Lesson 6.2 | Task Allocation and Human Authority

1. Lesson Overview

After understanding AI roles and complementarity (Lesson 6.1), we now turn to the practical work of deciding what to delegate and who holds authority. This lesson introduces a structured approach to decomposing workflows, allocating tasks among humans and AI, and mapping different types of human authority. You will learn to distinguish tasks that can be AI-supported from those that require ordinary human review, specialist review, or must remain entirely human.

We also introduce the concept of authority types—process ownership, information ownership, authorship, professional authority, review authority, decision authority, action authority, appeal authority, correction authority, and stop authority—and show how to create a human-AI responsibility matrix. This lesson is the second step in building a complete PARTNER collaboration design (the A and R of PARTNER).

Why this matters: Many AI failures are not failures of the model—they are failures of allocation. When decision authority, action authority, or stop authority is implicitly delegated to AI, accountability collapses. This lesson gives you the vocabulary and tools to allocate tasks and authority explicitly.

2. Learning Outcomes

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

Decompose a workflow into its constituent tasks and decisions.
Allocate each activity to one of seven categories: human-led, AI-supported, AI-generated + human-reviewed, specialist-reviewed, human-only, prohibited, escalation required.
Identify activities that must not be delegated to AI.
Distinguish process ownership, information ownership, authorship, professional authority, review authority, decision authority, action authority, appeal authority, correction authority, and stop authority.
Map authority roles to a specific workflow and identify gaps or overlaps.
Create a human-AI responsibility matrix for a bounded use case.
Apply the A (Allocate Tasks) and R (Retain Human Authority) elements of PARTNER.

3. Before You Begin

Assigned reading: Chapter 6, sections 6.4–6.6 (Task Allocation, Authority, and Responsibility).
Prerequisite knowledge: You should be able to classify AI roles (Lesson 6.1) and apply SCOPE and PCAFE-V to a workflow.
Optional AI tool: You may use an AI system to explore workflows, but do not submit AI-generated answers as your own.
Privacy reminder: Do not share confidential or institutional information.
Academic integrity: All submitted work must be your own analysis. Document any AI assistance.
Estimated preparation time: 30 min for reading and retrieval.

4. Retrieval Practice: Concepts from Lesson 6.1

Recall concepts from Units 4–5 and Lesson 6.1 to answer these questions.

Question 1: In Lesson 6.1, you learned that an AI role does not confer authority. What does this mean for task allocation?

Show answer

It means that even if an AI system is classified as a Recommender or Agent, it cannot make final decisions, approve actions, or assume accountability. Task allocation must reserve all decisions with consequences for human authority.

Question 2: How does SCOPE help you decide which tasks to allocate to AI?

Show answer

SCOPE (Source, Context, Output, Permission, Escalation) provides a structured way to define the boundaries of AI involvement. It helps you determine what sources the AI can use, what context it has, what outputs it produces, what permissions it needs, and when to escalate—all of which inform allocation decisions.

Question 3: What is the difference between an AI Assistant and an AI Agent in terms of task allocation?

Show answer

An Assistant supports human tasks but does not act independently or maintain state across sessions. An Agent can pursue multi-step goals, use tools, and may perform external actions. Task allocation for an Agent requires more restrictive permissions and oversight, and many actions (e.g., sending, modifying, approving) should be reserved for human authority.

5. Opening Scenario: "The 'Human-in-the-Loop' Illusion"

📌 Scenario — The "Human-in-the-Loop" Illusion

A municipal government deploys an AI system to review building-permit applications. The system classifies applications as low, medium, or high risk, and generates a recommended decision (approve, request changes, or deny). A junior clerk is designated as the "human in the loop" and is asked to review and approve the AI's recommendations. However, the clerk:

Does not own the application data;
Lacks architectural or engineering expertise;
Cannot reject the recommendation without supervisor approval;
Cannot stop the workflow if they disagree.

Questions:

Does the clerk have meaningful authority? What types of authority are missing?
What tasks should remain human-only in this context?
How could the workflow be redesigned to allocate tasks and authority more responsibly?

Show guided analysis

1. Authority analysis: The clerk lacks decision authority (cannot reject), stop authority, and information ownership. The review is nominal because the clerk does not have the expertise or the power to override the AI. This is a classic case of a "human-in-the-loop" that is actually human-on-the-loop or even human-out-of-the-loop in practice.

2. Human-only tasks: Decisions that require professional judgment (e.g., interpretation of building codes, safety considerations, unusual cases) should remain human-only. Also, any decision that has significant consequences for applicants or public safety must be made by a qualified professional, not delegated to AI or a junior clerk.

3. Redesign: Allocate the AI to an Assistant or Recommender role—it can flag issues and suggest classifications, but a qualified building inspector must review the full application, have the authority to reject the AI's recommendation, and stop the process if needed. The clerk could handle administrative tasks but not final decisions.

6. Decomposing Workflows

Before allocating tasks, you must decompose the workflow into its constituent stages. A typical workflow includes:

Problem definition — understanding the need, defining success criteria.
Source selection — identifying and gathering relevant information.
Context preparation — organising and structuring the information for the AI.
Generation — having the AI produce draft content, analyses, or options.
Analysis — interpreting the AI output, comparing alternatives, identifying patterns.
Recommendation — proposing a course of action (may be AI-generated or human-made).
Decision — choosing among options.
Action — implementing the decision.
Verification — checking that the action was correct and effective.
Review — evaluating the overall process and outcomes.
Correction — making adjustments if needed.
Appeal — handling challenges or requests for reconsideration.

Why decompose? Without a clear inventory, you risk overlooking critical steps, over-delegating to AI, or under-specifying human oversight.

Section summary

Decomposition identifies every stage in a workflow, from problem definition to appeal.
It prevents over-delegation and under-specification of oversight.
Every stage should be explicitly named and reviewed.

7. Task Allocation Categories

For each activity in the workflow, assign one of the following categories (adapted from Chapter 6):

Human-led — The human defines the task, directs the AI, and makes final decisions; the AI may assist but does not lead.
AI-supported — The AI performs a bounded function under human direction; the human reviews and integrates the output.
AI-generated but human-reviewed — The AI produces a draft or recommendation that requires thorough human review before use.
Specialist-reviewed — The output requires review by a subject-matter expert or professional.
Human-only — The activity must be performed entirely by a human, without AI involvement.
Prohibited — The activity is not permitted for AI under any circumstances (e.g., making final approvals, acting on behalf of a vulnerable person).
Escalation required — If the activity cannot be completed within defined parameters, it must be escalated to a higher authority.

Table 1. Task allocation examples
Activity	Allocation category	Rationale
Generating three draft email responses	AI-generated + human-reviewed	AI drafts, human selects and edits.
Selecting the final candidate for a job	Human-only	Requires professional judgment and accountability.
Flagging unusual patterns in financial data	AI-supported	AI identifies patterns; analyst investigates.
Interpreting complex legal clauses	Specialist-reviewed	Only a qualified lawyer can interpret; AI may assist with summarising.
Issuing a refund without human approval	Prohibited	Financial action requires authorisation.

Section summary

There are seven allocation categories, from human-led to prohibited and escalation required.
Allocation should consider consequence of error, reversibility, expertise required, and accountability.
Not every activity is suitable for AI involvement.

8. Types of Human Authority

Authority is not a single concept. In human-AI collaboration, we must distinguish several types of authority that may be held by different individuals or roles:

Purpose authority — Who defines the legitimate human purpose? (Usually the organisation or client.)
Process ownership — Who is responsible for the overall workflow design and execution?
Information ownership — Who has legitimate control over the data and sources used?
Authorship authority — Who is credited as the author or creator of the final product?
Professional authority — Who has the qualifications to make professional judgments in this domain?
Review authority — Who has the right to examine, critique, and approve or reject AI outputs?
Decision authority — Who makes the final choice among alternatives?
Action authority — Who is authorised to implement the decision (e.g., send an email, update a record, sign a contract)?
Appeal authority — Who hears and decides on appeals or challenges to the decision?
Correction authority — Who can correct errors after the fact?
Stop authority — Who has the power to halt the workflow or AI system if something goes wrong?

In a well-designed collaboration, these authorities are clearly assigned, and no single person should hold all of them without appropriate checks. The presence of a human reviewer does not guarantee that all these authorities exist.

Section summary

Authority has at least eleven distinct types relevant to human-AI collaboration.
Each type should be assigned to a specific role.
A "human reviewer" without decision or stop authority is not meaningful oversight.

9. Human-AI Responsibility Matrix

A responsibility matrix (sometimes called a RACI chart) is a table that maps each workflow activity to the roles responsible, accountable, consulted, and informed. For our purposes, we create a simplified matrix that includes:

Workflow stage / activity
AI role / activity (what the AI does)
Human role / activity (what the human does)
Specialist role (if applicable)
Authority types involved (e.g., decision authority, action authority)
Verification and documentation requirements

The matrix helps visualise who does what, who decides, and who is accountable. It is a key tool for Assignment 2.

Table 2. Simplified responsibility matrix example (drafting a report)
Stage	AI activity	Human activity	Specialist	Authority
Data aggregation	Collect and organise data from sources	Define data sources and quality criteria	—	Information ownership (human)
Draft generation	Generate draft report sections	Review draft, add context and interpretation	—	Review authority (human)
Fact-checking	Flag potential inconsistencies	Verify facts against original sources	Subject-matter expert for complex claims	Decision authority (human)
Final approval	None	Approve final report for distribution	—	Action authority, stop authority (human)

Section summary

A responsibility matrix maps each stage to AI, human, and specialist roles.
It includes authority types and verification requirements.
It is a key visual tool for designing and defending a workflow.

10. PARTNER Walkthrough — Allocate Tasks & Retain Human Authority

In Lesson 6.1 we covered P (Purpose). Now we move to A and R.

A: Allocate Tasks

Guiding questions for task allocation:

Which tasks are bounded and reviewable?
Which tasks require human interpretation or judgment?
Which tasks require professional or community authority?
Which actions must not be automated?
Who bears the verification work?

Use the allocation categories (human-led, AI-supported, etc.) to assign each activity. Consider the consequence of error, reversibility, need for human relationships, and AI permissions.

R: Retain Human Authority

Guiding questions:

Who has legitimate authority to make the final decision?
Can the human reviewer reject the AI output?
Can the workflow be interrupted?
Who is accountable for the final outcome?

Ensure that decision authority, action authority, and stop authority are clearly assigned to qualified humans. Do not allow the AI system to assume these authorities by default.

🧩 Brief application

Consider a workflow where an AI generates a candidate shortlist for a job. Which authority types must be retained by humans? Who should hold each?

Show example

Decision authority: Hiring manager decides final hire.
Review authority: HR and hiring manager review the shortlist.
Action authority: HR sends the offer.
Stop authority: Hiring manager can stop the process if concerns arise.
Information ownership: HR owns candidate data.

Section summary

PARTNER's A guides you to allocate tasks using explicit categories.
PARTNER's R guides you to retain decision, action, and stop authority with humans.
Authority is not implied—it must be assigned.

11. Process Walkthrough: From Workflow to Allocation to Authority Map

We continue the academic task from Lesson 6.1: generating discussion questions. We now decompose the workflow, allocate tasks, and map authority.

Step 1 — Decompose the workflow

Problem definition: Instructor defines learning objectives and desired discussion outcomes.
Source selection: Instructor selects readings and supplementary materials.
Context preparation: Instructor creates a context package with readings, objectives, and criteria.
Generation: AI generates draft questions based on context.
Analysis & recommendation: Instructor reviews drafts, selects promising questions, and requests refinements.
Decision: Instructor finalises the question set.
Action: Instructor publishes questions to the learning platform.
Verification: Instructor checks that questions align with objectives and are suitable for students.
Review: After the discussion, instructor assesses effectiveness and may revise future questions.

Step 2 — Allocate tasks

Table 3. Task allocation for discussion-question workflow
Activity	Allocation	Rationale
Problem definition	Human-led	Instructor defines purpose.
Source selection	Human-led	Instructor chooses materials.
Context preparation	Human-led / AI-supported	Instructor prepares; AI may assist with formatting.
Generation	AI-generated + human-reviewed	AI drafts; instructor reviews and revises.
Analysis & recommendation	Human-led	Instructor evaluates and selects.
Decision	Human-only	Instructor makes final choice.
Action	Human-only	Instructor publishes; AI does not act.
Verification	Human-led	Instructor checks quality.
Review	Human-led	Instructor assesses and refines.

Step 3 — Map human authority

Process owner: Instructor
Information owner: Instructor (readings) and institution (course materials)
Review authority: Instructor
Decision authority: Instructor
Action authority: Instructor
Stop authority: Instructor (can pause or stop use of AI at any time)
Appeal authority: Department head (if students appeal question quality)
Correction authority: Instructor

This walkthrough demonstrates that the instructor retains all consequential authorities, while the AI serves as a drafting assistant with no independent authority.

Walkthrough summary

Decomposition first, then allocation, then authority mapping.
Each stage has an explicit allocation category.
Authority is assigned to specific human roles.

12. Worked Example: "Customer-Support Ticket Triage"

A customer-support team uses an AI system to triage incoming tickets: the AI categorises the issue, suggests a priority level, and drafts a first response. A support agent reviews the ticket, the AI's suggestions, and then responds or escalates.

Decomposition

Ticket receipt
AI categorisation (technical, billing, account, etc.)
AI priority suggestion (high, medium, low)
AI draft response
Agent review of ticket and AI output
Agent decision: respond, escalate, or request more info
Agent action: send response, escalate to specialist, or follow up
Verification: agent checks response before sending
Customer feedback / appeal handling

Allocation

Categorisation, priority, draft response — AI-generated + human-reviewed
Agent review — Human-led (review authority)
Decision to respond, escalate, or follow up — Human-only (decision authority)
Sending response / escalating — Human action (action authority) — AI may assist with drafting but not send without approval.
Handling appeals — Specialist-reviewed (supervisor or specialist team)

Authority map

Process owner: Support team lead
Information owner: Customer data is owned by the organisation; agents have access.
Review authority: Support agent
Decision authority: Support agent (for standard issues); specialist for complex.
Action authority: Support agent (sending responses); specialist for escalations.
Stop authority: Agent can escalate or stop automated triage if the AI fails.
Appeal authority: Team lead or manager.

Show analysis of delegation boundaries

Boundaries respected: The AI does not make decisions; it supports the agent. The agent retains review, decision, and action authority.

Potential risks: If agents become over-reliant on the AI's priority suggestions, they may overlook high-priority tickets that the AI misclassified. Regular audits and stop conditions are needed.

14. Exercises

Complete these five exercises to apply task allocation and authority mapping.

📝 Exercise 6.2.1: Workflow Task Inventory

Learning outcome: Decompose a workflow
Estimated time: 12 min

Scenario: A marketing team uses an AI to generate social-media posts. The process includes: (a) defining campaign goals, (b) selecting key messages, (c) having the AI generate post drafts, (d) reviewing drafts, (e) editing and finalising, (f) scheduling posts, (g) monitoring engagement, (h) adjusting strategy based on data.

Instructions: List each activity and assign a preliminary allocation category (human-led, AI-supported, etc.).

Show hint

Think about which steps require human creativity, judgment, and accountability, and which are suitable for AI generation or support.

Show sample solution design

(a) Human-led
(b) Human-led
(c) AI-generated + human-reviewed
(d) Human-led (review authority)
(e) Human-only
(f) Human action (or AI-supported with approval)
(g) AI-supported (data collection) + human analysis
(h) Human-led

📝 Exercise 6.2.2: Allocate Tasks

Learning outcome: Allocate tasks among humans and AI
Estimated time: 15 min

Scenario: A financial analyst uses an AI system to: (1) gather financial data from multiple sources, (2) calculate key ratios, (3) generate a draft report, (4) identify anomalies, (5) recommend investment options. The analyst reviews all outputs, makes investment decisions, and writes the final report.

Instructions: For each numbered activity, allocate to one of the seven categories and justify.

Show hint

Distinguish between data gathering/calculation (AI-friendly) and judgment/decision (human-only).

Show sample answers

AI-supported — gathering is bounded.
AI-supported — calculation is routine.
AI-generated + human-reviewed — draft report needs review.
AI-supported — flagging anomalies.
AI-generated + human-reviewed — recommendations must be reviewed; decision is human-only.

📝 Exercise 6.2.3: Identify Activities Requiring Specialist Review

Learning outcome: Recognise specialist review needs
Estimated time: 10 min

Scenario: A healthcare organisation uses an AI to generate patient-discharge summaries based on clinical notes. The summaries are then reviewed by a nurse before being sent to the patient's primary care physician.

Instructions: Which activities in this workflow require specialist (medical professional) review? Justify.

Show hint

Any activity that involves clinical judgment, interpretation of symptoms, or treatment plans requires medical expertise.

Show sample answer

The review of the discharge summary must be done by a clinician (nurse or physician) because it involves medical accuracy, completeness, and safety. The AI generation itself can be AI-supported, but the final approval must be specialist-reviewed.

📝 Exercise 6.2.4: Map Human Authority

Learning outcome: Map authority types
Estimated time: 15 min

Scenario: A project manager uses an AI to generate weekly status reports. The AI pulls data from project management software, drafts the report, and sends a draft to the manager. The manager reviews, edits, and distributes the final report.

Instructions: Identify who holds each of the following authorities: process ownership, information ownership, review authority, decision authority, action authority, stop authority.

Show sample map

Process ownership: Project manager
Information ownership: Project team / data source owners (manager has access)
Review authority: Project manager
Decision authority: Project manager (what to include, what to emphasise)
Action authority: Project manager (distributes report)
Stop authority: Project manager (can stop AI use if issues arise)

📝 Exercise 6.2.5: Diagnose Inappropriate Delegation

Learning outcome: Identify improper delegation
Estimated time: 12 min

Scenario: A university uses an AI system to grade student essays. The AI assigns a score and provides feedback. A professor reviews a random sample of 10% of the essays and accepts all AI scores without changes.

Instructions: What is problematic about this delegation? Which authority is being undermined?

Show sample answer

The professor has delegated decision authority (grading) to the AI without adequate review. This is inappropriate because grading requires professional judgment and understanding of individual student performance. The review is nominal (only 10% sample). This undermines the professor's professional authority and accountability. The professor should retain decision authority and review every essay (or at least all borderline cases) or use AI only as a supporting tool for feedback, not final scores.

15. Homework Problems

Formative · Ungraded · Recommended for portfolio preparation and Assignment 1 evidence.

🧠 Problem 6.2.1: Task Allocation

A legal paralegal uses an AI to summarise depositions and extract key facts. The paralegal reviews the summaries, incorporates them into a case brief, and submits the brief to the supervising lawyer, who approves it.

Allocate the tasks (summarising, extracting facts, reviewing, incorporating, approving) to appropriate categories.

Show solution design

Summarising, extracting facts: AI-generated + human-reviewed (paralegal reviews).
Reviewing: Human-led (paralegal).
Incorporating into case brief: Human-led (paralegal).
Approving: Human-only (supervising lawyer).

🧠 Problem 6.2.2: Process Owner Identification

In a software development team, an AI system automatically generates test cases based on user stories. The test cases are then executed by a QA engineer, who reports any failures to the development lead. The development lead decides which bugs to fix.

Who is the process owner? Justify.

Show solution design

The development lead is the process owner because they oversee the quality assurance process and make decisions about bug fixes. The QA engineer executes tasks but does not own the process.

🧠 Problem 6.2.3: Information Authority

A data scientist uses an AI to analyse customer purchase data. The AI generates insights and recommendations. The data scientist presents the findings to the marketing director, who decides on campaign changes.

Who holds information authority? Explain.

Show solution design

The marketing director (or the organisation) holds information authority because they own the customer data and decide how it can be used. The data scientist has access but does not own the data.

🧠 Problem 6.2.4: Specialist Review Requirement

A team of urban planners uses an AI to generate traffic-flow models for a new city development. The models are based on historical data and proposed land use. The team reviews the models and presents them to the city council.

Which parts of this workflow require specialist review? Identify at least two.

Show solution design

Model validation: A traffic engineer or urban planner must review the assumptions and outputs for accuracy.
Interpretation and implications: A specialist must assess how the model impacts real-world planning decisions and community outcomes.

🧠 Problem 6.2.5: Decision Authority — AI vs. Human

A financial institution uses an AI to approve small personal loans. The AI evaluates credit scores, income, and repayment history, and makes an approval decision automatically for amounts under $5,000. A human officer reviews applications that are denied or over $5,000.

Is decision authority appropriately retained? Discuss.

Show solution design

Partially appropriate — for low-risk, small amounts, automated decision-making may be acceptable if the system is well-tested and transparent. However, even for small loans, there is a risk of discrimination or error. The human review for denials and larger amounts is good. But the institution should also have a clear appeal process and regular audits to ensure fairness. Decision authority for the automated decisions is effectively delegated to the AI, which is acceptable only if the institution takes full accountability and provides recourse.

🧠 Problem 6.2.6: Action Authority — Email Generation

A public relations assistant uses an AI to draft responses to media inquiries. The assistant reviews the drafts, makes edits, and then clicks a button that causes the AI to send the email automatically.

Who has action authority? Is it properly assigned?

Show solution design

The assistant has action authority because they initiate the sending. However, the AI is technically performing the action (sending). This is acceptable if the assistant is authorised to send emails on behalf of the organisation. The key is that the human makes the conscious decision to send, and the AI executes the mechanical step. This is a proper allocation, provided the assistant reviews the content and has the authority to communicate externally.

🧠 Problem 6.2.7: Appeal Authority

A university uses an AI to recommend course schedules for students. If a student disagrees with the recommendation, they can submit an appeal to an academic advisor, who can override the recommendation.

Who holds appeal authority? Why is this important?

Show solution design

The academic advisor holds appeal authority. This is important because it provides a human check on the AI's recommendations, ensures fairness, and allows for individual circumstances that the AI may not consider.

🧠 Problem 6.2.8: Stop Authority

An AI system monitors social media for a brand and automatically posts responses to positive comments. If a negative or contentious comment appears, the system flags it for human review. However, the system has no stop button; it continues to post responses to positive comments even if a human reviewer is overwhelmed.

What is missing from this design, and what risk does it pose?

Show solution design

Missing: A clear stop authority with the ability to pause the entire system if needed. The risk is that the system may continue to post inappropriate or off-brand responses if the human reviewer is unable to keep up, or if a crisis occurs, the system could amplify damage.

🧠 Problem 6.2.9: Responsibility Matrix Construction

For the customer-support triage example in the worked example, create a simplified responsibility matrix with columns for Stage, AI Activity, Human Activity, Specialist Role, and Authority.

Show solution design

Table 4. Responsibility matrix for customer-support triage
Stage	AI activity	Human activity	Specialist	Authority
Categorisation	Classify issue type	—	—	— (AI support)
Priority suggestion	Assign priority level	Review and adjust if needed	—	Review authority (agent)
Draft response	Generate draft	Review, edit, approve	—	Decision authority (agent)
Final response	None	Send response	—	Action authority (agent)
Escalation	Flag complex issues	Escalate to specialist	Specialist team	Specialist review, decision

🧠 Problem 6.2.10: Transfer Problem — Allocating for a New Context

You are a human-resources manager implementing an AI system to screen resumes and shortlist candidates for open positions. The AI ranks applicants based on keywords and experience. You will review the shortlist and conduct interviews.

Apply the concepts: Decompose the workflow, allocate tasks, and identify the authorities that must be retained by humans.

Show solution design

Decomposition: (1) Define job criteria, (2) Collect resumes, (3) AI screening and ranking, (4) HR review of shortlist, (5) Interview scheduling, (6) Interview and selection, (7) Offer and hiring.

Allocation: (1) Human-led, (2) Human-led (or AI-supported for collection), (3) AI-generated + human-reviewed, (4) Human-led, (5) AI-supported (scheduling) + human approval, (6) Human-only, (7) Human-only.

Authorities retained: Decision authority (hiring manager), action authority (HR to make offer), stop authority (hiring manager can stop process), review authority (HR and hiring manager).

16. Mini-Project: Human-AI Responsibility Matrix

📁 Project: Human-AI Responsibility Matrix

Purpose: Extend your Lesson 6.1 use case by developing a comprehensive responsibility matrix. This project prepares you for Assignment 2 by requiring you to decompose a workflow, allocate tasks, and map human authority in detail.
Estimated effort: 90–120 minutes.
Deliverable: A structured document (approx. 600–1000 words) addressing 15 items.

Instructions

Use the same bounded AI-supported activity you selected in Lesson 6.1 (or a new one).
Do not use sensitive or institutional information.
Use the structure provided.
Complete the self-review checklist.

Required Sections

Workflow stage
Human purpose for each stage
Input to each stage
AI activity (if any)
Human activity
Specialist role (if needed)
Expected output
Verification method
Decision authority
Action authority
Documentation
Exception handling
Escalation
Stop authority
Final accountability

Self-Review Checklist

Have I decomposed the workflow into at least 5 distinct stages?
For each stage, have I clearly defined AI and human activities?
Have I identified which tasks are human-only?
Have I mapped decision, action, and stop authorities to specific roles?
Have I specified verification and exception handling?
Is the final accountability clearly assigned?

Show Hint 1: Decomposition

Break your workflow into logical stages: e.g., planning, preparation, generation, review, finalisation, action, and follow-up.

Show Hint 2: Allocation categories

Use the seven allocation categories to label each stage's AI involvement. Be specific about what the AI does and what the human does.

Show Hint 3: Authority mapping

For each stage, ask: who decides, who acts, who stops, who reviews? Ensure that decision and action authorities are not delegated to AI without clear human oversight.

Show illustrative partial example (different use case)

Use case: Using AI to generate a project status report.

Table 5. Partial responsibility matrix for project status report
Stage	AI activity	Human activity	Authority
Data gathering	Pull data from project tools	Define data sources	Information ownership (PM)
Draft generation	Generate report narrative	Review and edit	Review authority (PM)
Final approval	None	Approve report	Decision, action, stop (PM)

Your project will be more detailed.

17. Key Takeaways

Remember these points

Decompose workflows to identify every activity and decision point.
Allocate tasks using clear categories: human-led, AI-supported, AI-generated + human-reviewed, specialist-reviewed, human-only, prohibited, escalation required.
Retain human authority over decisions, actions, and stopping. Authority types include process ownership, information ownership, review, decision, action, appeal, correction, and stop.
A responsibility matrix visualises who does what and who decides, ensuring accountability.
Never delegate decision, action, or stop authority to AI without explicit human oversight and accountability.
PARTNER A & R guide you to allocate tasks and retain human authority—these are essential for responsible collaboration.

18. Key Terms

Workflow decomposition: Breaking a workflow into its constituent stages and decisions.
Task allocation: Assigning each activity to a category such as human-led, AI-supported, or human-only.
Human-led: An allocation category in which the human defines the task, directs the AI, and makes final decisions.
AI-generated + human-reviewed: An allocation category in which the AI produces a draft or recommendation that requires thorough human review.
Specialist-reviewed: An allocation category in which output requires review by a subject-matter expert or professional.
Human-only: An allocation category in which an activity must be performed entirely by a human.
Prohibited: An allocation category in which an activity is not permitted for AI under any circumstances.
Escalation required: An allocation category in which an activity must be escalated to a higher authority if it cannot be completed within defined parameters.
Process ownership: Authority over the overall workflow design and execution.
Information ownership: Authority over data and sources used in the workflow.
Review authority: The right to examine, critique, and approve or reject AI outputs.
Decision authority: Authority to make the final choice among alternatives.
Action authority: Authority to implement a decision (e.g., send, sign, modify).
Appeal authority: Authority to hear and decide on appeals or challenges to a decision.
Correction authority: Authority to correct errors after the fact.
Stop authority: Authority to halt the workflow or AI system if something goes wrong.
Responsibility matrix: A table mapping each workflow stage to AI, human, and specialist roles with authority types.
PARTNER: A framework for human-AI responsibility and collaboration; A = Allocate Tasks, R = Retain Human Authority.

19. Looking Ahead: Lesson 6.3

You now have the tools to allocate tasks and map authority. In Lesson 6.3: Meaningful Oversight, Automation Bias, and Responsibility, you will learn how to design effective human review processes, recognise automation bias, and identify responsibility laundering. You will also apply the T (Test and Verify) element of PARTNER.

Ensure you have completed the quiz, exercises, and mini-project. These are crucial for Assignment 2, where you will produce a full responsibility matrix for multiple contexts.

Continue to Lesson 6.3: Meaningful Oversight, Automation Bias, and Responsibility Return to Unit 6 Home

20. Sources & Further Reading

Primary source: Chapter 6: Human-AI Collaboration, COMP 3270 textbook (sections 6.4–6.6).
Supporting sources: Appendix D (Responsible-Use Toolkit) and Appendix H (Sample AI Use Statements) from the COMP 3270 course materials.
Institutional guidance: [INSERT VERIFIED REFERENCE TO INSTITUTIONAL AI POLICY]
Further reading: [INSERT VERIFIED REFERENCE TO RELEVANT PEER-REVIEWED LITERATURE]

Note: All page numbers, specific definitions, and direct quotations from the textbook are to be confirmed from the supplied course materials. [PAGE NUMBER TO BE CONFIRMED]

⚠️ Note

Some textbook content, definitions, examples, and figures are marked with placeholders. These will be supplied by the course coordinator or subject-matter expert.

Task Allocation and Human Authority

Table of Contents

1. Lesson Overview

2. Learning Outcomes

3. Before You Begin

4. Retrieval Practice: Concepts from Lesson 6.1

5. Opening Scenario: "The 'Human-in-the-Loop' Illusion"

6. Decomposing Workflows

7. Task Allocation Categories

8. Types of Human Authority

9. Human-AI Responsibility Matrix

10. PARTNER Walkthrough — Allocate Tasks & Retain Human Authority

A: Allocate Tasks

R: Retain Human Authority

11. Process Walkthrough: From Workflow to Allocation to Authority Map

Step 1 — Decompose the workflow

Step 2 — Allocate tasks

Step 3 — Map human authority

12. Worked Example: "Customer-Support Ticket Triage"

Decomposition

Allocation

Authority map

13. Self-Check Quiz

14. Exercises

15. Homework Problems

16. Mini-Project: Human-AI Responsibility Matrix

17. Key Takeaways

18. Key Terms

19. Looking Ahead: Lesson 6.3

20. Sources & Further Reading