COMP 3270 / Module 3 / Unit 7 / Lesson 2

Lesson 7.2 · Week 5 · Reading: Chapter 9

Fairness, Equity, Accessibility, and Intersectionality

Distinguish equality from equity, examine how fairness criteria may conflict, identify accessibility barriers across language, disability, technology, and cost, and design meaningful participation that includes affected stakeholders.

Estimated study time: 130–160 min Prerequisite: Lesson 7.1 Module 3 · Week 5

Table of Contents

1. Lesson Overview

This lesson moves from identifying bias to examining what makes an AI-supported system fair, equitable, inclusive, and accessible. You will learn to distinguish equality from equity, explore why different fairness criteria may conflict, and examine how barriers—including language, disability, technology, and cost—affect access and participation.

You will also be introduced to intersectionality—the idea that people experience overlapping social positions and barriers that cannot be understood by examining one characteristic at a time. The lesson concludes with designing meaningful participation, ensuring that affected stakeholders have a voice in defining fairness, testing, and monitoring.

Why this matters: A system can be technically unbiased yet still be unfair if it does not account for different needs, barriers, or starting points. Fairness is not a single, universal formula—it requires context-sensitive evaluation, active inclusion, and accessible design.

2. Learning Outcomes

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

  1. Distinguish equality, equity, fairness, inclusion, and accessibility.
  2. Recognize that fairness may involve competing criteria that require explicit justification.
  3. Identify accessibility barriers in AI-supported content and services.
  4. Explain intersectionality conceptually and identify how multiple barriers may interact.
  5. Identify uneven benefits, burdens, and participation across different user groups.
  6. Avoid treating one demographic dimension as sufficient for fairness analysis.
  7. Design a participation plan that includes affected stakeholders in evaluation and monitoring.

3. Before You Begin

4. Retrieval Practice: Concepts from Lesson 7.1

These questions draw on Lesson 7.1. Answer each before reading the feedback.

Question 1: What is the difference between data bias and interpretation bias? Give a brief example of each.

Show answer

Answer: Data bias originates in the training data (e.g., historical underrepresentation of certain groups). Interpretation bias arises when humans misread or over-rely on AI outputs (e.g., accepting a ranking without questioning its criteria). Example: Data bias: a dataset of engineering profiles contains mostly men. Interpretation bias: a manager assumes the AI's candidate ranking is objective and ignores potential proxies.

Question 2: Why is it important to examine who is absent from AI-generated content, not just who is present?

Show answer

Answer: Absence can signal that certain groups or perspectives are not valued or are invisible, which can lead to representational harm, missed opportunities, and unequal access to visibility and voice. Omissions also reflect the values embedded in the data and workflow.

Question 3: Give one example of a default assumption that could appear in an AI-powered educational tool, and explain why it might be problematic.

Show answer

Answer: Assumption: all students have high-speed internet and a computer. This is problematic because students in rural or low-income areas may have limited connectivity or rely on mobile devices, leading to unequal access to the tool's features.

5. Opening Scenario: "AI-Powered Student Support Chatbot"

📌 Scenario — AI-Powered Student Support Chatbot

A university deploys an AI chatbot to answer student questions about enrolment, financial aid, and academic policies. The chatbot is designed to provide the same response to every student for a given query. The system is efficient and appears to reduce staff workload.

However, early feedback reveals disparities:

  • Students who are non-native English speakers receive responses that are less comprehensible, due to complex vocabulary and idiomatic expressions.
  • Students with visual impairments cannot access the chatbot's interface because it is not compatible with screen readers.
  • Students in rural areas with limited internet access experience long loading times and frequent disconnections.
  • Students with learning disabilities find the responses too text-heavy and lacking in alternative formats.
  • International students are unaware of the human-support options because the chatbot does not mention them.

The university acknowledges that the chatbot is convenient but questions whether it is equitable, inclusive, and accessible.

Questions to consider:

  1. Does the chatbot treat everyone equally? Does that equal treatment produce equitable outcomes?
  2. What barriers are affecting different groups? How do these barriers interact?
  3. Who is included in the chatbot's design and testing? Who is missing?
  4. What would make this system fairer and more inclusive?
Show guided analysis

1. Equality vs. equity: The chatbot provides the same response to everyone (equality), but this does not account for different needs, languages, or access barriers—so outcomes are not equitable.

2. Barriers: Language complexity, visual interface, connectivity, cognitive load, and lack of information about human alternatives are all barriers that affect different users.

3. Intersectionality: A student who is both a non-native English speaker and has a visual impairment faces overlapping barriers that are greater than the sum of each individually.

4. Inclusion gap: The design and testing likely did not involve users with disabilities, non-native speakers, or rural students, so their needs were not accounted for.

5. Fairer approach: Provide multilingual support, simplify language, make the interface accessible, offer offline/low-bandwidth options, and prominently inform users about human-support alternatives.

6. Equality, Equity, Fairness, Inclusion, and Accessibility

These terms are often used interchangeably, but they have distinct meanings in a sociotechnical context. Understanding these differences is essential for designing responsible AI systems.

Table 1. Distinguishing key concepts
Concept Definition AI-supported example
Equality Providing the same resource, process, or treatment to everyone. Every student receives the same AI-generated response to a question.
Equity Accounting for different needs, barriers, starting points, or effects to ensure fair outcomes. The chatbot offers simplified language for non-native speakers and audio alternatives for visually impaired users.
Fairness Evaluating whether treatment, process, performance, and outcomes are justifiable in context. Regular audits compare error rates and service quality across user groups, and decisions are made to adjust the system.
Inclusion Enabling meaningful participation, representation, access, and influence for all stakeholders. Users with disabilities and non-native speakers are involved in testing and governance of the chatbot.
Accessibility Designing information, systems, and services so that people with varied abilities and access needs can use them. The chatbot interface meets WCAG standards, provides text alternatives, and works with screen readers.
Key insight: Equality is about sameness; equity is about fairness. Treating everyone the same can actually produce unfair outcomes when people start from different positions or face different barriers.
Section summary
  • Equality and equity are distinct; equal treatment does not guarantee fair outcomes.
  • Fairness is contextual and requires evaluation of treatment, process, and outcomes.
  • Inclusion and accessibility are necessary for fairness to be meaningful.

7. Fairness Criteria May Conflict

Fairness is not a single, universal metric. Different stakeholders may prioritise different criteria, and these can conflict. For example:

There is no single "correct" answer—fairness requires explicit justification, stakeholder involvement, and context-sensitive trade-offs.

🧩 Quick reflection

Think of a situation where two fairness criteria might conflict. Which would you prioritise, and why?

Show example

Example: In a hiring tool, we might prioritise equal treatment (all candidates are processed the same way) over equitable support (extra help for candidates from underrepresented backgrounds). However, if the tool's criteria are biased, equal treatment may perpetuate inequality. A fairer approach might be to provide optional, transparent explanations and allow candidates to correct errors.

Section summary
  • Fairness criteria are not always compatible; trade-offs are inevitable.
  • Trade-offs require explicit justification and stakeholder involvement.
  • Different contexts call for different prioritisation.

8. Accessibility Barriers

Accessibility is an integral part of inclusion and fairness. Barriers can be grouped into several categories:

An AI-generated accommodation (e.g., a summary) must still be checked for accuracy and suitability—automation alone does not guarantee accessibility.

Important: Accessibility is not a one-time checkbox; it requires ongoing testing with real users who have diverse needs, and it must be integrated into the design and development process from the start.
Section summary
  • Accessibility barriers span visual, hearing, motor, cognitive, language, literacy, connectivity, cost, authentication, time, format, and human-support dimensions.
  • Accessibility requires ongoing testing with real users.
  • Automated accommodations must themselves be verified for quality and suitability.

9. Intersectionality

Definition — Intersectionality

Intersectionality is a concept that recognises that people may experience overlapping social positions and barriers that cannot be understood by examining one characteristic at a time. For example, a person may experience discrimination or disadvantage due to the combination of race, gender, disability, and socioeconomic status—not just each factor in isolation.

In the context of AI-supported systems, intersectionality means that:

Cautions:

Practical takeaway: When evaluating fairness, ask: Who might be affected by the combination of multiple barriers? This helps uncover hidden inequities.
Section summary
  • Intersectionality recognises that overlapping identities and barriers produce unique experiences.
  • Single-dimension fairness assessments can miss compounded harms.
  • Use intersectionality to surface potential harms, not to stereotype.

10. Uneven Benefits, Burdens, and Participation

Fairness analysis requires examining who benefits from an AI system, who bears burdens, who is missing from the conversation, and who holds authority.

An equitable system does not just avoid harm—it actively works to distribute benefits fairly and to include those who have been historically marginalised.

Section summary
  • Fairness requires examining benefits, burdens, participation, and authority.
  • An equitable system actively works to distribute benefits and include the marginalised.
  • Missing stakeholders represent a governance and fairness gap.

11. The Role of Human Authority

AI systems should not exercise final, accountable authority over decisions that affect people's rights, opportunities, or well-being. Human authority must be maintained in several ways:

AI can assist, but it cannot replace human judgment, especially in contexts involving ethics, care, or justice.

Section summary
  • Human authority over consequential decisions is non-negotiable.
  • Human review, accountability, contestability, and transparency are essential safeguards.
  • AI assists; it does not replace ethical judgment or accountability.

12. Process Walkthrough: Users, Barriers, Fairness Criteria, Accessibility, and Participation

This walkthrough applies the concepts of fairness, equity, accessibility, and intersectionality to the student-support chatbot scenario. It shows how to systematically examine user needs, barriers, and fairness criteria.

Table 2. Fairness and inclusion walkthrough for the chatbot
Stage Action Fairness/inclusion check Intervention
1. Identify users and non-users List all who interact with the chatbot, including those who may avoid it. Are there users who are excluded or self-exclude? (e.g., non-native speakers, visually impaired, rural students). Survey students to understand usage patterns and barriers.
2. Map barriers For each user group, identify accessibility barriers (language, visual, connectivity, etc.). How do barriers overlap for individuals with multiple characteristics? Create an intersectional barrier map.
3. Define fairness criteria Select the criteria that matter most (e.g., equal access, comparable error rates, equitable support). Do criteria conflict? Which ones are prioritised, and why? Engage stakeholders in priority-setting.
4. Assess equity Compare outcomes across groups (e.g., response quality, completeness, user satisfaction). Are outcomes equitable? Do some groups receive lower-quality service? Adjust system to provide additional support where needed.
5. Evaluate inclusion Examine who was involved in design, testing, and governance. Were affected users included? Were their perspectives valued? Establish a user advisory group.
6. Ensure accessibility compliance Test the system against accessibility standards (WCAG, etc.) and with assistive technologies. Are there remaining barriers? Are accommodations effective? Implement fixes and provide alternative human-support channels.
7. Design participation Create mechanisms for ongoing user input and feedback. Is participation meaningful, or is it tokenistic? Co-design improvements with users.
8. Assign human authority Identify who is responsible for oversight, complaints, and corrections. Is there a clear person or team accountable? Designate a human owner and provide training.
Walkthrough summary
  • Fairness requires systematic examination of users, barriers, criteria, and participation.
  • Accessibility and intersectionality must be considered from the start.
  • Human authority must be assigned and operational.

13. Worked Example: AI-Generated Health Information for Diverse Populations

Context: A public health agency deploys an AI chatbot to answer questions about vaccine eligibility, symptoms, and clinic locations. The chatbot uses a single, standardised response for each query, generated from official guidelines.

Concern: The chatbot's responses are written at a high literacy level and are available only in English. Users who are older adults, have low literacy, or speak other languages may not understand the information. Additionally, the chatbot does not provide audio or large-text options.

Worked Fairness Analysis

  1. Equality: All users receive the same response text (equal treatment).
  2. Equity: This equal treatment does not account for differences in literacy, language, or sensory needs. Some groups are likely to misunderstand or miss critical information.
  3. Fairness criteria: The agency must decide whether to prioritise equal access (everyone can use the chatbot) or equitable outcomes (all users understand the information). These may conflict if simplifying the text for some groups reduces precision for others.
  4. Accessibility barriers: Language (English only), cognitive (high literacy level), visual (no audio, small text), and possibly connectivity (if video is used).
  5. Intersectionality: An older adult with low literacy and limited digital skills faces multiple barriers that make the chatbot nearly unusable, even though each barrier alone might be manageable.
  6. Inclusion: The system was developed by subject-matter experts but did not involve community members or user testing with target populations.
  7. Participation: There is no mechanism for users to provide feedback or request human assistance.
  8. Human authority: No clear human owner is designated to handle complaints or correct misinformation.
  9. Mitigations:
    • Provide versions of responses at multiple literacy levels.
    • Offer translations in the most common languages of the community.
    • Implement audio-read-aloud features and high-contrast text.
    • Include a prominent option to speak with a human health navigator.
    • Establish a user advisory group to co-design improvements.
  10. Recommendation: Proceed with a pilot that includes these mitigations and a plan for ongoing evaluation and user feedback.
Show key insights from this example
  • Equal treatment does not guarantee equitable outcomes.
  • Accessibility is not an afterthought—it must be built in from the start.
  • Intersectional analysis reveals compounded disadvantages that single-dimension checks miss.
  • Meaningful participation requires involving affected communities, not just experts.

14. Guided Learning Activities

🧩 Activity 7.2.A: Equality, Equity, Fairness, Inclusion, or Accessibility?
  • Purpose: To practise distinguishing the five concepts.
  • Learning outcome: Distinguish equality, equity, fairness, inclusion, and accessibility.
  • Estimated time: 15 minutes.
  • Instructions: For each scenario below, identify which concept(s) are primarily at play and justify your answer.
  1. An AI tutoring system provides extra practice problems to students who scored lower on the initial diagnostic.
  2. A university ensures that all students receive the same AI-generated study plan.
  3. A company's AI hiring tool is tested with candidates from diverse backgrounds, and error rates are compared to ensure they are similar.
  4. A health app provides text-to-speech and high-contrast modes for users with visual impairments.
  5. A city's AI-powered public information system includes a feedback button that asks residents about their experience, and the responses are used to adjust the system.
Show sample classifications
  • 1. Equity — providing extra support based on need.
  • 2. Equality — providing the same treatment to all.
  • 3. Fairness — evaluating outcomes (error rates) across groups.
  • 4. Accessibility — designing for different abilities.
  • 5. Inclusion — enabling participation and feedback.
🧩 Activity 7.2.B: Fairness-Criteria Comparison
  • Purpose: To explore how different fairness criteria may conflict.
  • Learning outcome: Recognize that fairness criteria may conflict and require justification.
  • Estimated time: 15–20 minutes.
  • Scenario: An AI system recommends course schedules for university students. It can prioritise either (a) giving every student the same number of electives, or (b) giving students with demanding majors more flexibility.
  • Task: Which criterion would you choose? What are the trade-offs? Who should be involved in the decision?
Show possible analysis

Possible stance: I would prioritise flexibility for students with demanding majors (equity) because equal treatment (same number of electives) does not account for different workload capacities. However, this might disadvantage students in less demanding majors who also need electives. Trade-offs: Some students may perceive unfairness; communication about the rationale is essential. Stakeholders: Students, advisors, faculty, and disability/accessibility offices should be consulted.

🧩 Activity 7.2.C: Accessibility Barrier Audit
  • Purpose: To practise identifying accessibility barriers in AI-generated content.
  • Learning outcome: Identify accessibility barriers involving language, format, disability, technology, cost, etc.
  • Estimated time: 20 minutes.
  • Materials: A sample AI-generated response (provided below).
  • Task: Audit the following response for accessibility barriers. List at least five different barriers and propose a mitigation for each.

Synthetic AI response:

"To apply for financial aid, please complete the online form available at the link below. You will need to upload your most recent tax documents and provide your student ID. If you have questions, you can chat with our virtual assistant. The application deadline is March 15 at 5:00 PM EST. Late applications will not be accepted."
Show sample audit
  • Barrier 1: Complex language ("upload your most recent tax documents") — may be difficult for users with low literacy or non-native speakers. Mitigation: Provide a simplified version with step-by-step instructions.
  • Barrier 2: Only a link (no alt text) — screen reader users may not know the link's purpose. Mitigation: Include descriptive link text and a text-based alternative.
  • Barrier 3: Time-sensitive deadline (March 15, 5:00 PM EST) — users in different time zones may misinterpret the deadline. Mitigation: Show the deadline in the user's local time.
  • Barrier 4: "Chat with our virtual assistant" — assumes the user can chat online and has the necessary interface. Mitigation: Provide a phone number or email for human support.
  • Barrier 5: No mention of accessibility accommodations (e.g., alternative formats). Mitigation: Include a statement about available accommodations and how to request them.
🧩 Activity 7.2.D: Intersectional Scenario Analysis
  • Purpose: To explore how multiple barriers interact.
  • Learning outcome: Explain intersectionality conceptually and identify how barriers may interact.
  • Estimated time: 15–20 minutes.
  • Task: Using the opening scenario (student support chatbot), describe how a student who is both a non-native English speaker and has a learning disability might experience the chatbot differently from a student who has only one of these characteristics. What additional barriers would they face?
Show sample analysis

Analysis: A student who is a non-native English speaker may struggle with complex vocabulary and idioms. A student with a learning disability (e.g., dyslexia) may also struggle with dense text and cognitive overload. When both characteristics are present, the student may find the chatbot's responses not only difficult to understand but also cognitively exhausting, leading to frustration and abandonment. They may also be less likely to seek human help because they are unsure how to ask. The combination creates a compounded barrier that is not captured by considering language or learning disability alone. Intervention: Provide clear, simple language with audio and visual aids, and offer a direct human-assistance option.

🧩 Activity 7.2.E: Participation Plan
  • Purpose: To design meaningful stakeholder participation.
  • Learning outcome: Identify who should participate in defining needs, testing, reviewing, monitoring, correcting, and deciding.
  • Estimated time: 20 minutes.
  • Task: For the public health chatbot (worked example), propose a participation plan that includes affected stakeholders. Describe who would be involved, at what stages, and how their input would be used.
Show sample participation plan
  • Design phase: Recruit a diverse advisory panel including older adults, non-native English speakers, low-literacy individuals, and people with visual or cognitive disabilities. Conduct co-design workshops to understand their needs.
  • Testing phase: Perform usability testing with the same groups, observing how they interact with the chatbot and collecting feedback on clarity and accessibility.
  • Review phase: Establish a quarterly review committee that includes community representatives to evaluate performance data and complaint trends.
  • Monitoring phase: Implement a feedback mechanism within the chatbot that allows users to report issues and suggest improvements; review these reports regularly.
  • Correction phase: Create a clear process for users to request human review or correction of misinformation; ensure that the process is widely communicated.
  • Decision phase: Include community representatives in decisions about whether to continue, modify, or stop the chatbot.

15. Exercises

📝 Exercise 7.2.1: Distinguish Equality and Equity
  • Difficulty: Introductory
  • Estimated time: ~10 min
  • Learning outcome: Distinguish equality from equity.

Scenario: An AI-powered job-matching platform provides all job seekers with the same set of recommended job listings based on their resumes.

Task: Explain how this approach is equal but may not be equitable. Provide one example of an inequity that could arise.

Show sample answer

Answer: Equal treatment—all candidates receive the same recommendations. However, this may not be equitable if candidates have different levels of work experience, education, or access to networking opportunities. For example, a candidate who graduated from a less prestigious university may receive recommendations for lower-level jobs, even if they have strong skills, because the algorithm's criteria are based on historical data that favours certain institutions.

📝 Exercise 7.2.2: Compare Fairness Criteria
  • Difficulty: Intermediate
  • Estimated time: ~12 min
  • Learning outcome: Compare possible fairness criteria and identify conflicts.

Task: List three fairness criteria that might apply to an AI system for allocating public housing assistance. For each, explain why it might conflict with another criterion.

Show sample answer
  • Equal access: Everyone who qualifies can apply and receive a response. Conflict: may not account for urgent needs.
  • Equitable support: Those with the greatest need receive priority. Conflict: may be perceived as unfair by those with less urgent needs.
  • Consistent procedures: Same process for all applicants. Conflict: may not allow for individual circumstances (e.g., disability, language barriers).
📝 Exercise 7.2.3: Identify Inclusion Gaps
  • Difficulty: Intermediate
  • Estimated time: ~10 min
  • Learning outcome: Identify inclusion gaps in an AI-supported process.

Scenario: A university's AI-powered academic advising system is designed by a team of faculty and IT staff, with no student involvement.

Task: What inclusion gaps exist? Who is missing, and why does that matter?

Show sample answer

Answer: Students are missing from the design process. Their perspectives on usability, relevance, and fairness are not considered. This matters because students are the primary users, and their needs (e.g., accessibility, language, cultural context) may not be understood by faculty and staff. Without student input, the system may be ineffective or even harmful for some groups.

📝 Exercise 7.2.4: Audit Accessibility
  • Difficulty: Intermediate
  • Estimated time: ~15 min
  • Learning outcome: Identify accessibility barriers in AI-generated content.

Task: Audit the following AI-generated text for accessibility barriers, and propose improvements.

Text: "Please refer to the attached PDF for detailed instructions. The PDF contains tables and graphs that illustrate the key metrics. You can download the PDF by clicking the link below."

Show sample audit
  • Barrier 1: Assumes user can open a PDF (may not be accessible to screen readers if not tagged). Improvement: Provide an HTML version with structured headings.
  • Barrier 2: Tables and graphs may not be accessible to users with visual impairments. Improvement: Provide text summaries of the data.
  • Barrier 3: "Click the link" assumes mouse use. Improvement: Use "follow the link" and ensure keyboard accessibility.
📝 Exercise 7.2.5: Analyse Intersecting Barriers
  • Difficulty: Advanced
  • Estimated time: ~15 min
  • Learning outcome: Analyse how multiple barriers can interact.

Scenario: An AI-powered telehealth platform requires users to schedule appointments through a web interface, verify identity using a photo ID, and complete a survey about symptoms. The platform is text-based and available only in English.

Task: Describe how a user who is an older adult, has limited English proficiency, and has low digital literacy might experience compounded barriers. Propose one design change to reduce this intersectional burden.

Show sample answer

Compounded barriers: The user may struggle with the English-only text, find the web interface confusing, and have difficulty taking a photo of their ID. Each barrier is manageable alone, but together they create a significant obstacle that may prevent them from accessing telehealth services. Design change: Offer a phone-based scheduling option with multilingual staff, or provide a simplified, image-based interface with voice guidance.

📝 Exercise 7.2.6: Design Meaningful Participation
  • Difficulty: Advanced
  • Estimated time: ~15 min
  • Learning outcome: Design a participation plan for an AI-supported system.

Task: For a municipal AI system that handles citizen service requests (e.g., pothole reporting, noise complaints), propose a participation plan that includes affected citizens, workers, and community organisations.

Show sample answer
  • Citizens: Form a community advisory group with representatives from different neighbourhoods, language groups, and disability organisations. Run regular town halls to gather feedback.
  • Workers: Involve city employees who handle service requests in testing and monitoring; they can provide insights on how the AI affects their workflow and interactions with citizens.
  • Community organisations: Partner with local non-profits that serve vulnerable populations to co-design the system and help disseminate information.
  • Feedback loop: Create a transparent dashboard showing how citizen feedback has been used to make changes.

16. Homework Problems

Formative · Ungraded · Recommended preparation for Assignment 3.

🧠 Problem 7.2.1: Equality vs. Equity
  • Estimated time: ~8 min

Task: Provide an example from education or work where providing the same AI-supported resource to all users would be equitable, and another where it would be inequitable. Explain why.

Show sample answer

Equitable: Providing all students with the same AI-generated lecture summary (if the summaries are accurate and accessible)—no one is disadvantaged by having the same information.

Inequitable: Providing the same AI-generated study plan to all students, regardless of their prior knowledge or learning style—some will need more support, and others may find it too basic.

🧠 Problem 7.2.2: Fairness Criteria Conflict
  • Estimated time: ~10 min

Task: Describe a situation in which equal treatment and equitable outcomes conflict in an AI-supported system. Which would you prioritise, and why?

Show sample answer

Example: In an AI-powered loan-approval system, equal treatment would mean applying the same criteria to all applicants. However, if the criteria are based on historical data that penalises certain groups, equal treatment produces inequitable outcomes. I would prioritise equitable outcomes by auditing and adjusting the criteria, even if that means treating some applicants differently (e.g., considering alternative credit data).

🧠 Problem 7.2.3: Accessibility Barriers
  • Estimated time: ~10 min

Task: Identify three accessibility barriers that might be present in an AI-powered job-application screening system. For each, propose a mitigation.

Show sample answer
  • Barrier: The system requires a CV in a specific format (e.g., PDF) that may not be accessible. Mitigation: Accept multiple formats and provide guidance on accessible document creation.
  • Barrier: The system uses automated video interviews that may not be accessible to deaf or hard-of-hearing applicants. Mitigation: Provide captioning and allow text-based alternatives.
  • Barrier: The system has a time limit for completing assessments, disadvantaging users with cognitive disabilities. Mitigation: Offer extended time or untimed options upon request.
🧠 Problem 7.2.4: Inclusion Gap
  • Estimated time: ~8 min

Task: Give an example of a group that might be excluded from an AI-supported public service because their needs were not considered. Explain what the consequence of that exclusion might be.

Show sample answer

Example: Indigenous communities with limited internet access may be excluded from an AI-powered government benefits portal. Consequence: they may miss out on critical financial support, widening existing inequalities.

🧠 Problem 7.2.5: Intersectionality
  • Estimated time: ~10 min

Task: Explain why considering intersectionality is important when evaluating the fairness of an AI system. Give a concrete example of how focusing only on one identity dimension could miss important harms.

Show sample answer

Importance: People have multiple, overlapping social positions that interact to produce unique experiences of advantage or disadvantage. A system that appears fair when examined by gender alone might be unfair to women of colour with disabilities, because the combination of biases may compound.

Example: An AI hiring tool might be calibrated to reduce gender bias but still exhibit racial bias in certain geographic regions. If the tool is only audited for gender, the racial bias may go undetected, harming applicants from minority racial groups.

🧠 Problem 7.2.6: Participation
  • Estimated time: ~10 min

Task: Describe what "meaningful participation" means in the context of AI fairness, and give an example of symbolic participation (i.e., participation that does not lead to real influence).

Show sample answer

Meaningful participation: Affected stakeholders are actively involved in decision-making, their input is taken seriously, and they have the power to influence outcomes.

Symbolic participation: A company holds a public consultation but does not use the feedback to make any changes, or only asks for feedback after decisions are already made.

🧠 Problem 7.2.7: Human-Service Alternative
  • Estimated time: ~8 min

Task: Why is it important to provide a human-service alternative alongside an AI-supported system? Give an example of when a user might need human assistance.

Show sample answer

Importance: AI systems have limitations—they can make errors, misunderstand context, and lack empathy. A human alternative ensures that users who are unable to use the AI, who have complex needs, or who prefer human interaction are not excluded.

Example: A student with a unique disability accommodation may need to discuss their situation with a human advisor, as the AI system may not have the flexibility to handle their specific needs.

🧠 Problem 7.2.8: Transfer to a New Domain
  • Estimated time: ~12 min

Task: Choose a domain (e.g., legal aid, social work, environmental monitoring) and describe how the concepts of equity, accessibility, and intersectionality apply to AI use in that domain. Provide a concrete example of an equity concern.

Show sample answer

Domain: Legal aid. An AI chatbot helps low-income individuals with basic legal questions (e.g., tenant rights, family law). Equity concern: The chatbot is text-based and uses complex legal terms, making it inaccessible to users with low literacy or non-native English speakers. Intersectionality: a single mother with limited English and a disability may be unable to get the help she needs, leading to further legal disadvantage.

17. Mini-Project: Inclusive and Accessible AI Use Review

📁 Project: Inclusive and Accessible AI Use Review
  • Purpose: To apply the concepts of fairness, equity, accessibility, and intersectionality to a bounded synthetic case, and to design a participation plan.
  • Learning outcomes: Distinguish equality and equity; compare fairness criteria; identify inclusion gaps; audit accessibility; analyse intersecting barriers; design meaningful participation.
  • Estimated effort: 90–120 minutes.
  • Context choices: Use the opening scenario (student support chatbot) or a case of your own (synthetic).

Required deliverable: A structured review (approximately 600–900 words or equivalent) containing:

  1. Intended users
  2. Potentially affected non-users
  3. Equality analysis — what is the same for all users?
  4. Equity analysis — what differences in needs, barriers, or outcomes exist?
  5. Fairness criteria — list the criteria that matter most and explain potential conflicts.
  6. Inclusion analysis — who is included in design and testing, and who is missing?
  7. Accessibility analysis — identify at least four barriers and propose mitigations.
  8. Language and literacy considerations
  9. Technology and cost barriers
  10. Intersecting barriers — provide one example of overlapping barriers.
  11. Human-service alternative — describe what human support exists.
  12. Participation plan — who should be involved, at what stages, and how.
  13. Evidence required — what additional information would help?
  14. Proposed changes — specific recommendations for improvement.
  15. Residual limitations — what remains uncertain?
  16. Preliminary recommendation — proceed, proceed with changes, pilot, or redesign?

Privacy and respectful-use boundaries: Use only synthetic, public, or generalised information. Do not use real personal, medical, or employment records. Do not ask the AI to infer protected characteristics.

Self-review checklist

  • Have I distinguished equality from equity with a concrete example?
  • Have I identified at least two fairness criteria and considered how they might conflict?
  • Have I identified at least four accessibility barriers with proposed mitigations?
  • Have I provided an intersectional example of overlapping barriers?
  • Have I described a meaningful participation plan?
  • Have I described human-service alternatives?
  • Have I made a clear, conditional recommendation?
Show solution architecture
  1. Scope: Define the system and its purpose.
  2. Stakeholder mapping: Identify all users and non-users, considering demographics, abilities, and access.
  3. Equality vs. equity: Compare the uniform treatment with the actual outcomes for different groups.
  4. Fairness criteria: Select and justify which criteria are most relevant.
  5. Barrier audit: Systematically list barriers across categories (visual, cognitive, language, etc.).
  6. Intersectional analysis: Identify overlapping barriers and their compounded effects.
  7. Participation design: Propose how to involve stakeholders meaningfully.
  8. Mitigations: For each barrier, suggest a concrete fix.
  9. Residual uncertainty: Acknowledge what you cannot know without testing.
  10. Recommendation: Make a conditional, evidence-based decision.
Show illustrative partial example (different context)

Context: AI-powered library search and recommendation system for a public library.

Intended users: Library patrons of all ages and backgrounds.

Non-users: Patrons with visual impairments, non-native speakers, those without internet access.

Equality: All users see the same search interface and recommendation algorithm.

Equity: Some users struggle with the interface (small text, English only, complex navigation).

Fairness criteria: Equal access vs. equitable outcomes (e.g., all users should find relevant materials).

Inclusion: The system was designed by IT staff and librarians, but no patrons were involved.

Accessibility barriers: Text too small, no screen reader compatibility, no language selection, requires high-speed internet.

Intersecting barriers: An older patron with low vision and limited digital literacy may be unable to use the system.

Human alternative: Librarians are available at the desk, but the system does not prominently direct users to them.

Participation plan: Form a patron advisory group including seniors, non-native speakers, and disability advocates. Conduct usability testing. Use feedback to adjust the interface and add features.

18. Self-Check Quiz

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

Question 1: What is the primary difference between equality and equity in the context of AI-supported systems?

Select one:
Show answer and explanation

Correct answer: b.

Explanation: Equality means sameness of treatment or resources; equity means adjusting to ensure fair outcomes given different starting points and barriers. The other options are either incorrect or conflate the terms.

Question 2: Which of the following are accessibility barriers that can appear in AI-generated content? Select all that apply.

Select all that apply:
Show answer and explanation

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

Explanation: Each of these can be a barrier for different users. Accessibility is broad and includes language, sensory, cognitive, and technical factors.

Question 3: Match each scenario to the concept it best illustrates.

Table 3. Matching scenarios to concepts
Scenario Concept
1. A university provides extra tutoring to first-generation students. A. Equality
2. A website offers its AI chatbot in multiple languages. B. Equity
3. An AI system gives the same response to all users. C. Accessibility
Show answer and explanation
  • 1 — B. Equity: Extra support based on need.
  • 2 — C. Accessibility: Language options address a barrier.
  • 3 — A. Equality: Same response for all.

Question 4: What does intersectionality mean in the context of AI fairness, and why is it important?

Show answer and explanation

Answer: Intersectionality recognises that people have multiple, overlapping social identities and barriers that interact to produce unique experiences of advantage or disadvantage. It is important because focusing on a single dimension (e.g., gender or race alone) can miss compounded effects, leading to incomplete fairness assessments.

Question 5: True or false: Providing the same AI-generated information to all users is always the fairest approach.

Select one:
Show answer and explanation

Answer: False.

Explanation: While equal treatment is one form of fairness, it may not be equitable if users have different needs, backgrounds, or access barriers. Fairness often requires differentiated support to achieve comparable outcomes.

Question 6: Give an example of how two fairness criteria might conflict, and suggest one way to resolve the conflict.

Show answer and explanation

Example: Equal access (everyone can use the system) vs. equitable support (extra help for disadvantaged groups). They conflict if providing extra support requires resources that could be used for universal access. Resolution: Prioritise equitable support for critical needs while ensuring a baseline of access for all; monitor outcomes and adjust.

Question 7: Which of the following is an example of a procedural fairness concern?

Select one:
Show answer and explanation

Correct answer: b.

Explanation: Procedural fairness concerns the processes and mechanisms for decision-making, including transparency, consistency, and the ability to challenge. Option a is a quality-of-service concern, c is allocation, and d is transparency (which can be a procedural issue but is more about explanation). Option b directly addresses contestability.

Question 8: Why is it important to have human authority over AI decisions, rather than allowing the AI to make final decisions?

Show answer and explanation

Answer: AI systems lack moral agency, contextual judgment, and accountability. Human authority ensures that decisions affecting people's rights and well-being are subject to ethical review, oversight, and correction. It also provides a clear point of responsibility for outcomes and complaints.

19. Assignment Connections

Assignment 3: Responsible AI Case Analysis

Lesson 7.2 builds evidence for Assignment 3: Responsible AI Case Analysis. Preserve the following:

  • Fairness criteria — which criteria you considered and why.
  • Equality and equity analysis — differences in treatment and outcomes.
  • Accessibility barriers — specific barriers you identified and proposed mitigations.
  • Intersectional considerations — how overlapping barriers may affect certain groups.
  • Participation plan — who should be involved and how.
  • Human-service alternatives — what human support exists or should exist.

These elements will directly inform the fairness and inclusion sections 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 did you previously understand the difference between equality and equity? Has this changed?
  2. What fairness criteria do you think are most important for AI systems in your field, and why?
  3. Recall an AI-generated resource you have used. What accessibility barriers did you notice (or not notice)?
  4. How might intersectionality affect the fairness of an AI system you are familiar with?
  5. What does meaningful participation look like to you? How could it be implemented in practice?
  6. How does the concept of human authority shape your view of responsible AI use?

Reflection planning checklist:

  • I have distinguished equality from equity with a concrete example.
  • I have identified at least two fairness criteria and considered how they might conflict.
  • I have identified an accessibility barrier in an AI system I have encountered.
  • I have explained intersectionality and given an example of overlapping barriers.
  • I have reflected on who should participate in AI governance.

20. Key Takeaways

Remember these points
  • Equality ≠ Equity: Equal treatment does not guarantee equitable outcomes; fairness often requires differentiated support.
  • Fairness criteria conflict: Different stakeholders may prioritise different criteria (e.g., equal access vs. equitable outcomes), requiring transparent trade-offs.
  • Accessibility is integral: Barriers related to language, disability, technology, and cost must be addressed to achieve fairness.
  • Intersectionality matters: Overlapping barriers create compounded disadvantages that single-dimension analyses miss.
  • Inclusion requires participation: Affected stakeholders must be involved in design, testing, monitoring, and decision-making.
  • Human authority is essential: AI systems should not exercise final authority; human review, accountability, and contestability are necessary.

21. Key Terms

Equality
Providing the same resource, process, or treatment to everyone.
Equity
Accounting for different needs, barriers, starting points, or effects to ensure fair outcomes.
Fairness
Evaluating whether treatment, process, performance, and outcomes are justifiable in context.
Inclusion
Enabling meaningful participation, representation, access, and influence for all stakeholders.
Accessibility
Designing information, systems, and services so that people with varied abilities and access needs can use them.
Intersectionality
The recognition that people experience overlapping social positions and barriers that interact and cannot be understood in isolation.
Fairness criteria
Standards used to evaluate fairness, such as equal access, equal treatment, equitable support, comparable error rates, etc.
Procedural fairness
Fairness of the processes and mechanisms for decision-making, including transparency, consistency, and contestability.
Participation
The involvement of affected stakeholders in design, testing, monitoring, and governance of AI systems.
Human authority
The principle that final decisions affecting people's rights and well-being must be subject to human oversight, accountability, and correction.

22. Looking Ahead: Lesson 7.3

In Lesson 7.2, you have explored fairness, equity, accessibility, and intersectionality. Next, in Lesson 7.3: Differential Performance, Proxy Variables, and Disparate Impact, you will examine how AI systems perform across different groups, how proxy variables can reproduce inequality, and how to identify disparate impact even when intent is absent.

After Lesson 7.3, you will move to Lesson 7.4: FAIRER, Participation, Contestability, and Remedy, where you will apply the complete FAIRER framework and design comprehensive fairness reviews.

Continue to Lesson 7.3: Differential Performance, Proxy Variables, and Disparate Impact Return to Unit 7 Home

23. Sources & Further Reading

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

👤