COMP 3270 / Unit 3 / Unit Home / Lesson 4

Lesson 3.4 · Week 2 · Reading: Chapter 3

Integrated Responsible‑Use Decisions and Personal Boundaries

Bring together everything from Unit 3 into a complete responsible‑use decision — and establish your own personal AI boundaries.

Suggested time: 140–170 minutes Concludes Unit 3 Builds on Lessons 3.1–3.3

Table of Contents

1. Lesson Overview

In Lessons 3.1–3.3 you have learned to define a human problem, assess suitability, evaluate risk, and design human authority, oversight, transparency, and accountability. Now you will bring all these elements together into an integrated responsible‑use decision.

This lesson teaches you to apply the complete decision sequence, revisit the AI Paradox and ORBIT with a fuller understanding, recognise when responsible refusal is the right choice, and establish your own personal AI boundaries. By the end, you will be able to make a reasoned recommendation about whether a proposed AI use should proceed, proceed with safeguards, be redesigned, or be set aside. You will also complete the foundational reflection for Module 1, preparing you to move on to prompt engineering and deeper collaboration with AI.

Why this matters

The ultimate test of responsible AI use is not whether you can generate a risk assessment or an oversight plan in isolation, but whether you can synthesise all the dimensions and arrive at a judgment that is balanced, defensible, and transparent. This lesson is the capstone of Module 1, giving you the confidence to make and explain responsible decisions in your own context.

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. Apply the complete responsible‑use decision sequence to a proposed AI use.
  2. Revisit the AI Paradox and ORBIT with integrated risk and oversight considerations.
  3. Select and justify a final recommendation (proceed, proceed with safeguards, redesign, use alternative, specialist review, defer, or do not proceed).
  4. Recognize situations in which responsible refusal is the appropriate decision.
  5. Develop preliminary personal AI boundaries based on purpose, risk, and accountability.
  6. Document their reasoning and remaining uncertainty.
  7. Connect the learning from Unit 3 to Assignment 1 and to future modules.

3. Before You Begin

4. Retrieval Practice

These questions activate knowledge from Lessons 3.1–3.3.

Q1. What is the difference between effectiveness and efficiency in AI use?

Show feedback

Sample answer: Effectiveness is about achieving a legitimate human purpose with adequate quality; efficiency is about speed or low cost. A fast, cheap AI output that requires extensive correction may be efficient but not effective.

Q2. What factors determine whether a use is suitable for AI?

Show feedback

Sample answer: Clarity of purpose, legitimacy, information authority, consequence of error, verifiability, reversibility, scale, stakeholder vulnerability, required expertise, and availability of safeguards.

Q3. What makes human oversight meaningful rather than nominal?

Show feedback

Sample answer: Meaningful oversight includes expertise, access to original evidence, sufficient time, independence, authority to disagree, and accountability. Nominal oversight lacks one or more of these.

Q4. Why should accountability never be assigned to an AI system?

Show feedback

Sample answer: AI systems lack human agency, moral responsibility, and the ability to answer for outcomes. Accountability requires a human who can be held to account.

5. Opening Scenario

📋 Scenario: From convenience to consequence

Alex is a graduate student working on a research paper. They start using a generative AI tool to brainstorm ideas and organise their notes. This seems harmless and helpful.

As the deadline approaches, Alex begins using the AI to draft entire sections of the paper, including the literature review and discussion. They also ask the AI to suggest references and to check for "plausible" arguments. Alex does not verify the references or the AI's claims, because they trust the AI's fluency.

The AI generates a polished, convincing draft. Alex submits it, and later, a reviewer discovers that several citations are fabricated and that a key argument misrepresents a source. Alex faces academic integrity proceedings and reputational damage.

Questions to consider:

  1. At what point did Alex's use of AI cross from low‑risk assistance into consequential (and problematic) advice or action?
  2. What evidence requirements should have increased as the AI's role grew?
  3. What professional or academic authority should Alex have sought?
  4. When should Alex have limited or stopped using AI for the paper?
Show guided analysis

1. Crossing point: The risk changed when Alex moved from using AI for brainstorming (low risk) to using it to generate substantive content without verification. The consequence of error shifted from minor inconvenience to academic misconduct and reputational harm.

2. Evidence requirements: As the AI's role grew, Alex should have required verification of references, cross‑checking of claims against original sources, and independent validation of arguments. The AI's output should have been treated as a draft to be thoroughly checked, not as a final product.

3. Authority: Alex should have consulted the course instructor or academic integrity office about allowable uses of AI. Professional authority (academic standards) would have set boundaries.

4. Limiting or stopping: Alex should have stopped using AI for content generation when they realised they were not verifying the output. A reasonable stop condition would be: "If I cannot verify the AI's claims and references, I cannot use them."

6. The Complete Responsible‑Use Decision Sequence

The following sequence integrates all the elements from Lessons 3.1–3.3. It provides a systematic way to evaluate any proposed AI use and arrive at a reasoned recommendation.

📋 Introductory Responsible‑Use Decision Sequence
  1. Define the human problem. What is the need, and what evidence shows it exists?
  2. Establish a legitimate purpose. Is the purpose lawful, ethical, clear, proportionate, and consistent with human authority?
  3. Identify intended users and affected stakeholders. Who benefits, and who bears burden?
  4. Consider non‑AI alternatives. What other approaches exist, and how do they compare?
  5. Define the proposed AI role. Which stages (information, evidence, analysis, recommendation, decision, action) will the AI assist with?
  6. Identify the required information. What information is needed, and do you have authority to use it?
  7. Examine capability and limitations. What can the AI do, and what are its known limitations?
  8. Identify opportunities and risks. Use the AI Paradox: capability → opportunity → corresponding risk.
  9. Assess consequence, detectability, and reversibility. How serious are possible errors, and can they be detected and corrected?
  10. Determine required evidence. What evidence must be available for human review?
  11. Assign human authority and oversight. Who has the authority to define purpose, use information, make decisions, etc.? Design meaningful oversight.
  12. Establish transparency and accountability. What will be disclosed, and who is accountable for what?
  13. Define escalation and stop conditions. When to escalate to senior review, and when to stop.
  14. Make and document a preliminary decision. Choose from the allowed recommendations and justify.

The allowed decisions are:

7. AI Paradox and ORBIT Revisited

Earlier in Unit 3, you learned the AI Paradox (capability creates opportunity and corresponding risk) and the ORBIT framework (Opportunities, Risks, Beneficiaries and Burden Bearers, Intervention and Oversight, Transparency and Accountability). With the full set of concepts from this unit, you can now apply them more deeply.

🔍 Revisiting the AI Paradox

For any proposed AI use, identify:

  • Capability: What the AI can do (e.g., generate text, summarise, recommend).
  • Opportunity: The potential benefit or improvement for the human purpose.
  • Risk: The corresponding adverse consequences, considering context and safeguards.
  • Context: The specific conditions of use (purpose, stakeholders, information).
  • Safeguards: The measures that mitigate risk.
  • Residual uncertainty: What remains unknown or unaddressed.
🔄 ORBIT with integrated depth
  • Opportunities – now weighted by legitimacy and suitability.
  • Risks – now assessed with consequence, detectability, reversibility, scale, vulnerability, and autonomy.
  • Beneficiaries and Burden Bearers – identified explicitly.
  • Intervention and Oversight – now designed as meaningful oversight, not just human presence.
  • Transparency and Accountability – now detailed with specific disclosures and assignment of responsibility.

Use ORBIT as a checklist to ensure you have covered all dimensions before making a final recommendation.

8. Responsible Refusal

Responsible refusal means declining or limiting AI use when the conditions are not right. It is a sign of competence and integrity, not a failure to innovate.

Situations that may call for responsible refusal:

✅ Refusal is a responsible decision

Choosing not to use AI in a given situation is not a sign of being "behind the times." It is a thoughtful judgment that prioritises human welfare, fairness, and accountability.

9. Personal AI Boundaries

Every individual should establish personal AI boundaries – a set of guidelines for their own use of generative AI. These boundaries reflect your values, your context (student, professional, personal), and your assessment of risk.

Consider defining boundaries for:

⚠️ Boundaries are personal and dynamic

Your boundaries may change as you gain experience, as technology evolves, and as your context changes. The important thing is to have explicit, thoughtful boundaries rather than drifting into unreflective use.

10. Decision‑Process Walkthrough

The diagram below illustrates the full responsible‑use decision process for Unit 3.

[INSERT VERIFIED COMPLETE RESPONSIBLE‑USE DECISION DIAGRAM]
Figure 1. Complete responsible‑use decision process (Unit 3).

Walkthrough of the full sequence:

  1. Start with the human problem – define need, stakeholders, current process.
  2. Purpose and alternatives – establish legitimate purpose, compare non‑AI alternatives.
  3. Suitability – assess whether the task is a good fit for AI.
  4. Risk assessment – identify errors, consequences, sensitivity, detectability, reversibility, scale, vulnerability, autonomy.
  5. Proportional safeguards – design controls matched to risk.
  6. Human authority and oversight – assign authority, design meaningful review.
  7. Transparency and accountability – disclose, assign responsibility.
  8. Escalation and stop – define triggers.
  9. Recommendation – choose from allowed options and justify.
  10. Documentation – record reasoning and uncertainty.
🔑 Integration is key

The final recommendation is not based on any single factor but on the weight of evidence across all dimensions. A use may be suitable and low‑risk but still fail on authority or accountability. The decision must be holistic.

11. Worked Example

We now present a complete integrated assessment for a new use case: a small business wants to use AI to generate product descriptions for an online store.

📝 Use case: AI‑generated product descriptions

A small e‑commerce business owner wants to use generative AI to write product descriptions for hundreds of items. The AI would take product specifications (provided by the owner) and generate engaging, SEO‑optimised text. The owner would review each description before publishing.

1. Human problem

Current situation: The owner writes product descriptions manually, which takes about 15 minutes per product. With hundreds of products, this is unsustainable and limits the business's growth.

Evidence of need: The owner has a backlog of products without descriptions, and sales are lower for products with poor descriptions.

Affected stakeholders: Owner, customers, and potentially the supply chain if descriptions affect inventory.

Strengths to preserve: Accurate product information, brand voice, and customer trust.

2. Legitimate purpose

To increase the efficiency of product listing creation while maintaining accurate, appealing product descriptions that inform customers and drive sales.

Legitimacy check: Lawful, ethical, clear, proportionate to the task, and consistent with the owner's authority.

3. Intended outcomes and stakeholders

4. Non‑AI alternatives

5. Proposed AI role

The AI will generate draft descriptions from product specifications. The owner will review and edit each draft before publishing.

6. Information required

Product specifications (e.g., size, colour, material, features). No sensitive personal data. The owner has authority to use this information.

7. Capability and limitations

8. AI Paradox: opportunities and risks

9. Consequences, detectability, reversibility

10. Evidence required

Owner must have access to the original product specifications to verify accuracy. The AI should provide a clear source for each claim.

11. Human authority and oversight

12. Transparency and accountability

13. Escalation and stop conditions

14. Recommendation

Proceed with safeguards. The use is suitable, the risk is moderate but manageable with human review. The owner should pilot the AI on a small set of products, measure accuracy, and refine the process. Safeguards: mandatory review of every description, cross‑check with specs, and a clear policy for correcting errors.

Show discussion of the worked example

This example demonstrates how the integrated sequence leads to a balanced recommendation. The business owner gains efficiency without compromising accuracy because of strong human oversight. The safeguards are proportionate to the moderate risk, and the owner retains full accountability.

If the owner could not commit to reviewing every description, the recommendation might shift to "redesign" or "use a non‑AI alternative." The key is that the decision is based on the specific context and capabilities.

12. Guided Learning Activities

🧩 Activity 1: Apply the Complete Sequence
  • Purpose: Practise applying the full responsible‑use decision sequence.
  • Learning outcome: Apply the complete responsible‑use sequence.
  • Estimated time: 25 minutes.

Instructions: For the following scenario, work through the 14‑step sequence and arrive at a preliminary recommendation. You may use the worked example as a guide.

Scenario: A university wants to use AI to generate draft feedback on student assignments. The AI would highlight strengths and weaknesses and suggest improvements. Instructors would review and edit the feedback before sharing with students.

Show model analysis (condensed)
  • Problem: Instructors spend significant time providing personalised feedback, limiting the amount of feedback they can give.
  • Purpose: To increase the quantity and consistency of feedback while maintaining quality and fairness.
  • Alternatives: Manual feedback, peer review, rubric‑based comments.
  • AI role: Generate draft feedback (analysis and recommendation).
  • Risks: Inaccurate or inappropriate feedback, bias, privacy of student work.
  • Safeguards: Instructor review, access to rubric, training, and clear accountability.
  • Recommendation: Proceed with safeguards, including pilot, instructor training, and a process for student appeals.
🧩 Activity 2: Revise an ORBIT Assessment
  • Purpose: Revisit an earlier ORBIT analysis with new risk and oversight information.
  • Learning outcome: Revise an ORBIT assessment using new risk information.
  • Estimated time: 15 minutes.

Instructions: Use the scenario from Activity 1. First, perform a brief ORBIT analysis. Then, suppose the university learns that the AI system has a known bias against certain student populations. Revise the ORBIT analysis and adjust the recommendation.

Show model revision

Initial ORBIT (brief):

  • O: Efficient, consistent feedback.
  • R: Inaccurate feedback, bias.
  • B: Students benefit from more feedback; instructors bear review burden.
  • I: Instructor review.
  • T: Disclosure to students, accountability of instructors.

Revised with bias information:

  • R: Increased risk of unfair treatment – bias could affect student grades and perceptions.
  • I: Stronger oversight needed – regular audits for bias, additional training, and a formal appeal process.
  • T: Enhanced transparency – disclose that AI is used and provide a way for students to challenge feedback.

Revised recommendation: Proceed with enhanced safeguards (bias audit, appeal process) and consider a non‑AI alternative if bias cannot be mitigated.

🧩 Activity 3: Identify Responsible‑Refusal Conditions
  • Purpose: Recognise situations where refusing AI use is the responsible decision.
  • Learning outcome: Identify responsible‑refusal conditions.
  • Estimated time: 15 minutes.

Instructions: For each scenario, decide whether the responsible decision is to refuse, restrict, or proceed. Justify.

  1. A journalist wants to use AI to generate articles on sensitive political topics, with minimal human oversight.
  2. A researcher wants to use AI to analyse medical records for a study, but the AI service does not meet privacy standards.
  3. A teacher wants to use AI to provide emotional support to students, replacing counselling services.
Show model analysis
  • Scenario 1: Refuse – high risk of misinformation, lack of authority, and professional standards require human judgment.
  • Scenario 2: Refuse (or restrict) – privacy violation; if the researcher cannot use a secure service, they should not proceed.
  • Scenario 3: Refuse – AI cannot replace counselling; this is a safety‑critical and relationship‑based role.
🧩 Activity 4: Develop Preliminary Personal Boundaries
  • Purpose: Begin drafting your own personal AI boundaries.
  • Learning outcome: Develop preliminary personal boundaries.
  • Estimated time: 20 minutes.

Instructions: Using the categories listed in the core content, draft a first version of your personal AI boundary charter. Consider your role (student, professional, etc.) and the contexts in which you use AI.

Show reflection prompts
  • Permitted low‑risk uses: What everyday tasks do you feel comfortable using AI for?
  • Uses requiring verification: When will you always fact‑check and verify?
  • Uses requiring caution: Which activities would you approach with extra care?
  • Protected information: What data will you never enter?
  • Professional referral: When will you consult a human expert instead of AI?
  • No‑AI activities: What will you always do without AI?
  • Stop conditions: What will make you stop using AI in a given context?
  • Documentation: How will you keep a record of your AI use?
  • Review date: When will you revisit your boundaries?

Note: There is no single "correct" charter. Your boundaries should reflect your personal values and context.

13. Exercises

Exercise 3.4.1 – Apply the complete responsible‑use sequence
  • Learning outcome: Apply the complete responsible‑use sequence.
  • Estimated time: 15 minutes.
  • Scenario: A local government wants to use AI to draft responses to routine citizen inquiries about permits and services. The responses would be reviewed by a staff member before sending.
  • Task: Outline the key steps in the decision sequence, focusing on the most critical points. What is your preliminary recommendation?
Show sample solution
  • Problem: Staff time spent on repetitive inquiries, delays in responses.
  • Purpose: Improve response time and consistency.
  • Alternatives: Expand FAQ, hire more staff, use rules‑based automation.
  • AI role: Generate draft responses (recommendation).
  • Risks: Inaccurate information, misleading citizens, legal liability.
  • Safeguards: Mandatory human review, source verification, and a correction process.
  • Recommendation: Proceed with safeguards, pilot with low‑risk inquiries, and gradually expand.
Exercise 3.4.2 – Revise an ORBIT assessment
  • Learning outcome: Revise an ORBIT assessment using new risk information.
  • Estimated time: 10 minutes.
  • Scenario: A hospital uses AI to triage emergency room patients. The AI assigns a priority score. Nurses review the scores and make the final triage decision. Initially, the ORBIT analysis rated the risk as moderate. Later, it is discovered that the AI systematically underestimates the severity of symptoms in elderly patients.
  • Task: Revise the ORBIT analysis and the recommendation.
Show sample answer
  • R: Risk increased – bias against elderly could lead to delayed treatment.
  • B: Elderly patients bear the burden.
  • I: Stronger oversight needed – additional checks for elderly patients, regular audits.
  • T: Disclose the bias and mitigation efforts.
  • Recommendation: Proceed with enhanced safeguards and consider a non‑AI alternative for elderly triage if bias cannot be fixed.
Exercise 3.4.3 – Select and justify a final recommendation
  • Learning outcome: Select and justify a final recommendation.
  • Estimated time: 10 minutes.
  • Scenario: A financial advisor wants to use AI to generate personalised investment recommendations for clients. The advisor would review each recommendation but has a high client load and limited time for deep review.
  • Task: Choose from the allowed recommendations and justify.
Show sample answer

Recommendation: Redesign or Obtain specialist review.

Justification: The advisor's limited time for review undermines meaningful oversight. The consequence of error is high (financial loss). A better approach would be to reduce the client load, use AI only for screening, or have a dedicated review team. Alternatively, use a non‑AI approach if the AI cannot be adequately overseen.

Exercise 3.4.4 – Identify responsible‑refusal conditions
  • Learning outcome: Identify responsible‑refusal conditions.
  • Estimated time: 10 minutes.
  • Scenario: A non‑profit wants to use AI to translate sensitive personal testimonies from survivors of trauma for a research study. The AI service is public and stores data for training.
  • Task: Should the non‑profit proceed? If not, what alternatives exist?
Show sample answer

Responsible refusal: Do not proceed with the public AI service. The information is highly sensitive, and the storage of data poses a privacy risk. Alternatives: use a secure, on‑premises AI system, or use professional human translators with confidentiality agreements.

Exercise 3.4.5 – Develop preliminary personal boundaries
  • Learning outcome: Develop preliminary personal boundaries.
  • Estimated time: 15 minutes.
  • Task: Write a short personal AI boundary charter for yourself as a student. Include at least five specific boundaries.
Show sample charter (illustrative)
  • I will use AI for brainstorming and organising notes, but I will write the final content myself.
  • I will never use AI to generate citations without verifying them.
  • I will not use AI for exams or graded assignments unless explicitly permitted.
  • I will not share personal or confidential information with any AI service.
  • I will document my AI use and be transparent about it when required.

Note: Your charter should be personal and reflect your own values and context.

14. Homework Problems

Formative · Ungraded · Recommended preparation for the Unit 3 consolidation and Assignment 1 reflection.

Problem 1 – Integrated use‑case decision
  • Learning outcome: Apply the complete responsible‑use sequence.
  • Scenario: A large retailer wants to use AI to generate personalised product recommendations for customers based on their purchase history.
  • Task: Work through the decision sequence and make a recommendation. What are the key risks and safeguards?
Show hint
Consider privacy, accuracy, and customer trust.
Show sample solution
  • Problem: Improve customer experience and sales by providing relevant recommendations.
  • Purpose: Legitimate (improving service).
  • Risks: Privacy invasion, inaccurate recommendations, bias.
  • Safeguards: Transparency about data use, opt‑out, human oversight of the recommendation logic, and regular audits.
  • Recommendation: Proceed with safeguards, including clear privacy policy and customer control.
Problem 2 – Revised ORBIT analysis
  • Learning outcome: Revise an ORBIT assessment using new risk information.
  • Scenario: A school uses AI to grade student essays. Initially, the ORBIT analysis was favourable. Later, it is found that the AI penalises essays written by non‑native English speakers.
  • Task: Revise the ORBIT analysis and the recommendation.
Show hint
Consider fairness and the impact on students.
Show sample answer
  • R: Bias risk increased.
  • B: Non‑native English speakers bear the burden.
  • I: Enhanced oversight needed – regular bias audits, human review of disputed grades.
  • T: Disclose the use and provide a way for students to appeal.
  • Recommendation: Proceed only if bias can be mitigated; otherwise, use non‑AI grading.
Problem 3 – Safeguard vs. redesign
  • Learning outcome: Decide between adding safeguards and redesigning the use.
  • Scenario: A social media platform uses AI to moderate content. The AI flags and removes posts that violate community guidelines. Human moderators review flagged posts, but the volume is high, and they often approve without reviewing.
  • Task: Should the platform add more safeguards, or redesign the process? Justify.
Show hint
Consider the effectiveness of the current oversight.
Show sample answer

Redesign. The current oversight is nominal. Adding more safeguards (e.g., more moderators) may not address the root problem. The process should be redesigned to ensure meaningful review, possibly by lowering the volume of flags (using better AI), or by using a tiered review system where only high‑risk flags go to humans.

Problem 4 – Non‑AI alternative preference
  • Learning outcome: Determine when a non‑AI alternative is preferable.
  • Scenario: A law firm wants to use AI to draft legal briefs. The AI would generate a draft, and lawyers would review and edit. However, the AI often makes subtle errors that take significant time to correct.
  • Task: Compare the AI approach with a non‑AI alternative (drafting from scratch). Which is preferable, and why?
Show hint
Consider the total time and quality.
Show sample answer

Non‑AI alternative (drafting from scratch) may be preferable if the time saved by AI is offset by the time needed for extensive correction. If the AI's errors are subtle and hard to catch, the risk of missed errors is high. The firm should pilot the AI and measure total time; if the AI does not provide a net benefit, revert to non‑AI.

Problem 5 – Responsible refusal
  • Learning outcome: Recognize responsible‑refusal conditions.
  • Scenario: A health app wants to use AI to provide real‑time mental health support via chat, with no human backup.
  • Task: Is this an appropriate AI use? If not, what is the responsible refusal?
Show hint
Consider the consequences and the need for human empathy.
Show sample answer

Unsuitable. Mental health support requires human empathy, professional judgment, and the ability to handle crises. The AI cannot provide this. The responsible refusal is to not use AI for this purpose; instead, the app should connect users to human professionals.

Problem 6 – Personal boundary
  • Learning outcome: Develop personal boundaries.
  • Task: Write a personal boundary regarding the use of AI for creative writing (e.g., poems, stories). What is your boundary, and why?
Show hint
Consider the value of personal expression and originality.
Show sample answer

Boundary: I will use AI to brainstorm ideas and overcome writer's block, but I will write the final draft myself without AI assistance. This preserves my creative voice and ensures originality.

Problem 7 – Uncertainty documentation
  • Learning outcome: Document reasoning and remaining uncertainty.
  • Scenario: You have assessed a proposed AI use and recommend proceeding with safeguards. However, there is uncertainty about the AI's performance on edge cases.
  • Task: Write a short note documenting the uncertainty and how it will be managed.
Show hint
State what is unknown and what monitoring will be done.
Show sample answer

Uncertainty: The AI's accuracy on rare scenarios is unknown.

Management: We will monitor the AI's performance on a sample of edge cases, escalate any errors to human review, and adjust the system based on findings.

Problem 8 – Final recommendation
  • Learning outcome: Select and justify a final recommendation.
  • Scenario: A small business wants to use AI to draft personalised email responses to customer inquiries. The owner will review each response, but the owner has limited time and may not always catch errors.
  • Task: What is your final recommendation? Justify.
Show hint
Consider the owner's capacity for meaningful review.
Show sample answer

Recommendation: Pilot or limit the use.

Justification: The owner's limited review capacity is a concern. Start with a pilot for low‑stakes inquiries, track accuracy, and if the owner cannot maintain quality, revert to a non‑AI approach or redesign the workflow (e.g., hire a part‑time reviewer).

Problem 9 – Transfer problem
  • Learning outcome: Apply the integrated decision framework to a new context.
  • Scenario: A community organisation wants to use AI to generate summaries of public meetings for citizens who cannot attend.
  • Task: Assess the use and make a recommendation, considering all dimensions.
Show hint
Consider the need for accuracy and the trust of the community.
Show sample answer
  • Problem: Citizens miss important information from public meetings.
  • Purpose: Improve civic engagement.
  • AI role: Generate summaries from transcripts.
  • Risks: Inaccurate summaries could mislead citizens.
  • Safeguards: Human review by a community member, clear disclaimers, and a way to correct errors.
  • Recommendation: Proceed with safeguards, especially strong human review and a feedback mechanism.
Problem 10 – Module 1 synthesis
  • Learning outcome: Synthesize learning from Units 1–3.
  • Scenario: Reflect on what you have learned about generative AI, its capabilities, limitations, and responsible use.
  • Task: Write a brief reflection on how your understanding has changed from Unit 1 to Unit 3.
Show hint
Consider the shift from "what can AI do?" to "should we use it, and how?"
Show sample reflection

Sample reflection: At the start of Unit 1, I was impressed by what AI could do. Now, I see that capability is only the beginning. I understand that effectiveness and responsibility require careful assessment of purpose, suitability, risk, and human control. I am more cautious and more deliberate in my own use of AI.

15. Mini‑Project: Introductory Responsible AI Use Decision

📁 Introductory Responsible AI Use Decision
  • Purpose: Integrate all Unit 3 components into a preliminary decision report.
  • Learning outcomes: All Unit 3 learning outcomes.
  • Estimated effort: 90–120 minutes.

Instructions: Select a bounded AI use from one of the following contexts (or choose your own with instructor approval):

  • Education / learning
  • Professional practice
  • Organisational activity
  • Communication / creative work
  • Public information / government
  • Everyday life

Develop a comprehensive report that includes all sections listed below. Use the worked example in this lesson as a model.

Required sections:

  1. Use‑case title
  2. Human problem
  3. Evidence of need
  4. Legitimate purpose
  5. Intended outcome
  6. Users and affected stakeholders
  7. Current or non‑AI process
  8. Proposed AI role
  9. Information required
  10. Technical capability
  11. Technical limitation
  12. Opportunity‑risk pair (AI Paradox)
  13. ORBIT analysis
  14. Consequences of error
  15. Detectability
  16. Reversibility
  17. Scale and vulnerability
  18. Preliminary risk category
  19. Evidence requirement
  20. Human authority
  21. Oversight plan
  22. Transparency requirement
  23. Accountability
  24. Escalation condition
  25. Stop condition
  26. Capability or relationship effect
  27. Residual uncertainty
  28. Final recommendation (choose from allowed list)
  29. Reasons
  30. Connection to Assignment 1 reflection

Allowed recommendations:

  • Proceed
  • Proceed with safeguards
  • Pilot or limit
  • Redesign
  • Use a non‑AI alternative
  • Obtain specialist review
  • Defer
  • Do not proceed

Deliverable: Submit a written report (approximately 1,500–2,000 words or equivalent structured submission).

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
  • ☐ All sections are addressed.
  • ☐ The reasoning is clear and justified.
  • ☐ Risks are identified and mitigated with proportionate safeguards.
  • ☐ Human authority and oversight are meaningful.
  • ☐ Transparency and accountability are explicit.
  • ☐ Escalation and stop conditions are defined.
  • ☐ The recommendation is justified.
  • ☐ Residual uncertainty is documented.
  • ☐ The report is well‑structured and professional.
Show solution architecture

Structure for a strong report:

  1. Title and overview – brief summary.
  2. Problem and purpose – define the human problem and legitimate purpose.
  3. Stakeholders and outcomes – identify who is affected and what success looks like.
  4. Alternatives and AI role – compare non‑AI alternatives and specify the AI's role.
  5. Risk assessment – use the AI Paradox and ORBIT, assess consequences, detectability, reversibility.
  6. Risk classification and safeguards – classify risk and design proportionate safeguards.
  7. Authority and oversight – assign human authority and design meaningful oversight.
  8. Transparency and accountability – draft statements.
  9. Escalation and stop – define triggers.
  10. Capability and relationship – consider human skills and relationships.
  11. Residual uncertainty and recommendation – document unknowns and state final decision.
  12. Reflection – connect to Assignment 1.

Note: Use headings and subheadings for clarity.

Show illustrative partial example (different context)

Use case: AI‑generated product descriptions (same as worked example).

Partial outline:

  • Human problem: Slow manual process limiting growth.
  • Purpose: Increase efficiency while maintaining accuracy.
  • AI role: Generate draft descriptions from specs.
  • Risk: Inaccurate descriptions could mislead customers.
  • Safeguards: Mandatory human review, source verification.
  • Accountability: Owner is accountable.
  • Recommendation: Proceed with safeguards.

This is a partial example; your report should be more detailed and include all sections.

16. Self‑Check Quiz

Test your understanding of integrated decision‑making and personal boundaries.

Q1. Which of the following is not an allowed recommendation in the responsible‑use decision framework?

Select one:
Show answer and explanation

Correct answer: c) Automate without any human oversight

Explanation: The framework includes only responsible options; automation without oversight is not a responsible recommendation because it lacks human authority and accountability.

Q2. What is the purpose of revisiting the AI Paradox in the integrated decision sequence?

Select one:
Show answer and explanation

Correct answer: b) To identify opportunities and corresponding risks in context.

Explanation: The AI Paradox helps to connect capability with both opportunity and risk, which is essential for a balanced decision.

Q3. Responsible refusal is:

Select one:
Show answer and explanation

Correct answer: b) A competent decision to limit AI use when conditions are not right.

Explanation: Responsible refusal is a sign of good judgment, not a failure.

Q4. Which of the following is a valid stop condition?

Select one:
Show answer and explanation

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

Explanation: This is a serious violation of authority and a clear stop condition.

Q5. True or false: Personal AI boundaries should be fixed and never changed.

Select one:
Show answer and explanation

Correct answer: False.

Explanation: Boundaries should be reviewed and updated as you gain experience, as technology changes, and as your context evolves.

Q6. In the integrated decision sequence, what is the role of escalation?

Select one:
Show answer and explanation

Correct answer: b) To involve more senior humans when a case exceeds routine parameters.

Explanation: Escalation ensures that complex or risky cases receive appropriate human attention.

Q7. Which of the following is a component of ORBIT?

Select one:
Show answer and explanation

Correct answer: d) All of the above

Explanation: ORBIT stands for Opportunities, Risks, Beneficiaries and Burden Bearers, Intervention and Oversight, Transparency and Accountability.

Q8. A recommendation to "redesign" means:

Select one:
Show answer and explanation

Correct answer: b) Change the AI role, safeguards, or scope to address issues.

Explanation: Redesign means revising the approach to make it responsible, not necessarily abandoning it.

17. Assignment 1 Journal Connection

The following prompts are designed to help you reflect on integrated decision‑making and personal boundaries for your Assignment 1 journal.

📓 Reflective prompts
  1. How has your understanding of responsible AI use changed from the beginning of Unit 1 to the end of Unit 3?
  2. Describe a situation where you would exercise responsible refusal. What would be your reasons?
  3. What personal AI boundaries have you established, and why?
  4. Reflect on the relationship between the AI Paradox, ORBIT, and the integrated decision sequence. How do they work together?

Note: These prompts are for your own reflection and can be used as material for your journal entries.

Show reflection planning checklist
  • ☐ I have considered how my thinking has evolved.
  • ☐ I have identified a specific situation where I might refuse AI.
  • ☐ I have drafted personal boundaries.
  • ☐ I have connected the frameworks (AI Paradox, ORBIT, decision sequence).
  • ☐ I have reflected on what I have learned and how it applies to my own context.

18. Key Takeaways

19. Key Terms

Integrated responsible‑use decision
A comprehensive judgment that synthesizes purpose, suitability, risk, oversight, transparency, accountability, escalation, and stop conditions.
Responsible refusal
The decision to decline or limit AI use when conditions are not appropriate, based on legitimate concerns.
Personal AI boundaries
An individual's self‑imposed guidelines for when and how they use generative AI, reflecting their values and context.
Allowed recommendations
The set of responsible outcomes from the decision process: proceed, proceed with safeguards, pilot/limit, redesign, use non‑AI, obtain specialist review, defer, or do not proceed.
Residual uncertainty
The remaining unknowns or risks after safeguards have been applied, which should be documented and monitored.
AI Paradox (revisited)
The insight that AI's capabilities create both opportunities and risks, which must be balanced with context and safeguards.
ORBIT (revisited)
A framework for structured responsible‑use analysis: Opportunities, Risks, Beneficiaries and Burden Bearers, Intervention and Oversight, Transparency and Accountability.
Stop condition
A predefined trigger that halts the AI process to prevent harm.
Escalation
The process of involving more senior or specialist humans when a case exceeds routine parameters.

20. Looking Ahead

Congratulations on completing Unit 3! You have built a strong foundation for responsible AI use. You now understand how to define problems, assess suitability, evaluate risks, design oversight, and make integrated decisions.

Next, you will move to Module 2: Prompt Engineering and Human‑AI Collaboration, beginning with Unit 4: Prompt Design Fundamentals. There, you will learn to craft effective prompts that leverage AI's capabilities while maintaining control and quality. The skills you have developed in Unit 3 will inform how you design prompts that are responsible and aligned with human purposes.

Before moving on, complete the Unit 3 Consolidation Exercise and the Cumulative Project to reinforce your learning. You may also want to revisit your personal AI boundaries and update them as you gain more experience.

🎉 End of Module 1

You have completed the foundational module of COMP 327. You are now equipped to approach generative AI with a critical, responsible, and human‑centred mindset.

Continue to Unit 3 Consolidation Return to Unit 3 Home

21. Sources & Further Reading

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

👤