COMP 3270 / Module 2 / Unit 6 / Lesson 5

Lesson 6.5 · Week 4 · Reading: Chapter 6

PARTNER Responsibility Matrix and Module Integration

Apply the author-approved PARTNER framework, integrate PCAFE-V and SCOPE, define review and stop gates, and complete Assignment 2 preparation.

Suggested time: 90–120 minutes Final lesson in Unit 6 Completes Module 2

Table of Contents

1. Lesson Overview

This final lesson in Unit 6 brings together all the concepts you have developed about human-AI collaboration. You will apply the PARTNER framework to create a complete responsibility matrix for a bounded AI-supported workflow. You will also integrate PCAFE-V (from Unit 4) and SCOPE (from Unit 5) into a single, coherent workflow record.

You will learn to define review gates (where human judgment must be applied before proceeding) and stop gates (conditions under which the workflow must be paused or terminated). The lesson concludes with preparation for Assignment 2: Prompt Engineering Portfolio, which is completed in Unit 6.

Why this matters: A responsible human-AI workflow is not just a collection of tools and prompts. It is a system of accountabilities — clear roles, meaningful oversight, and explicit conditions for proceeding, pausing, or stopping. PARTNER provides a structured way to map these accountabilities so that no AI system exercises final authority over consequential decisions.

2. Learning Outcomes

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

  1. Explain the purpose and operational elements of the author-approved PARTNER framework.
  2. Allocate workflow tasks among the user, AI system, specialist reviewer, decision-maker, action authority, and stop authority.
  3. Build a complete responsibility matrix for a bounded AI-supported workflow.
  4. Integrate PCAFE-V, SCOPE, and PARTNER into a single workflow record.
  5. Define review gates that require evidence, alternatives, uncertainty, and reasons.
  6. Define stop gates, escalation pathways, fallback options, and appeal mechanisms.
  7. Complete the final self-review of Assignment 2 and document AI use appropriately.
  8. Identify personal responsible-use boundaries and prepare for Module 3.

3. Before You Begin

4. Retrieval Practice

These questions draw on Lessons 6.1–6.4. Answer each before reading the feedback.

Question R1 (AI roles)

What is the difference between an AI system acting as a recommender and acting as an agent?

Show answer

Answer: A recommender suggests options or scores that a human then evaluates and decides upon. An agent can take actions in the world on behalf of a user (e.g., sending messages, making purchases, modifying files) without requiring explicit human approval for each action. Agents therefore require much stricter human oversight and stop conditions.

Question R2 (Meaningful oversight)

List four conditions that make human review meaningful rather than nominal.

Show answer
  • Informed: The reviewer has the necessary information and evidence.
  • Competent: The reviewer has the relevant expertise or training.
  • Independent: The reviewer can challenge the AI without pressure.
  • Empowered: The reviewer has the authority to revise, reject, escalate, or stop.

Question R3 (Automation bias)

Define automation bias and give one example of how it can undermine meaningful oversight.

Show answer

Answer: Automation bias is the tendency to over-rely on automated outputs, even when contrary evidence exists. Example: A manager accepts an AI-generated performance ranking without reviewing the underlying criteria, missing a systematic error that disadvantages certain employees.

Question R4 (Responsibility laundering)

What is responsibility laundering, and why is it problematic?

Show answer

Answer: Responsibility laundering occurs when a person or organisation attempts to transfer responsibility for a decision or action to the AI system. It is problematic because AI systems cannot exercise moral, legal, professional, or institutional authority. Responsibility must remain with identifiable humans or institutions.

5. Core Content

5.1 The Purpose of PARTNER

The PARTNER framework is the primary tool for human-AI responsibility allocation in COMP 327. Its purpose is to ensure that every task in an AI-supported workflow is assigned to an appropriate actor, with clear authority, accountability, and safeguards.

Operational definition (placeholder)

[INSERT AUTHOR-APPROVED PARTNER EXPANSION]. Until the expansion is confirmed, treat PARTNER operationally as a framework addressing the allocation of tasks, authority, review, decision, action, correction, appeal, remedy, and stop conditions across human and AI actors.

PARTNER builds on the role classifications and oversight concepts from Lessons 6.1–6.4. It does not replace PCAFE-V or SCOPE; it complements them by focusing specifically on who does what, who decides, who is accountable, and when the workflow must stop.

Key principle: AI systems may assist, analyze, draft, recommend, simulate, or critique. They must not be the final decision-maker, the accountable authority, or the remedy owner for consequential matters.

5.2 Actors in the Responsibility Matrix

A complete PARTNER matrix identifies the following actors for each task or stage:

Table 1. Actors in a PARTNER responsibility matrix
Actor Role Example
User The person initiating or operating the workflow. A student using an AI tool to draft a literature review.
AI system The AI tool performing a bounded task (drafting, summarizing, retrieving, scoring, etc.). A large language model generating a summary.
Specialist reviewer A person with relevant expertise who checks the AI output for accuracy, bias, or suitability. A subject-matter expert or accessibility reviewer.
Decision-maker The person authorized to make the final decision based on the reviewed output. A manager, instructor, or committee chair.
Action authority The person authorized to execute or publish the outcome. An administrator who sends the final communication.
Correction authority The person responsible for correcting errors and handling appeals. A designated appeals officer or supervisor.
Stop authority The person empowered to pause or terminate the workflow if stop conditions are met. A senior manager or safety officer.

5.3 Review Gates and Stop Gates

Two types of gates are essential for responsible human-AI collaboration:

Common pitfall: Treating review as a formality. A review gate that only asks "Do you approve?" is not meaningful. A meaningful review gate asks: What is the evidence? What alternatives were considered? What uncertainty remains? Why is this the appropriate decision?

6. The PARTNER Framework

Below are the operational elements of PARTNER (using placeholders until the approved expansion is supplied). Each element addresses a distinct dimension of human-AI responsibility allocation.

Table 2. Operational elements of PARTNER (placeholder)
Element Definition (placeholder) Guiding questions Common mistake
[PARTNER ELEMENT 1] [INSERT AUTHOR-APPROVED DEFINITION] Who initiates the task? What is the purpose? Skipping purpose definition and moving directly to tool use.
[PARTNER ELEMENT 2] [INSERT AUTHOR-APPROVED DEFINITION] What is the AI's role? What are its limits? Assigning decision authority to an AI system.
[PARTNER ELEMENT 3] [INSERT AUTHOR-APPROVED DEFINITION] Who reviews? What evidence must they consider? Treating review as a formality or rubber-stamp.
[PARTNER ELEMENT 4] [INSERT AUTHOR-APPROVED DEFINITION] Who decides? Who has final authority? Allowing the AI to make the final decision.
[PARTNER ELEMENT 5] [INSERT AUTHOR-APPROVED DEFINITION] Who acts? Who is accountable for the outcome? Confusing recommendation with action.
[PARTNER ELEMENT 6] [INSERT AUTHOR-APPROVED DEFINITION] How are errors corrected? How are appeals handled? Assuming AI can remedy its own errors.
[PARTNER ELEMENT 7] [INSERT AUTHOR-APPROVED DEFINITION] When must the workflow stop? Who can stop it? Omitting stop conditions or failing to empower a stop authority.
Operational note: Until the author-approved expansion is supplied, use the elements above as a checklist. When the expansion is provided, replace the placeholders and use the approved terminology consistently.

7. Building the Responsibility Matrix

A responsibility matrix maps each workflow task to an actor (or actors) and specifies the authority and accountability at each stage. Use the template below.

Table 3. PARTNER responsibility matrix template
Workflow stage AI role Human reviewer Decision-maker Action authority Correction authority Stop authority Stop condition
Problem definition None User User User User User Purpose unclear or unauthorized
Source selection Recommender User User User User User Sources are biased or unauthorized
Drafting Assistant User User User User User Output contains harmful or fabricated content
Verification None User or specialist User User User User Claims cannot be verified
Final approval None Specialist reviewer Decision-maker Action authority Correction authority Stop authority Evidence insufficient or authority unclear
Publication / action None N/A N/A Action authority Correction authority Stop authority Error detected or complaint received
Correction / remedy None N/A Correction authority Correction authority Correction authority Stop authority Remedy cannot be provided

How to use the matrix:

  1. List every stage in your workflow, from problem definition to remedy.
  2. For each stage, identify the AI's role (if any).
  3. Assign a human reviewer, decision-maker, action authority, correction authority, and stop authority.
  4. Specify one or more stop conditions for each stage.
  5. Review the matrix to ensure that no consequential authority rests with the AI system.

8. Integrating PCAFE-V, SCOPE, and PARTNER

Module 2 has introduced three complementary frameworks. Together they provide a complete approach to responsible prompt engineering and human-AI collaboration.

Table 4. Complementary frameworks in Module 2
Framework Focus Key questions
PCAFE-V Prompt design and verification What is the purpose, context, audience, format, evidence, and verification plan?
SCOPE Information literacy and source evaluation What are the sources, context, omissions, purpose, and evidence?
PARTNER Human-AI responsibility allocation Who does what? Who reviews, decides, acts, corrects, and stops?
Integration workflow:
  1. Use PCAFE-V to design the prompt and plan verification.
  2. Use SCOPE to evaluate sources and identify omissions.
  3. Use PARTNER to allocate tasks and authority across human and AI actors.
  4. Define review gates where human judgment must be applied.
  5. Define stop gates and assign a stop authority.
  6. Document the entire workflow in a single record for Assignment 2.

9. Review Gates and Stop Gates

Gates are decision points. Review gates require human judgment before proceeding; stop gates require pausing or terminating the workflow.

9.1 Review Gates

At each review gate, the reviewer should be able to answer:

9.2 Stop Gates

Stop gates should be defined in advance. Common stop conditions include:

Stop authority must be named. It is not enough to say "the workflow should stop if X occurs." Specify who is responsible for detecting the condition and who has the authority to pause or terminate the workflow.

10. Worked Example: AI-Assisted Literature Review

Context: A graduate student uses an AI tool to help draft a literature review for a course paper. The AI is used to summarize articles, identify themes, and suggest additional sources.

Step 1: Define the purpose and AI role (PCAFE-V)

Step 2: Evaluate sources and omissions (SCOPE)

Step 3: Build the PARTNER responsibility matrix

Table 5. PARTNER matrix for AI-assisted literature review
Stage AI role Reviewer Decision-maker Action authority Correction authority Stop authority Stop condition
Topic selection None Student Student Student Student Student Topic unclear or unauthorized
Source search Recommender Student Student Student Student Student Sources biased or inaccessible
Summarization Assistant Student Student Student Student Student Summary inaccurate or fabricated
Theme identification Critic Student Student Student Student Student Themes do not reflect source evidence
Final draft None Student Student Student Student Student Claims unverified or authorship unclear
Submission None Instructor (if requested) Student Student Student Student Disclosure missing or integrity concern

Step 4: Define review gates and stop gates

Show key insights from this example
  • The AI's role is bounded to assistance and critique; it does not decide or act.
  • The student retains all decision, action, correction, and stop authority.
  • Review gates require active verification, not passive approval.
  • Stop conditions are specific and actionable.

11. Guided Learning Activities

Activity 11.1: Build Your PARTNER Matrix
  • Purpose: To apply the PARTNER framework to a workflow of your choice.
  • Learning outcome: Build a complete responsibility matrix.
  • Estimated time: 30–40 minutes.
  • Instructions: Choose a bounded AI-supported workflow (e.g., drafting a report, generating a study guide, summarizing a meeting). Using the template from Section 7, create a PARTNER matrix with at least five stages. For each stage, identify the AI role, reviewer, decision-maker, action authority, correction authority, stop authority, and at least one stop condition.
Show example matrix (study guide workflow)
  • Stage 1: Define learning objectives — AI: None. Reviewer: Student. Decision: Student. Action: Student. Correction: Student. Stop: Student. Stop condition: Objectives unclear.
  • Stage 2: Generate practice questions — AI: Assistant. Reviewer: Student. Decision: Student. Action: Student. Correction: Student. Stop: Student. Stop condition: Questions contain errors or bias.
  • Stage 3: Review questions against course material — AI: None. Reviewer: Student. Decision: Student. Action: Student. Correction: Student. Stop: Student. Stop condition: Material unavailable or misaligned.
  • Stage 4: Finalize study guide — AI: None. Reviewer: Student. Decision: Student. Action: Student. Correction: Student. Stop: Student. Stop condition: Unverified claims remain.
Activity 11.2: Define Review Gates
  • Purpose: To design meaningful review gates.
  • Learning outcome: Define review gates that require evidence, alternatives, and reasons.
  • Estimated time: 15–20 minutes.
  • Instructions: For the workflow you used in Activity 11.1, identify at least three review gates. For each, write the specific questions the reviewer must answer before proceeding.
Show sample review gate questions
  • What evidence supports this output?
  • What alternatives were considered?
  • What uncertainty remains?
  • Why is this the appropriate decision?
  • Could this disadvantage any group?
  • Is the use of AI and data authorized?
Activity 11.3: Define Stop Gates
  • Purpose: To define stop conditions and assign stop authority.
  • Learning outcome: Define stop gates, escalation pathways, and fallback options.
  • Estimated time: 15–20 minutes.
  • Instructions: For the same workflow, define at least three stop conditions. For each, specify who is responsible for detecting the condition and who has the authority to pause or terminate the workflow. Describe what happens after the stop (escalation, fallback, correction).
Show sample stop conditions
  • Stop condition: AI output contains fabricated citations. Detector: Student. Stop authority: Student. After stop: Remove fabricated content, verify sources, restart drafting.
  • Stop condition: Reviewers unavailable. Detector: Student. Stop authority: Student. After stop: Pause workflow until reviewer available or choose non-AI alternative.
  • Stop condition: Use of AI is not authorized for this task. Detector: Student. Stop authority: Student. After stop: Discontinue AI use, complete task without AI.
Activity 11.4: Integrate PCAFE-V, SCOPE, and PARTNER
  • Purpose: To create a single integrated workflow record.
  • Learning outcome: Integrate the three frameworks.
  • Estimated time: 20–25 minutes.
  • Instructions: Create a one-page integrated workflow record for your chosen task. Include: (a) PCAFE-V prompt design summary, (b) SCOPE source evaluation summary, (c) PARTNER responsibility matrix, (d) review gates, and (e) stop gates.
Show integrated workflow template
  1. Purpose and AI role (PCAFE-V): [one sentence]
  2. Sources and omissions (SCOPE): [one sentence]
  3. Responsibility matrix (PARTNER): [table or bullet list]
  4. Review gates: [list]
  5. Stop gates: [list]
  6. Verification plan: [how claims will be checked]
  7. Disclosure plan: [how AI use will be documented]

12. Exercises

Exercise 1: Identify the Actors
  • Learning outcome: Identify actors in a responsibility matrix.
  • Estimated time: 8 minutes.
  • Task: For an AI-supported customer service workflow, list which actor should fill each role: user, AI system, specialist reviewer, decision-maker, action authority, correction authority, stop authority.
Show sample answer
  • User: Customer service representative.
  • AI system: Drafts responses to customer inquiries.
  • Specialist reviewer: Senior representative or subject-matter expert.
  • Decision-maker: Customer service representative (with senior approval for complex cases).
  • Action authority: Customer service representative (sends the response).
  • Correction authority: Senior representative or quality assurance officer.
  • Stop authority: Customer service manager.
Exercise 2: Design a Review Gate
  • Learning outcome: Design a meaningful review gate.
  • Estimated time: 10 minutes.
  • Task: For the AI-supported customer service workflow, design one review gate. Write the specific questions the reviewer must answer before proceeding.
Show sample review gate

Review gate: Before sending an AI-drafted response, the customer service representative must confirm:

  • Does the response accurately address the customer's inquiry?
  • Is the tone appropriate and respectful?
  • Are there any factual claims that need verification?
  • Does the response include any personal or confidential information that should not be shared?
  • Is the response consistent with company policy?
Exercise 3: Define a Stop Condition
  • Learning outcome: Define a stop condition and assign stop authority.
  • Estimated time: 8 minutes.
  • Task: For the customer service workflow, define one stop condition and specify who detects it and who has the authority to stop the workflow.
Show sample answer

Stop condition: The AI-drafted response contains information that is factually incorrect or could be harmful to the customer.

Detector: Customer service representative.

Stop authority: Customer service manager.

After stop: The response is not sent. The representative escalates to the manager, who reviews the case and decides whether to correct the response or handle the inquiry personally.

Exercise 4: Integrate the Frameworks
  • Learning outcome: Integrate PCAFE-V, SCOPE, and PARTNER.
  • Estimated time: 12 minutes.
  • Task: Write a one-paragraph summary showing how PCAFE-V, SCOPE, and PARTNER would work together for an AI-supported research task.
Show sample answer

Sample integration: PCAFE-V is used to design the prompt and plan verification (purpose, context, audience, format, evidence, verification). SCOPE is used to evaluate sources and identify omissions (sources, context, omissions, purpose, evidence). PARTNER is used to allocate responsibility (who reviews, decides, acts, corrects, and stops). Together, they ensure that the workflow is purposeful, well-sourced, and accountable.

Exercise 5: Assignment 2 Self‑Review
  • Learning outcome: Complete the final self-review of Assignment 2.
  • Estimated time: 15 minutes.
  • Task: Review your Prompt Engineering Portfolio against the Assignment 2 checklist in the Unit 6 landing page. List three strengths and three areas for improvement.
Show sample self-review

Strengths:

  • Clear documentation of prompt revisions.
  • Strong verification of claims and sources.
  • Explicit disclosure of AI use.

Areas for improvement:

  • Need more detail on human review and correction.
  • Need to add more reflection on automation bias.
  • Need to strengthen the PARTNER matrix.

13. Homework Problems

Formative · Ungraded · Recommended preparation for Assignment 2

Homework 1: PARTNER Element Matching
  • Estimated time: 5 minutes.
  • Task: Match each operational PARTNER element (placeholder) to its guiding question.
Show sample matching
  • Element 1 → Who initiates the task and why?
  • Element 2 → What is the AI's role and what are its limits?
  • Element 3 → Who reviews and what evidence must they consider?
  • Element 4 → Who decides and who has final authority?
  • Element 5 → Who acts and who is accountable?
  • Element 6 → How are errors corrected and appeals handled?
  • Element 7 → When must the workflow stop and who can stop it?
Homework 2: Build a Responsibility Matrix
  • Estimated time: 10 minutes.
  • Task: Create a PARTNER responsibility matrix for a simple AI-supported task (e.g., drafting an email, summarizing an article). Include at least four stages.
Show sample matrix
  • Stage 1: Define purpose — AI: None. Reviewer: User. Decision: User. Action: User. Correction: User. Stop: User. Stop: Purpose unclear.
  • Stage 2: Draft content — AI: Assistant. Reviewer: User. Decision: User. Action: User. Correction: User. Stop: User. Stop: Content harmful or inaccurate.
  • Stage 3: Verify claims — AI: None. Reviewer: User. Decision: User. Action: User. Correction: User. Stop: User. Stop: Claims unverified.
  • Stage 4: Send/publish — AI: None. Reviewer: User. Decision: User. Action: User. Correction: User. Stop: User. Stop: Authorization missing.
Homework 3: Identify Human‑Only Work
  • Estimated time: 5 minutes.
  • Task: List three activities that should remain human-only in any AI-supported workflow, and explain why.
Show sample answer
  • Final decision-making: Requires judgment, accountability, and authority that AI cannot exercise.
  • Ethical judgment: Requires moral reasoning and contextual understanding.
  • Relationship and care: Requires empathy, trust, and human connection.
Homework 4: Automation‑Bias Countermeasures
  • Estimated time: 8 minutes.
  • Task: For the customer service workflow, propose three countermeasures to reduce automation bias.
Show sample answer
  • Train representatives on the limitations of AI and the importance of independent judgment.
  • Require representatives to document their reasoning for accepting or modifying AI-drafted responses.
  • Randomly audit a sample of responses to check for over-reliance on AI.
Homework 5: Review Gate Design
  • Estimated time: 8 minutes.
  • Task: Design a review gate for an AI-supported workflow. List at least four questions the reviewer must answer.
Show sample answer

Review gate questions:

  • What evidence supports this output?
  • What alternatives were considered?
  • What uncertainty remains?
  • Why is this the appropriate decision?
  • Could this disadvantage any group?
Homework 6: Stop Gate Design
  • Estimated time: 8 minutes.
  • Task: Define two stop conditions for an AI-supported workflow, and specify who detects each condition and who has stop authority.
Show sample answer
  • Stop condition 1: AI output contains fabricated information. Detector: User. Stop authority: Supervisor. After stop: Correct information and re-verify.
  • Stop condition 2: A qualified reviewer is unavailable. Detector: User. Stop authority: Supervisor. After stop: Pause until reviewer available or use non-AI alternative.
Homework 7: Integrate the Frameworks
  • Estimated time: 8 minutes.
  • Task: Write a short paragraph explaining how PCAFE-V, SCOPE, and PARTNER complement each other in a responsible AI workflow.
Show sample answer

Answer: PCAFE-V guides the design of the prompt and the verification plan. SCOPE ensures that sources are evaluated and omissions identified. PARTNER allocates responsibility across human and AI actors, defining who reviews, decides, acts, corrects, and stops. Together, they create a workflow that is purposeful, well-sourced, and accountable.

Homework 8: Transfer
  • Estimated time: 10 minutes.
  • Task: Apply the PARTNER framework to a domain outside your usual area (e.g., healthcare, law, education, public administration). Describe the key actors and one stop condition.
Show sample answer

Domain: Healthcare — AI-assisted triage.

Actors: Patient (user), AI (recommender), triage nurse (reviewer), physician (decision-maker), physician (action authority), quality assurance officer (correction authority), chief of emergency medicine (stop authority).

Stop condition: If the AI recommends a lower triage priority for a patient with symptoms of a serious condition, and the nurse cannot verify or override the recommendation, the workflow stops and escalates to the physician immediately.

14. Mini‑Project: Integrated PARTNER Workflow Record

Project title: Integrated PARTNER Workflow Record for a Bounded AI-Supported Task
  • Purpose: To produce a complete, integrated workflow record that demonstrates your ability to apply PCAFE-V, SCOPE, and PARTNER together.
  • Learning outcomes: Build a responsibility matrix; define review and stop gates; integrate the three frameworks; document AI use and verification.
  • Estimated effort: 90–120 minutes.
  • Context choices: Choose a bounded task from your studies or professional practice (e.g., drafting a report, creating a study guide, summarizing a meeting, preparing a presentation).

Required deliverable: A structured workflow record (approximately 800–1,200 words or equivalent) with the following sections:

  1. Task title and description
  2. Legitimate purpose
  3. PCAFE-V summary (purpose, context, audience, format, evidence, verification)
  4. SCOPE summary (sources, context, omissions, purpose, evidence)
  5. AI role classification (tool, assistant, critic, simulator, recommender, agent)
  6. PARTNER responsibility matrix (all actors and stages)
  7. Review gates (at least three)
  8. Stop gates (at least three, with stop authority named)
  9. Escalation and fallback
  10. Verification plan
  11. Disclosure plan
  12. Residual uncertainty and limitations

Privacy and respectful-use boundaries: Use only synthetic, public, or authorized information. Do not enter personal, confidential, or sensitive data. Do not use AI to make consequential decisions or take actions on your behalf.

Show solution architecture
  1. Define the task and purpose clearly.
  2. Design the prompt and verification plan using PCAFE-V.
  3. Evaluate sources and omissions using SCOPE.
  4. Classify the AI role.
  5. Build the PARTNER responsibility matrix.
  6. Define review gates and stop gates.
  7. Specify escalation, fallback, verification, and disclosure.
  8. Document residual uncertainty.
Show illustrative partial example (different context)

Context: AI-assisted meeting summarization.

  • Purpose: To produce an accurate summary of a team meeting for those who could not attend.
  • AI role: Assistant — transcribes audio and drafts a summary.
  • PCAFE-V: Prompt specifies format (bullet points), audience (team members), and verification (check against recording).
  • SCOPE: Sources are the meeting recording; omissions may include non-verbal cues or side conversations.
  • PARTNER matrix: User (team lead) reviews, decides, acts, corrects, and stops. AI does not decide or act.
  • Review gate: Before sharing the summary, the team lead verifies key decisions and action items against the recording.
  • Stop gate: If the summary misattributes a decision or omits a critical action item, the team lead pauses and corrects before distribution.

15. Self‑Check Quiz

These questions are formative and ungraded. Use them to check your understanding.

Quiz 1 — Multiple choice

What is the primary purpose of the PARTNER framework?

Select one:
Show answer and explanation

Correct answer: B.

Explanation: PARTNER focuses on human-AI responsibility allocation. PCAFE-V focuses on prompt design (A); SCOPE focuses on source evaluation (C). PARTNER does not replace human judgment (D).

Quiz 2 — Multiple response

Which of the following are actors in a PARTNER responsibility matrix? Select all that apply.

Select all that apply:
Show answer and explanation

Correct answers: All of the above (A–E).

Explanation: A complete responsibility matrix includes the user, AI system, specialist reviewer, decision-maker, action authority, correction authority, and stop authority.

Quiz 3 — Short answer

What is the difference between a review gate and a stop gate?

Show answer and explanation

Answer: A review gate is a point where human judgment must be applied before proceeding. A stop gate is a condition under which the workflow must be paused, escalated, or terminated.

Quiz 4 — True or false

An AI system can be assigned stop authority if it is designed to detect errors.

Select one:
Show answer and explanation

Answer: False.

Explanation: Stop authority must remain with an identifiable human or institution. AI can flag potential issues, but the decision to stop must be made by a person with the appropriate authority and accountability.

Quiz 5 — Short answer

How do PCAFE-V, SCOPE, and PARTNER complement each other?

Show answer and explanation

Answer: PCAFE-V guides prompt design and verification. SCOPE evaluates sources and omissions. PARTNER allocates responsibility across human and AI actors. Together, they ensure that the workflow is purposeful, well-sourced, and accountable.

Quiz 6 — Multiple choice

Which of the following is a meaningful review gate question?

Select one:
Show answer and explanation

Correct answer: B.

Explanation: Meaningful review gates require evidence, alternatives, uncertainty, and reasons — not just passive approval.

Quiz 7 — Short answer

List three stop conditions that could apply to an AI-supported workflow.

Show answer and explanation
  • The purpose is unclear, unauthorized, or has changed materially.
  • The output contains harmful, fabricated, or biased content.
  • A qualified reviewer or stop authority is unavailable.

Quiz 8 — Short answer

What should be included in an integrated workflow record for Assignment 2?

Show answer and explanation

Answer: An integrated workflow record should include: purpose and AI role (PCAFE-V), source evaluation and omissions (SCOPE), responsibility matrix (PARTNER), review gates, stop gates, verification plan, and disclosure plan.

16. Assignment 2 Connection

This lesson completes your preparation for Assignment 2: Prompt Engineering Portfolio (15% of final grade, submitted at the end of Module 2). Use the integrated workflow record from your mini-project as a key component of your portfolio.

Final Assignment 2 checklist
  • ☐ Purpose, task, audience, format, criteria, and verification are explicit.
  • ☐ Initial and revised prompts and outputs are preserved.
  • ☐ Context is necessary, minimized, and authorized.
  • ☐ Sources and evidence boundaries are documented.
  • ☐ The task is decomposed into reviewable stages where appropriate.
  • ☐ AI and human roles are clearly distinguished.
  • ☐ Meaningful review, approval, escalation, fallback, and stopping are operational.
  • ☐ Claims and sources are verified as required.
  • ☐ Corrections and final human decisions are explained.
  • ☐ AI use is documented and disclosed according to assignment instructions.

Do not submit this lesson's mini-project as your Assignment 2. It is preparatory work. Your portfolio should be a polished, reflective document that demonstrates your development as a responsible prompt engineer.

17. Assignment 1 Journal Connection

Reflect on the following prompts for your Unit 6 journal entry. Do not provide model responses — write your own reflections.

Reflection prompts for Lesson 6.5
  1. How has your understanding of human-AI collaboration changed after building a PARTNER matrix?
  2. What was the most challenging part of defining review gates and stop gates?
  3. How do you now think about the relationship between prompt design, source evaluation, and responsibility allocation?
  4. What responsible-use boundaries have you established for your own AI use?
  5. How will you carry forward the lessons from Module 2 into Module 3?
Show reflection planning checklist
  • ☐ I have reflected on the role of the PARTNER framework.
  • ☐ I have considered how review gates change the quality of human oversight.
  • ☐ I have identified a personal boundary or commitment for responsible AI use.
  • ☐ I have connected Module 2 learning to future modules.

18. Key Takeaways

19. Key Terms

PARTNER framework
[INSERT AUTHOR-APPROVED EXPANSION]. Operationally, a framework for allocating tasks, authority, review, decision, action, correction, appeal, remedy, and stop conditions across human and AI actors.
Responsibility matrix
A table mapping each workflow stage to the actors responsible for review, decision, action, correction, and stopping.
Review gate
A point in the workflow where human judgment must be applied before proceeding, requiring evidence, alternatives, uncertainty, and reasons.
Stop gate
A predefined condition under which the workflow must be paused, escalated, or terminated, with a named stop authority.
Stop authority
The person empowered to pause or terminate the workflow if stop conditions are met.
Correction authority
The person responsible for correcting errors and handling appeals.
Action authority
The person authorized to execute or publish the outcome of the workflow.
Decision-maker
The person authorized to make the final decision based on the reviewed output.
Specialist reviewer
A person with relevant expertise who checks the AI output for accuracy, bias, or suitability.
Integrated workflow record
A single document combining PCAFE-V, SCOPE, and PARTNER to provide a complete account of purpose, sources, roles, authority, review, and stop conditions.
Responsibility laundering
Attempting to transfer responsibility for a decision or action to the AI system.

20. Looking Ahead

You have now completed Unit 6 and Module 2. You are ready to move to Module 3: Responsible and Ethical AI Use, beginning with Unit 7: Bias, Fairness, and Inclusion.

In Module 3, you will build on the responsibility allocation and oversight skills you developed here. You will examine how bias enters AI-supported workflows, how to assess fairness, and how to design systems that are inclusive, accessible, and accountable.

The PARTNER framework will continue to be relevant as you evaluate who is responsible for identifying and addressing bias, who participates in fairness reviews, and who holds the authority to stop or remedy an unfair system.

Continue to Module 3: Responsible and Ethical AI Use Return to Unit 6 Home
👤