COMP 3270 / Module 3 / Unit 7 / Lesson 4

Lesson 7.4 · Week 5 · Reading: Chapter 9 & FAIRER

FAIRER, Participation, Contestability, and Remedy

Apply the complete FAIRER framework, design meaningful participation, build contestability and remedy pathways, define monitoring and stop conditions, and make a justified recommendation about a bounded AI-supported use.

Estimated study time: 140–170 min Prerequisites: Lessons 7.1, 7.2, and 7.3 Module 3 · Week 5

Table of Contents

1. Lesson Overview

This culminating lesson integrates all the concepts from Unit 7 into a comprehensive fairness review framework. You will learn to apply the FAIRER framework—a structured approach to evaluating fairness, inclusion, and remedy in AI-supported workflows. You will also design meaningful participation for affected stakeholders, build contestability pathways that include notice, explanation, correction, human review, and challenge, and establish remedy mechanisms and ongoing monitoring.

Finally, you will make a justified recommendation about whether an AI-supported use should proceed, be limited, redesigned, escalated, or rejected. This lesson prepares you for the Unit 7 cumulative project and for Assignment 3.

Why this matters: A fairness evaluation is incomplete without mechanisms for affected people to challenge outcomes, obtain corrections, and access remedy. The FAIRER framework provides a systematic way to ensure that fairness is not just assessed but operationalised through participation, contestability, and ongoing accountability.

2. Learning Outcomes

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

  1. Apply the FAIRER framework to a bounded AI-supported case.
  2. Identify appropriate forms of participation for affected stakeholders.
  3. Design mechanisms for notice, explanation, correction, human review, contestability, and remedy.
  4. Establish monitoring indicators and reassessment triggers.
  5. Define fairness-related escalation and stop conditions.
  6. Develop a justified recommendation concerning an AI-supported use.
  7. Document remaining uncertainty and limitations.
  8. Connect FAIRER with ORBIT, SCOPE, and PARTNER frameworks.

3. Before You Begin

4. Retrieval Practice: Concepts from Lessons 7.1–7.3

These questions draw on Lessons 7.1–7.3. Answer each before reading the feedback.

Question 1: What is a proxy variable, and why can it produce biased outcomes even when explicit demographics are not used?

Show answer

Answer: A proxy variable is a measurable attribute that correlates with a protected characteristic. It can produce biased outcomes because the AI system may use the proxy as a substitute for the protected characteristic, indirectly reproducing discrimination even if the protected characteristic itself is not included.

Question 2: Explain the difference between disparate impact and discriminatory intent.

Show answer

Answer: Disparate impact refers to a materially unequal effect that occurs even without discriminatory intent; the process may appear neutral but produces unequal outcomes. Discriminatory intent involves a deliberate desire to treat a group unfairly. A system can have a disparate impact even if it was designed with good intentions.

Question 3: List four forms of harm that AI systems can cause, and give a brief example of each.

Show answer
  • Allocation: Unequal job opportunities.
  • Quality-of-service: Lower translation accuracy for certain languages.
  • Representational: Stereotyping in generated content.
  • Procedural: Opaque appeals process.

5. Opening Scenario: "AI-Powered Public Benefits Eligibility Tool"

📌 Scenario — AI-Powered Public Benefits Eligibility Tool

A state agency deploys an AI system to assist caseworkers in determining eligibility for public benefits (e.g., food assistance, housing support). The system processes applications and provides a recommendation score, which caseworkers use to make final decisions. The system was developed with the goal of increasing efficiency and consistency.

However, community advocates raise concerns:

  • Applicants from certain regions receive lower scores, even when other factors are similar (suggesting possible proxy variables).
  • The system's explanations are available only in English and are highly technical, making it difficult for applicants to understand decisions.
  • There is no clear appeal process; applicants are told they can "request a review" but no details are provided.
  • Caseworkers feel pressured to follow the AI's recommendation, and there is no training on the system's limitations.
  • No community representatives were involved in the system's design or testing.

The agency wants to improve the system's fairness but is unsure how to proceed.

Questions to consider:

  1. What does a complete fairness response require in this situation?
  2. Who should be involved in defining fairness and designing remedies?
  3. What mechanisms for contestability and remedy should be established?
  4. How should the agency monitor the system going forward?
  5. What conditions would justify pausing or stopping the system?
Show guided analysis

Complete fairness response requires:

  • Participation: Involve applicants, community advocates, caseworkers, and subject-matter experts in reviewing the system.
  • Contestability: Provide clear explanations in accessible language, a simple process for requesting human review, and a pathway to appeal decisions.
  • Remedy: Offer correction of errors, reconsideration, and, where appropriate, compensation or restoration of benefits.
  • Monitoring: Track approval rates, error types, complaint patterns, and disparities; use this data to trigger reassessment.
  • Stop conditions: Pause the system if disparities cannot be explained or corrected, if accessibility barriers persist, or if contestability is not functional.

6. The FAIRER Framework

FAIRER framework

[INSERT AUTHOR-APPROVED FAIRER EXPANSION] — Until the expansion is confirmed, treat FAIRER operationally as a framework addressing fairness goals, affected stakeholders, inclusion and accessibility, representation and differential effects, evidence and evaluation, and redress, remedy, and reassessment.

FAIRER provides a structured approach to evaluating and ensuring fairness in AI-supported workflows. It is not a guarantee but a process for systematic inquiry, stakeholder engagement, and continuous improvement.

Below are the operational elements of FAIRER (using placeholders until the approved expansion is supplied):

Table 1. Operational elements of FAIRER (placeholder)
Element Definition (placeholder) Guiding questions Common mistake
[FAIRER ELEMENT 1] [INSERT AUTHOR-APPROVED DEFINITION] Who defines fairness? What are the goals? Assuming one universal fairness metric.
[FAIRER ELEMENT 2] [INSERT AUTHOR-APPROVED DEFINITION] Who are the affected stakeholders? Who is missing? Overlooking indirect stakeholders (e.g., family members).
[FAIRER ELEMENT 3] [INSERT AUTHOR-APPROVED DEFINITION] Is the system inclusive and accessible? What barriers exist? Equating representation with inclusion.
[FAIRER ELEMENT 4] [INSERT AUTHOR-APPROVED DEFINITION] Are there differential effects? Are proxies operating? Ignoring intersectional effects.
[FAIRER ELEMENT 5] [INSERT AUTHOR-APPROVED DEFINITION] What evidence is needed? How is it gathered and evaluated? Relying solely on aggregate accuracy.
[FAIRER ELEMENT 6] [INSERT AUTHOR-APPROVED DEFINITION] What remedy and reassessment are available? Treating disclosure as remedy.
Key insight: FAIRER is not a checklist but a cycle of inquiry and action. It requires ongoing engagement with stakeholders, continuous monitoring, and a willingness to adjust or stop the system when needed.
Section summary
  • FAIRER provides a systematic process for fairness evaluation and remedy.
  • It is not a one-time checklist but an ongoing cycle.
  • Each element requires stakeholder engagement and contextual judgment.

7. Participation

Meaningful participation means that affected stakeholders have a genuine voice and influence in the design, testing, monitoring, and governance of AI systems. Participation can take many forms:

Participation must be meaningful rather than symbolic—that is, it must have the potential to influence decisions, and stakeholders must be aware that their input is valued and acted upon.

When designing participation, consider:

Section summary
  • Meaningful participation gives stakeholders genuine influence, not just a voice.
  • Participation can take many forms, from consultation to co-design.
  • Resources and power imbalances must be addressed for participation to be meaningful.

8. Contestability

Contestability means that affected people can:

A contestability pathway should be designed with the user in mind, ensuring that it is accessible, easy to use, and well-communicated. Key design elements:

Section summary
  • Contestability requires notice, explanation, correction, human review, and appeal.
  • The pathway must be accessible and well-communicated.
  • Documentation supports monitoring and continuous improvement.

9. Remedy

Remedy refers to the actions taken to correct harms caused by an AI-supported system. Possible remedies include:

Remedies should be proportionate to the harm caused and should be designed in consultation with affected stakeholders. Importantly, disclosure alone is not remedy—telling someone about a problem is not the same as fixing it.
Section summary
  • Remedy includes correction, reconsideration, restoration, compensation, and system changes.
  • Remedies must be proportionate and designed with affected stakeholders.
  • Disclosure is not remedy.

10. Monitoring and Reassessment

Fairness is not a one-time achievement; it requires ongoing monitoring and reassessment. Key monitoring indicators may include:

Monitoring should be accompanied by reassessment triggers—thresholds or events that prompt a deeper review. For example:

Section summary
  • Monitoring indicators should cover performance, complaints, accessibility, and outcomes.
  • Reassessment triggers prompt deeper reviews when issues emerge.
  • Monitoring and reassessment are ongoing responsibilities.

11. Escalation and Stop Conditions

Fairness-related escalation and stop conditions define when a system should be paused, restricted, or withdrawn. Examples:

These conditions should be defined in advance and communicated to all stakeholders. They are not a sign of failure but a responsible mechanism for ensuring that fairness is taken seriously.

Section summary
  • Stop conditions specify when a system should be paused, restricted, or withdrawn.
  • They should be defined in advance and communicated clearly.
  • Stopping is a sign of responsible governance, not failure.

12. Connecting FAIRER with ORBIT, SCOPE, and PARTNER

FAIRER builds on and extends the earlier frameworks you learned:

Together, these frameworks provide a comprehensive approach to responsible AI use: from defining the problem and allocating roles, to identifying biases, to designing a fair and accountable system.

Section summary
  • ORBIT, SCOPE, PARTNER, and FAIRER are complementary frameworks.
  • Together they cover purpose, context, roles, fairness, and remedy.
  • Using all four provides a comprehensive approach to responsible AI.

13. Process Walkthrough: Fairness Evidence → Participation → Contestability → Remedy → Monitoring → Decision

This walkthrough applies the FAIRER framework to the public benefits eligibility scenario. It shows how to move from evidence to a justified recommendation.

Table 2. FAIRER walkthrough for the public benefits system
Stage Activity Fairness check Output / decision
1. Assemble evidence Collect performance data, complaints, user feedback, and audit findings. Are there disparities? What are the error types? Who is affected? Evidence summary: regional disparities, lack of explanations, no appeal process.
2. Identify stakeholders Map affected groups: applicants, caseworkers, community organisations, disability advocates. Who is missing from current discussions? Who has authority? Stakeholder map: applicants, caseworkers, advocates, agency leadership.
3. Design participation Plan how to involve stakeholders in review, redesign, and monitoring. Is participation meaningful? Are power imbalances addressed? Advisory group with community reps; regular feedback sessions.
4. Build contestability Design notice, explanation, correction, human review, and appeal processes. Is the process accessible? Is it easy to use? Multilingual explanation letters, online and phone appeal options.
5. Define remedies Specify what remedies are available (correction, reconsideration, etc.). Are remedies proportionate? Do they address root causes? Correction of errors, human review of denials, compensation for delays.
6. Set monitoring indicators Select metrics and triggers for review. Will monitoring capture emerging disparities? Disaggregated approval rates, complaint trends, override rates.
7. Define stop conditions Specify when to pause or stop the system. Are conditions clear and actionable? Stop if regional disparities exceed threshold and cannot be explained.
8. Make recommendation Based on all evidence, decide whether to proceed, redesign, limit, or stop. Is the recommendation justified and documented? Proceed with safeguards (participation, contestability, monitoring) and pilot with community oversight.
Walkthrough summary
  • FAIRER walkthrough moves from evidence to a justified recommendation.
  • Each stage adds a layer of fairness assurance.
  • Recommendations are conditional and include safeguards.

14. Worked Example: AI-Powered Educational Recommendation System

Context: A university uses an AI system to recommend courses to students, based on their declared majors, grades, and past enrolments. The system is intended to help students plan their schedules efficiently.

Concerns: The system recommends courses that are predominantly taken by students of certain demographic groups, potentially reinforcing stereotypes. There is no explanation for recommendations, and students cannot appeal. The system was not tested with students with disabilities.

Worked FAIRER Analysis

  1. Fairness goals: Ensure that all students receive useful, relevant recommendations that do not limit their academic exploration.
  2. Stakeholders: Students, advisors, faculty, disability services, and student advocacy groups.
  3. Inclusion and accessibility: The system interface is not screen-reader compatible; explanations are missing. Need to provide accessible explanations and alternative formats.
  4. Representation and differential effects: Analyse recommendation patterns by major, gender, and race. Examine if certain groups are steered away from certain courses.
  5. Evidence: Collect student feedback, advisor observations, and system logs. Compare recommendation diversity across groups.
  6. Redress: Provide a correction mechanism (students can flag irrelevant recommendations), human review (advisor can override), and a formal appeal for students who feel disadvantaged.
  7. Monitoring: Track recommendation diversity, student satisfaction, and course-enrolment patterns. Trigger review if disparities emerge.
  8. Participation: Involve students and advisors in regular reviews; run focus groups to gather qualitative feedback.
  9. Stop conditions: Pause the system if certain groups consistently receive lower-quality recommendations or if accessibility barriers persist.
  10. Recommendation: Proceed with a phased rollout, with the safeguards described, and a commitment to annual fairness audits.
Show key insights from this example
  • FAIRER requires concrete actions: stakeholder involvement, contestability, monitoring, and stop conditions.
  • Participation must be ongoing, not one-time.
  • Recommendations are conditional and include safeguards.
  • Residual uncertainty is documented (e.g., long-term effects not yet known).

15. Guided Learning Activities

🧩 Activity 7.4.A: FAIRER Walkthrough
  • Purpose: To apply FAIRER to a synthetic case.
  • Learning outcome: Apply FAIRER elements to a bounded case.
  • Estimated time: 25 minutes.
  • Scenario: Use the public benefits scenario (Section 5).
  • Task: For each operational FAIRER element, write a brief description of how it applies to the scenario. Include at least one guiding question for each.
Show sample walkthrough
  • Element 1 (Fairness goals): Fairness means all eligible applicants receive fair consideration, with no disparate impact by region or other proxies. Guiding question: What metrics define "fair consideration"?
  • Element 2 (Stakeholders): Applicants, caseworkers, advocates, agency leaders. Guiding question: How are missing stakeholders (e.g., non-English speakers) identified and included?
  • Element 3 (Inclusion/accessibility): The system must provide multilingual explanations and be usable by people with disabilities. Guiding question: What accessibility standards apply?
  • Element 4 (Differential effects): Analyse approval rates by region and other factors. Guiding question: What proxy variables might be at play?
  • Element 5 (Evidence): Collect disaggregated performance data, user feedback, and independent audits. Guiding question: What evidence is missing?
  • Element 6 (Redress): Provide a clear appeal process, correction of errors, and human review. Guiding question: How is remedy monitored for effectiveness?
🧩 Activity 7.4.B: Stakeholder Participation Design
  • Purpose: To design a participation plan for an AI system.
  • Learning outcome: Identify appropriate forms of participation.
  • Estimated time: 20 minutes.
  • Task: For the public benefits scenario, design a participation plan that includes affected people, workers, specialists, accessibility reviewers, community authorities, decision-makers, and remedy owners. Describe how each group would be involved and how their input would influence the system.
Show sample plan
  • Affected applicants: Conduct focus groups and surveys; include them in advisory board.
  • Caseworkers: Train them on the system and collect feedback on usability and fairness.
  • Specialists (fairness, AI): Conduct bias audits and provide recommendations.
  • Accessibility reviewers: Test the system with assistive technologies.
  • Community authorities: Engage community organisations to reach underrepresented groups.
  • Decision-makers: Agency leaders to make final decisions based on input.
  • Remedy owners: Designate a specific office to handle appeals and corrections.
🧩 Activity 7.4.C: Contestability Pathway
  • Purpose: To design a contestability pathway.
  • Learning outcome: Design notice, explanation, correction, human review, challenge, and remedy.
  • Estimated time: 20 minutes.
  • Task: For the public benefits scenario, design a detailed contestability pathway for applicants. Describe each step, the responsible parties, and the expected timelines.
Show sample pathway
  1. Notice: Applicants receive a letter (in multiple languages) explaining that AI was used and the key factors considered.
  2. Explanation: The letter includes a plain-language summary of the decision and the factors that influenced it.
  3. Correction: Applicants can submit corrected information online or by phone.
  4. Human review: If the applicant disagrees, they can request a review by a senior caseworker within 10 business days.
  5. Challenge/appeal: If the review is unsatisfactory, the applicant can appeal to an independent review board.
  6. Response: The board must respond within 20 business days.
  7. Remedy: If the appeal is successful, the applicant receives the appropriate benefit and any back-pay.
  8. Documentation: All steps are recorded for monitoring purposes.
🧩 Activity 7.4.D: Fairness Monitoring Plan
  • Purpose: To design a monitoring plan.
  • Learning outcome: Select monitoring indicators and reassessment triggers.
  • Estimated time: 15 minutes.
  • Task: For the public benefits scenario, select at least five monitoring indicators and describe what would trigger a reassessment.
Show sample plan
  • Indicators: Approval rates by region, complaint rates, override rates (caseworkers overruling AI), user satisfaction scores, and accessibility test results.
  • Triggers: A 10% or greater disparity in approval rates between regions, a 20% increase in complaints from a specific group, or three consecutive accessibility complaints.
🧩 Activity 7.4.E: Proceed, Redesign, Restrict, or Stop?
  • Purpose: To practice making a justified recommendation.
  • Learning outcome: Make and justify a final recommendation.
  • Estimated time: 15 minutes.
  • Task: Based on the public benefits scenario and your analysis, make a final recommendation: Proceed, Proceed with safeguards, Pilot with monitoring, Limit the use, Redesign, Require specialist review, Provide a non-AI pathway, Pause, Do not proceed, or Retire. Justify your answer with evidence from the scenario.
Show sample recommendation

Recommendation: Pilot with monitoring and strong safeguards (proceed with conditions).

Justification: The system has potential efficiency benefits, but the observed regional disparities and lack of contestability are serious concerns. By implementing a pilot with the proposed participation, contestability, and monitoring measures, the agency can gather real-world evidence and make adjustments before full deployment. If disparities persist after the pilot, escalate to redesign or pause.

16. Exercises

📝 Exercise 7.4.1: Apply FAIRER
  • Difficulty: Intermediate
  • Estimated time: ~12 min
  • Learning outcome: Apply FAIRER to a given case.

Task: Choose a synthetic AI-supported case (e.g., the career-profile tool from Lesson 7.1). For each of the six operational FAIRER elements, write one sentence describing how it applies.

Show sample answer
  • Fairness goals: Ensure diverse and non-stereotyped career representation.
  • Stakeholders: Students, career advisors, alumni, and employers.
  • Inclusion: Provide profiles for multiple cultures, genders, and abilities.
  • Differential effects: Check if certain groups are underrepresented in certain fields.
  • Evidence: Conduct a diversity audit of profiles and user feedback.
  • Redress: Allow users to suggest additional profiles and flag stereotypes.
📝 Exercise 7.4.2: Identify Meaningful Participation
  • Difficulty: Intermediate
  • Estimated time: ~10 min
  • Learning outcome: Distinguish meaningful from symbolic participation.

Task: Give an example of symbolic participation in an AI project and explain why it is not meaningful. Then, propose a more meaningful alternative.

Show sample answer

Symbolic: A company holds a public webinar to present the AI system and collects comments via chat, but does not respond to or incorporate the comments.

Meaningful alternative: Hold workshops with affected communities where their input is recorded, and a follow-up report is shared showing how the input influenced design changes.

📝 Exercise 7.4.3: Design Accessible Notice and Explanation
  • Difficulty: Advanced
  • Estimated time: ~12 min
  • Learning outcome: Design accessible notice and explanation.

Task: Write a sample notice (100–150 words) for a public benefits decision that includes: (a) a clear statement that AI was used, (b) the key factors considered, (c) how to get more information, and (d) how to request a review. Write it at an accessible literacy level (Grade 8–10).

Show sample notice

Sample notice:

Dear Applicant,

We used a computer system to help us review your application for benefits. The system looked at your income, family size, and housing situation. It helped us make a decision faster, but a person still checked the result.

If you have questions about this decision, please call us at 1-800-555-0199 or visit our website for a detailed explanation. You can also request a human review if you think something is wrong. We will respond within 10 days.

Sincerely,
Benefits Office

📝 Exercise 7.4.4: Build a Contestability Pathway
  • Difficulty: Advanced
  • Estimated time: ~15 min
  • Learning outcome: Design a contestability pathway.

Task: For the educational recommendation system (worked example), design a simple contestability pathway for students. Include notice, explanation, correction, human review, and appeal.

Show sample pathway
  • Notice: When a student receives recommendations, an email explains that the system uses their past courses and major to suggest options.
  • Explanation: The email includes a list of the factors considered and a link to a more detailed FAQ.
  • Correction: Students can update their academic interests or flag a recommendation as irrelevant.
  • Human review: If a student is unsatisfied, they can meet with an academic advisor to review their plan.
  • Appeal: If the advisor cannot resolve the issue, the student can appeal to the department head.
📝 Exercise 7.4.5: Select Monitoring Indicators
  • Difficulty: Intermediate
  • Estimated time: ~10 min
  • Learning outcome: Select appropriate monitoring indicators.

Task: For the educational recommendation system, list three monitoring indicators that would help detect bias, and explain what each would reveal.

Show sample answer
  • Indicator 1: Recommendation diversity by major — reveals if certain majors are over- or under-recommended to certain demographic groups.
  • Indicator 2: Student satisfaction by group — reveals if certain groups are less satisfied with the recommendations.
  • Indicator 3: Advisor override rate — shows how often advisors override the system, and for which students, indicating potential biases.
📝 Exercise 7.4.6: Develop Remedy and Stop Conditions
  • Difficulty: Advanced
  • Estimated time: ~12 min
  • Learning outcome: Define remedy and stop conditions.

Task: For the public benefits system, define at least three possible remedies for a mistaken denial, and three stop conditions that would trigger a system pause.

Show sample answer

Remedies:

  • Reconsideration of the application with corrected information.
  • Retroactive benefits if the denial was wrongful.
  • An apology and explanation of what went wrong.

Stop conditions:

  • If regional approval disparities exceed 15% for three consecutive months.
  • If more than 10% of appeals result in overturned decisions due to systematic error.
  • If accessibility complaints are not resolved within 30 days.

17. Homework Problems

Formative · Ungraded · Recommended preparation for Assignment 3.

🧠 Problem 7.4.1: FAIRER Element Matching
  • Estimated time: ~5 min

Task: Match the FAIRER operational elements to their descriptions (placeholder). Write the element number next to each description.

Show sample matching

Descriptions:

  1. Identifying who is affected and who is missing — Element 2
  2. Designing correction and appeal processes — Element 6
  3. Determining what fairness means in context — Element 1
🧠 Problem 7.4.2: FAIRER Case Analysis
  • Estimated time: ~10 min

Task: Choose a synthetic AI use case (e.g., the translation service from Lesson 7.3). For each FAIRER element, write one sentence describing how it applies.

Show sample answer
  • Fairness goals: Ensure equal quality of translation across all languages.
  • Stakeholders: Users of each language, translators, public health officials.
  • Inclusion: Provide translations in all relevant languages with appropriate cultural nuance.
  • Differential effects: Monitor translation accuracy per language.
  • Evidence: Use human evaluation and user feedback.
  • Redress: Allow users to report translation errors and suggest corrections.
🧠 Problem 7.4.3: Participation
  • Estimated time: ~8 min

Task: Describe a situation where participation might be tokenistic, and explain how to make it more meaningful.

Show sample answer

Tokenistic: A company invites a few community members to a one-hour meeting but does not incorporate their feedback into the final design.

More meaningful: Form a community advisory group that meets regularly, provides compensation for members' time, and has a formal role in decision-making, with minutes and action items shared publicly.

🧠 Problem 7.4.4: Contestability
  • Estimated time: ~8 min

Task: What are the minimum requirements for a contestability pathway? List at least four.

Show sample answer
  • Notice that AI was used.
  • An understandable explanation of the decision.
  • A way to correct inaccurate information.
  • A process to request human review or appeal.
🧠 Problem 7.4.5: Explanation Design
  • Estimated time: ~10 min

Task: Write a short explanation (100 words) for a loan denial that is clear and non-technical, suitable for someone with high school education. Include the main factors and how to contest.

Show sample answer

Sample explanation:

Your loan application was not approved because our system reviewed your income, current debts, and credit history, and we found that your debt-to-income ratio is higher than our guidelines allow. This is a standard measure we use to ensure that loans are affordable.

If you believe we have incorrect information, please call us at 1-800-555-0123 to correct it. You can also request a human review of your application within 30 days.

🧠 Problem 7.4.6: Remedy
  • Estimated time: ~8 min

Task: Give an example of a remedy that is proportionate to the harm caused by an AI error. Explain why it is proportionate.

Show sample answer

Example: A student was wrongly denied access to a course because the AI system misread their prerequisites. The remedy: promptly enrol the student in the course, offer a make-up session for missed material, and provide an apology.

Why proportionate: The harm was limited to a delay in enrollment, and the remedy restores the student's opportunity without over-compensating.

🧠 Problem 7.4.7: Monitoring
  • Estimated time: ~8 min

Task: Select two monitoring indicators for an AI system that allocates housing assistance, and explain what each indicator would reveal about fairness.

Show sample answer
  • Indicator 1: Approval rates by neighbourhood — reveals if certain areas are disproportionately denied.
  • Indicator 2: Appeal success rates — shows whether appeals are effective for all groups or if some groups are less successful.
🧠 Problem 7.4.8: Stop Condition
  • Estimated time: ~8 min

Task: Define a stop condition for an AI system used for job screening. Explain why that condition is appropriate.

Show sample answer

Stop condition: If the system's rejection rate for candidates from any demographic group exceeds the group's representation in the applicant pool by more than 20% for two consecutive quarters, and the cause cannot be explained by legitimate job requirements.

Why appropriate: This threshold indicates a potentially serious disparate impact that warrants investigation and possible suspension.

🧠 Problem 7.4.9: Final Recommendation
  • Estimated time: ~10 min

Task: For the public benefits system, make a final recommendation (choose from the allowed list) and provide a justification of 100–150 words.

Show sample answer

Recommendation: Pilot with monitoring and strong safeguards.

Justification: The system offers potential efficiency gains, but the observed disparities and lack of contestability are serious. A pilot with the proposed participation, contestability, and monitoring measures will allow the agency to test the system in a controlled manner, gather evidence, and make adjustments. If the disparities persist after six months, we should escalate to redesign or pause.

🧠 Problem 7.4.10: Transfer
  • Estimated time: ~10 min

Task: Apply the FAIRER framework to a domain of your choice (e.g., legal tech, healthcare, education). Write a brief outline of how you would apply each element.

Show sample answer

Domain: Healthcare AI for diagnostic support.

  • Fairness goals: Equal diagnostic accuracy across demographic groups.
  • Stakeholders: Patients, doctors, nurses, medical researchers.
  • Inclusion: Ensure the system works for diverse populations (age, ethnicity, language).
  • Differential effects: Monitor error rates by group.
  • Evidence: Use clinical trials and real-world data.
  • Redress: Provide a mechanism for doctors to override and patients to request second opinions.

18. Mini-Project: Complete FAIRER Review

📁 Project: Complete FAIRER Review for a Bounded AI-Supported Case
  • Purpose: To integrate all Unit 7 learning into a comprehensive fairness review.
  • Learning outcomes: Apply FAIRER, design participation, contestability, remedy, and monitoring, and make a justified recommendation.
  • Estimated effort: 100–150 minutes.
  • Context: Choose one of the following synthetic cases, or use a case from your experience (syntheticised).
  • Option A: Public benefits eligibility (as in the scenario).
  • Option B: AI-powered academic advising system.
  • Option C: AI-generated news summarisation for public consumption.

Required deliverable: A structured report with all sections listed below (approximately 800–1200 words or equivalent).

  1. Use-case title
  2. Human problem
  3. Legitimate purpose
  4. AI role
  5. Stakeholders
  6. Bias entry points
  7. Representation analysis
  8. Default assumptions
  9. Equality analysis
  10. Equity analysis
  11. Fairness criteria
  12. Inclusion analysis
  13. Accessibility analysis
  14. Intersectional considerations
  15. Differential-performance evidence
  16. Proxy-variable analysis
  17. Representational harms
  18. Allocation harms
  19. Quality-of-service harms
  20. Procedural harms
  21. Relational or capability harms
  22. Disparate-impact analysis
  23. Automation-bias risk
  24. Participation plan
  25. Human authority
  26. Notice and explanation
  27. Correction process
  28. Human-review process
  29. Contestability
  30. Remedy
  31. Monitoring indicators
  32. Escalation
  33. Stop conditions
  34. Complete FAIRER analysis
  35. Residual uncertainty
  36. Final recommendation
  37. Assignment 3 connection
  38. Assignment 1 reflection connection
  39. Preparation for Unit 8

Allowed recommendations: Proceed, Proceed with safeguards, Pilot with monitoring, Limit the use, Redesign, Require specialist or community review, Provide a non-AI pathway, Pause, Do not proceed, Retire the use.

Privacy and ethical boundaries: Use synthetic information only.

Self-review checklist

  • Have I applied all six FAIRER elements?
  • Have I identified at least two proxy variables?
  • Have I classified at least three forms of harm?
  • Have I designed a participation plan that includes affected people?
  • Have I designed a contestability pathway?
  • Have I defined remedy, monitoring, and stop conditions?
  • Have I made a clear, justified recommendation?
  • Have I documented residual uncertainty?
Show solution architecture
  • Start with a clear description of the case and its purpose.
  • Map stakeholders and identify who is affected.
  • Analyse bias entry points and representation.
  • Apply equality, equity, and fairness criteria.
  • Assess accessibility and intersectionality.
  • Examine differential performance and proxies.
  • Classify harms and assess disparate impact.
  • Design participation and contestability.
  • Define monitoring and stop conditions.
  • Apply FAIRER and make a justified recommendation.
  • Document residual uncertainty.
Show illustrative partial example (different context)

Context: AI-powered job matching platform.

  • Stakeholders: Job seekers, employers, recruiters.
  • Bias entry points: Historical data (past hiring patterns), proxy variables (postal code, school name).
  • Fairness criteria: Equal access to job listings, equitable matching quality.
  • Contestability: Allow job seekers to flag irrelevant matches and provide feedback.
  • Remedy: Adjust algorithm if bias is detected; provide human review for complaints.
  • Recommendation: Proceed with safeguards (regular bias audits, user feedback, and a non-AI alternative for job search).

19. Self-Check Quiz

Eight questions · Formative · Ungraded. Expand each item to see feedback.

Question 1: What is the primary purpose of the FAIRER framework?

Select one:
Show answer and explanation

Correct answer: b.

Explanation: FAIRER is a structured approach to evaluating fairness, inclusion, and remedy. It is not a technical formula (a) nor does it replace human judgment (c). It also does not guarantee perfect unbiased outputs (d).

Question 2: Which of the following are elements of a contestability pathway? Select all that apply.

Select all that apply:
Show answer and explanation

Correct answers: All of the above (a–e).

Explanation: Contestability requires notice, explanation, correction, human review, and appeal. All are necessary for meaningful contestability.

Question 3: Match each FAIRER operational element (placeholder) to its focus.

Table 3. Matching FAIRER elements to their focus
Element Focus
1. [Element 1] A. Stakeholder involvement
2. [Element 2] B. Defining fairness goals
3. [Element 6] C. Redress and reassessment
Show answer and explanation
  • 1 — B (Defining fairness goals)
  • 2 — A (Stakeholder involvement)
  • 3 — C (Redress and reassessment)

Question 4: What is the difference between disclosure and remedy?

Show answer and explanation

Answer: Disclosure is simply telling someone about a problem or a decision. Remedy involves taking action to correct the problem, address the harm, and prevent recurrence. Disclosure alone does not fix the issue.

Question 5: True or false: Once a fairness review is completed, the system can be considered permanently fair.

Select one:
Show answer and explanation

Answer: False.

Explanation: Fairness is not static; it requires ongoing monitoring, reassessment, and adaptation as contexts change, new data emerges, and stakeholder needs evolve.

Question 6: Give an example of a stop condition for an AI-supported system.

Show answer and explanation

Example: If the system's error rate for a particular demographic group exceeds 20% and cannot be explained or corrected, the system should be paused pending a full review.

Question 7: Which of the following is a sign of meaningful participation?

Select one:
Show answer and explanation

Correct answer: b.

Explanation: Meaningful participation involves influence and feedback, not just being informed or consulted without effect.

Question 8: How does FAIRER relate to the ORBIT, SCOPE, and PARTNER frameworks?

Show answer and explanation

Answer: FAIRER builds on these frameworks: ORBIT provides initial purpose and workflow framing, SCOPE helps identify omissions and evidence gaps, PARTNER guides human-AI role allocation. FAIRER adds a systematic fairness evaluation, stakeholder participation, contestability, remedy, and ongoing monitoring, making the overall approach more comprehensive.

20. Assignment Connections

Assignment 3: Responsible AI Case Analysis

This lesson is directly aligned with Assignment 3. Preserve the following evidence:

  • FAIRER analysis — your application of each element.
  • Participation plan — who you would involve and how.
  • Contestability and remedy design — the pathways you designed.
  • Monitoring indicators and stop conditions — what you would track and when to stop.
  • Final recommendation — your justified conclusion.
  • Residual uncertainty — what you do not yet know.

These sections will form a significant portion of your Assignment 3 case analysis.

📝 Assignment 1: Reflective Journal

Reflect on the following prompts for your Unit 7 journal entry.

Show reflection prompts and planning checklist
  1. How has your understanding of fairness changed after learning about FAIRER?
  2. What does meaningful participation mean to you, and how could it be implemented in your field?
  3. Why is contestability essential for responsible AI? Have you ever experienced a lack of contestability?
  4. What remedies would you consider appropriate for AI-related harms?
  5. How have your responsible-use boundaries shifted after completing Unit 7?

Reflection planning checklist:

  • I have reflected on how FAIRER adds to other frameworks.
  • I have considered a situation where contestability was lacking.
  • I have thought about appropriate remedies for AI harms.
  • I have identified a personal boundary or commitment for responsible AI use.

21. Key Takeaways

Remember these points
  • FAIRER provides a systematic fairness review process—it is not a one-time checklist but an ongoing cycle.
  • Participation must be meaningful—stakeholders must have genuine influence, not just symbolic consultation.
  • Contestability is a right—affected people need notice, explanation, correction, human review, and appeal.
  • Remedy goes beyond disclosure—it requires action to correct harms and prevent recurrence.
  • Monitoring is continuous—fairness requires ongoing data collection, stakeholder feedback, and reassessment.
  • Stop conditions are essential—knowing when to pause or stop is a sign of responsible governance.
  • Recommendations are conditional—they should include safeguards and acknowledge residual uncertainty.

22. Key Terms

FAIRER framework
A structured approach to evaluating fairness, inclusion, and remedy in AI-supported workflows. [INSERT AUTHOR-APPROVED EXPANSION].
Participation
The involvement of affected stakeholders in design, testing, monitoring, and governance, with genuine influence.
Contestability
The ability of affected people to challenge decisions, correct information, and obtain human review and remedy.
Remedy
Action taken to correct harms, including correction, reconsideration, restoration, compensation, or system changes.
Monitoring
Ongoing observation of performance, complaints, and outcomes to detect emerging fairness issues.
Reassessment trigger
A threshold or event that prompts a deeper review of the system's fairness.
Stop condition
A predefined circumstance under which the system should be paused, restricted, or withdrawn.
Escalation
The process of moving a case or issue to a higher level of review or authority.
Symbolic participation
Participation that does not lead to real influence, often used to give an appearance of consultation.
Residual uncertainty
What remains unknown despite the available evidence, requiring caution and ongoing inquiry.

23. Looking Ahead: Unit 8

Congratulations—you have completed all four lessons of Unit 7! You are now prepared to:

Next, you will move on to Unit 8: Privacy, Security, and Data Protection, where you will learn about protecting personal information, confidentiality, and the PROTECT framework.

Continue to Unit 8: Privacy, Security, and Data Protection Return to Unit 7 Home

24. Sources & Further Reading

[VERIFICATION REQUIRED] — Please confirm all source details with the course coordinator and textbook publisher.

⚠️ Note

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

👤