the AI's role and the human's role at each stage. Where is the boundary?
Show hint
Think about who makes the final decision to change the timing.
Show sample solution
  • Analysis: AI analyses data and identifies patterns.
  • Recommendation: AI suggests adjustments.
  • Decision: City engineers decide which adjustments to implement.
  • Action: Engineers or technicians implement the changes.
  • Boundary: AI stops at recommendation; human makes decision and action.
Problem 2 – Human authority
  • Learning outcome: Identify legitimate human authority.
  • Scenario: A school district uses AI to generate report card comments based on student grades and attendance. The comments are reviewed by teachers before being sent to parents.
  • Task: Who has the authority to decide the final comments? What authority does the teacher have that the AI lacks?
Show hint
Consider the teacher's professional judgment and knowledge of the student.
Show sample answer

Final authority: The teacher.

Teacher's authority: The teacher has professional judgment, knowledge of the student's personal circumstances, and the ability to tailor the message. The AI lacks these human and contextual insights.

Problem 3 – Oversight effectiveness
  • Learning outcome: Explain what makes human oversight meaningful.
  • Scenario: A hospital uses AI to prioritise emergency room patients based on triage scores. A nurse reviews the AI's triage scores before assigning beds, but the nurse is often overwhelmed and rarely changes the AI's ranking.
  • Task: Is the nurse's oversight meaningful? Why or why not? What could be improved?
Show hint
Consider the nurse's time, independence, and ability to disagree.
Show sample answer

Not meaningful. The nurse lacks time and is pressured to accept the AI's ranking. This is nominal oversight.

Improvements: Reduce nurse workload, provide better training, allow more time for review, and ensure that the nurse can override the AI without penalty.

Problem 4 – Reviewer capacity
  • Learning outcome: Recognise nominal oversight.
  • Scenario: A university uses AI to flag potential plagiarism in student submissions. An academic integrity officer must review each flag, but the officer receives hundreds of flags per day and can only spend a few seconds per case.
  • Task: What is the problem with this oversight? What should be done?
Show hint
Volume exceeds capacity, making review superficial.
Show sample answer

Problem: The officer cannot meaningfully review each flag. This is nominal oversight and may lead to unfair outcomes.

Solution: Set a reasonable review quota, use a risk‑based triage (e.g., only review high‑risk flags), or increase staffing.

Problem 5 – Transparency requirements
  • Learning outcome: Identify appropriate transparency and disclosure requirements.
  • Scenario: A news organisation uses AI to generate summaries of long reports. The summaries are published on the website, but readers are not told that AI was used.
  • Task: What transparency is needed? Why?
Show hint
Think about journalistic standards and reader trust.
Show sample answer

Transparency needed: Readers should be informed that the summary was AI‑assisted and that human editors reviewed it for accuracy. This maintains trust and allows readers to assess the reliability of the information.

If the AI use is not disclosed, it may mislead readers and violate journalistic ethics.

Problem 6 – Accountability assignment
  • Learning outcome: Assign accountability for outputs, decisions, and actions.
  • Scenario: A software company uses AI to suggest code fixes for bugs. The suggestions are reviewed by a senior developer before being committed.
  • Task: Who is accountable for the correctness of the code after it is committed? Who is accountable if the fix causes a new bug?
Show hint
The senior developer reviewed and approved the code.
Show sample answer

Accountability: The senior developer who reviewed and approved the code is accountable for the final code quality. The company is also accountable for the overall product.

Note: The AI system cannot be held accountable; the human reviewer takes responsibility.

Problem 7 – Escalation condition
  • Learning outcome: Define escalation conditions.
  • Scenario: An AI system is used to screen job applicants for a large retailer. The system flags candidates who match certain criteria. The HR team reviews the flags, but sometimes the AI misses qualified candidates.
  • Task: What escalation conditions should be in place? When should a candidate be manually reviewed?
Show hint
Consider cases where the AI's confidence is low, or where the applicant has unusual qualifications.
Show sample answer
  • Escalate if the AI's confidence score is below a threshold.
  • Escalate if the candidate has a disability or requests accommodation.
  • Escalate if the candidate's qualifications are unusual or not well‑represented in the training data.
  • Escalate if the candidate appeals the decision.
Problem 8 – Stop condition
  • Learning outcome: Define stop conditions.
  • Scenario: A government agency uses AI to classify social media posts for content moderation. The AI is supposed to flag inappropriate content, but a spike in false positives occurs.
  • Task: What stop conditions should be in place? When should the system be halted?
Show hint
Consider the impact of false positives on users and the agency's reputation.
Show sample answer
  • Stop if false positive rate exceeds a certain threshold.
  • Stop if the system's output is being used to restrict speech without human review.
  • Stop if the system exhibits bias against a protected group.
  • Stop if the agency cannot provide adequate human oversight.
Problem 9 – Transfer problem
  • Learning outcome: Apply oversight concepts to a new context.
  • Scenario: A law firm uses AI to review contracts and flag unusual clauses for lawyers to examine. The lawyers must review each flag but often delegate to junior associates who lack experience.
  • Task: Is the oversight meaningful? What changes would you recommend?
Show hint
Consider the expertise of the reviewer and the consequences of missing a clause.
Show sample answer

Not fully meaningful. Junior associates may lack the expertise to spot important issues. This is a risk.

Recommendations: Ensure that reviews are conducted by qualified lawyers, provide training, or have a senior lawyer review the junior's assessments. Also, set clear criteria for when to escalate to a partner.

18. Mini‑Project: Human Authority and Oversight Plan

📁 Human Authority and Oversight Plan
  • Purpose: Extend your use case from Lessons 3.1 and 3.2 by adding human authority, oversight, transparency, and accountability.
  • Learning outcomes: Assign human authority, design meaningful oversight, establish transparency and accountability, and define escalation and stop conditions.
  • Estimated effort: 60–90 minutes.

Instructions: Using the same use case you developed in Lessons 3.1 and 3.2 (or a new one with instructor approval), add the following sections to your plan:

  1. Human‑defined purpose: Who defines the purpose and scope?
  2. Information owner: Who authorises the use of information?
  3. AI role: Which stages (information, evidence, analysis, recommendation) will AI assist with?
  4. Evidence required: What evidence must be available for human review?
  5. Human reviewer: Who will review the AI output? What role or title?
  6. Reviewer expertise: What expertise does the reviewer need?
  7. Evidence available to reviewer: What evidence (original data, context) will the reviewer have?
  8. Review timing: When will review occur (before action, after?)?
  9. Decision authority: Who makes the final decision?
  10. Action authority: Who authorises or executes actions?
  11. Transparency statement: What will be disclosed, to whom, and how?
  12. Accountability statement: Who is accountable for outcomes, corrections, and appeals?
  13. Escalation route: When and to whom will cases be escalated?
  14. Stop authority: Who can stop the process, and under what conditions?
  15. Correction process: How will errors be corrected?
  16. Preliminary review of oversight capacity: Does the reviewer have the time, independence, and authority to be effective?

Deliverable: Submit a written plan (approximately 800–1,000 words) that includes all sections above. Use the worked example as a model.

Privacy and integrity boundaries:
  • Use only synthetic, public, or authorised information.
  • Do not use real personal, confidential, or proprietary information.
  • Do not use a use case that involves a high‑impact decision unless you have the appropriate expertise and authority.
Show self‑review checklist
  • ☐ I have identified the AI's role in terms of stages.
  • ☐ I have assigned human authority for each key decision.
  • ☐ I have specified the reviewer's expertise, evidence, time, and independence.
  • ☐ I have described transparency and accountability.
  • ☐ I have defined escalation and stop conditions.
  • ☐ I have checked that oversight is meaningful, not nominal.
  • ☐ I have considered correction and appeal processes.
Show solution architecture

Structure for a strong plan:

  1. Brief recap of use case (purpose, suitability, risk).
  2. AI role specification – clearly state which stages AI handles.
  3. Authority map – table or list of authorities and who holds them.
  4. Oversight design – reviewer, expertise, evidence, timing, independence.
  5. Transparency and accountability – draft statements.
  6. Escalation and stop – list triggers and processes.
  7. Capacity check – confirm that oversight is feasible.
  8. Conclusion – summarise and state that the use can proceed responsibly.
Show illustrative partial example (different context)

Use case: AI‑generated donor thank‑you letters (from previous lessons).

  • AI role: Generates a draft letter (recommendation).
  • Reviewer: Development coordinator.
  • Expertise: Knows donors and institutional voice.
  • Evidence: Donor name, amount, previous communications.
  • Authority: Coordinator approves and sends.
  • Transparency: Donors are told the letter was AI‑assisted but reviewed by staff.
  • Accountability: Coordinator is accountable for accuracy and tone.
  • Escalation: If donor complains, escalate to director.
  • Stop: If AI generates offensive content, stop and investigate.

Note: This is a partial example; your plan should be more detailed.

19. Self‑Check Quiz

Test your understanding of authority, oversight, transparency, and accountability.

Q1. Which of the following is not a requirement for meaningful human oversight?

Select one:
Show answer and explanation

Correct answer: c) The reviewer is pressured to approve quickly.

Explanation: Pressure to approve quickly undermines meaningful oversight; it is a characteristic of nominal oversight.

Q2. In an AI‑supported process, who should be accountable for the final decision?

Select one:
Show answer and explanation

Correct answer: c) The human who makes the decision.

Explanation: Accountability must be assigned to a specific human. The AI system cannot be accountable.

Q3. What is an example of a stop condition?

Select one:
Show answer and explanation

Correct answer: c) The AI is processing sensitive information without authorisation.

Explanation: Unauthorised processing is a serious stop condition because it violates information authority.

Q4. Which of the following best describes nominal oversight?

Select one:
Show answer and explanation

Correct answer: c) A reviewer who sees only the AI's recommendation and rarely rejects it.

Explanation: This is a classic sign of nominal oversight: limited access to evidence and automatic approval.

Q5. True or false: Transparency means disclosing the use of AI, and that disclosure alone makes the use responsible.

Select one:
Show answer and explanation

Correct answer: False.

Explanation: Disclosure is important but does not make an unsuitable use acceptable. The use must also be legitimate, suitably governed, and properly overseen.

Q6. When should a process be escalated?

Select one:
Show answer and explanation

Correct answer: c) When the case is unusual and requires professional judgment.

Explanation: Escalation is triggered by complexity, uncertainty, or high stakes, not by routine disagreement or busyness.

Q7. Who has the authority to define the purpose of an AI use?

Select one:
Show answer and explanation

Correct answer: d) The human who has legitimate authority over the activity.

Explanation: Purpose authority resides with the person or role that is responsible for the activity and has the right to define its scope and objectives.

Q8. Which of the following is a characteristic of meaningful oversight?

Select one:
Show answer and explanation

Correct answer: b) The reviewer is independent and can disagree.

Explanation: Independence and the ability to disagree are essential for meaningful oversight. The other options are indicators of nominal oversight.

20. Assignment 1 Journal Connection

The following prompts are designed to help you reflect on authority, oversight, transparency, and accountability for your Assignment 1 journal.

📓 Reflective prompts
  1. What does meaningful human oversight mean to you? Provide an example from your own experience or from this lesson.
  2. Why is it important that human presence is not the same as human control?
  3. What transparency and accountability practices would you expect when AI is used in a context that affects you personally?
  4. Describe a situation where you would stop an AI use or escalate it to a more senior person.

Note: These reflections can be used as material for your journal entries.

Show reflection planning checklist
  • ☐ I have described a specific scenario.
  • ☐ I have identified the authority and accountability structures.
  • ☐ I have considered whether oversight is meaningful.
  • ☐ I have reflected on transparency and how it builds trust.
  • ☐ I have thought about when to escalate or stop.

21. Key Takeaways

22. Key Terms

Human authority
The legitimate right to perform an action, make a decision, or control a process.
Meaningful oversight
Human review that includes expertise, access to evidence, sufficient time, independence, authority to disagree, and accountability.
Nominal oversight
Human review that is superficial, lacks genuine authority, and does not meaningfully check the AI's output.
Transparency
The practice of disclosing relevant information about an AI use in a way that is understandable and actionable.
Accountability
The obligation of a specific person or role to answer for the outcomes of an AI‑supported process.
Escalation
The process of involving more senior or specialist humans when a case exceeds routine parameters.
Stop condition
A predefined trigger that halts an AI process to prevent harm or unauthorised action.
Rubber‑stamping
Another term for nominal oversight; approving without genuine review.
Information authority
The right to use specific information in an AI system.
Decision authority
The right to make the final decision in a process.

23. Looking Ahead

In Lesson 3.4, you will integrate all the elements from Lessons 3.1–3.3 into a complete responsible‑use decision. You will apply the AI Paradox and ORBIT to an integrated analysis, develop a reasoned recommendation, and establish preliminary personal AI boundaries.

After completing Unit 3, you will move to Module 2: Prompt Engineering and Human‑AI Collaboration, starting with Unit 4: Prompt Design Fundamentals.

Continue to Lesson 3.4: Integrated Responsible‑Use Decisions and Personal Boundaries Return to Unit 3 Home

24. Sources & Further Reading

All references to the textbook are to the COMP 3270 course materials. Page numbers to be confirmed.

👤