AI Bias & Algorithmic Justice
Use each case to practise ethical reasoning rather than to seek a single correct answer. Ākonga must distinguish evidence from assumption, listen for affected perspectives, and revise a claim when counter-evidence is stronger.
- Scaffold: Structured academic controversy with claim, evidence, counterclaim, and revision prompts
- Seminar outcome: Revised ethical judgement explaining who bears risk and who owes action
🌅 Opening & Evidence Protocol
🎥 Media Anchor
Video: "3 types of bias in AI | Machine learning" — Google
- What bias risks appear when datasets are incomplete or unbalanced?
- Which mitigation strategy should be standard in AI design?
Evidence protocol: describe the documented case before explaining its possible cause or reach.
Opening Protocol (5 minutes)
- Locate the evidence: Name the study, investigation, or source behind each case.
- Keep its scope: Do not transfer a finding to another group, place, or system without evidence.
- Reserve the decision: Identify who is affected and who must shape, approve, challenge, or stop the system.
🎯 Learning Objectives & Success Criteria
By the end of this lesson, ākonga will be able to:
- Identify: Recognise AI bias in real-world systems and understand its sources
- Analyse: Examine how algorithmic bias affects different communities disproportionately
- Evaluate: Assess AI systems using named justice and Māori data-sovereignty sources
- Design: Propose solutions for creating more equitable AI systems
Success Criteria — Ākonga will demonstrate:
- ✓ Ability to identify bias in AI systems through case study analysis
- ✓ Understanding of how bias affects Māori and other marginalized communities
- ✓ Accurate use of a named source, including its scope and authority limits
- ✓ Creative problem-solving for algorithmic justice solutions
Phase 1: AI Bias Investigation — Uncovering Hidden Inequality (25 minutes)
Bias Detective Challenge
20 minutes investigation + 5 minutes synthesis📄 Activity Handouts: Use AI Ethics & Bias for foundational concepts, and AI Bias Lab Activity for structured investigation worksheets.
🌟 Evidence set: The four cases below support different, bounded conclusions: measured error gaps in three gender-classification systems; a reported experimental hiring tool; bias risks in New Zealand health-record data and model design; and a model of predictive-policing feedback. Do not merge them into one claim or extend them to groups the source did not study.
Real-World AI Bias Cases (15 minutes):
Working in groups of 4, investigate one of these documented cases of AI bias:
Case 1: Facial Recognition & Race
The evidence: In Gender Shades (Buolamwini & Gebru, 2018), three commercial gender-classification systems had error rates of up to 34.7% for darker-skinned women, while the maximum for lighter-skinned men was 0.8%.
Investigation Questions:
- Why might training data cause this bias?
- How does this affect surveillance and policing?
- Which affected groups are established by this study, and which local impacts would need separate evidence?
- How could this be fixed?
Case 2: Hiring Algorithms & Gender
The evidence: A 2018 Reuters investigation reported that Amazon abandoned an experimental recruiting system after it learned patterns that penalised résumés containing terms such as “women’s”.
Investigation Questions:
- How did historical hiring patterns bias the AI?
- What careers or opportunities could be affected?
- What evidence would be required before making a claim about wāhine Māori?
- What safeguards should exist?
Case 3: Bias Risks in New Zealand Health Data and Models
The evidence: Yogarajan et al. (2022) found bias arising from data and model-design choices in New Zealand health records and argued that Māori control and participation are needed so health algorithms do not worsen inequity. This study does not establish that every health-AI tool performs worse for Māori.
Investigation Questions:
- Which data and model-design risks do the authors identify?
- What evidence would be needed to assess a specific health-AI system?
- Which possible health-equity impacts are supported, and which remain hypotheses?
- What control and participation requirements do the authors identify, and who would hold the next design decision?
Case 4: Predictive Policing & Communities of Colour
The evidence: Ensign et al. (2018) modelled how predictive-policing systems updated with discovered-crime data such as arrest counts can repeatedly send police to the same neighbourhoods regardless of the true crime rate.
Investigation Questions:
- How do historical policing biases get amplified by AI?
- What evidence would be needed before applying this model’s findings to policing in Aotearoa?
- How does this relate to justice system inequality?
- What monitoring, appeal, and stop conditions would address the documented feedback-loop risk?
Bias Pattern Analysis (5 minutes):
Each group presents their case in 1 minute, then class identifies common patterns:
- Where does bias come from in AI systems?
- Who is most likely to be harmed by AI bias?
- How does AI bias connect to existing social inequalities?
- Why might companies and institutions be slow to address these issues?
Phase 2: Source, Authority & Algorithmic Justice (25 minutes)
Justice-Centred AI Evaluation Workshop
Māori data-sovereignty source check (15 minutes):
Use Te Mana Raraunga’s published principles, or a more relevant named Māori source, to answer only what that source can establish. Record the source and its scope.
Authority to control
Applied to AI: Whose data or knowledge is involved, and who does the source say may decide how it is collected, interpreted, used, or reused?
- Who currently makes those decisions?
- Who can challenge, change, or stop the system?
- Which decision remains with the affected Māori authority?
- What evidence supports each answer?
Provenance and whakapapa
Applied to AI: What provenance, purpose, context, and parties are recorded for the data and output?
- Where did the data come from?
- For what purpose was it collected?
- Which relationships or contexts are missing?
- What cannot be inferred from the available metadata?
Consent, accountability, access, and benefit
Applied to AI: What does the source require of people or organisations using the data?
- Was consent obtained, and can it be withdrawn?
- Who is accountable for harm or misuse?
- Who has access, and who receives the benefit?
- What safeguard follows directly from the named source?
Scope and uncertainty
Applied to AI: What can this evidence support, and what remains outside the source’s scope?
- Which affected groups are actually represented?
- Which impact claims need another source?
- What cultural or local decision cannot be made in class?
- How should the conclusion be limited?
Comparative Justice Framework Analysis (10 minutes):
Groups also apply one additional justice framework to their AI case:
Distributive Justice
Key Question: How are benefits and harms distributed?
- Who benefits most/least from this AI system?
- Are resources allocated fairly?
- How are risks shared across different groups?
Procedural Justice
Key Question: Are the processes fair and transparent?
- Can people understand how AI decisions are made?
- Is there accountability for AI decisions?
- Can affected people participate in AI governance?
Recognition Justice
Key Question: Are all groups recognised and valued?
- Whose knowledge and experiences are valued in AI design?
- How does the AI handle cultural differences?
- What voices are missing from AI development?
Phase 3: Design Justice Solutions — Creating Equitable AI (20 minutes)
AI Justice Design Challenge
Solution Design Process (15 minutes):
Groups redesign their AI system to embody justice principles:
Step 1: Problem Reframing (3 minutes)
- What is the real problem this AI should solve?
- Whose needs should be centred in the solution?
- What would success look like for marginalized communities?
Step 2: Inclusive Design Principles (5 minutes)
- Community Participation: How would affected communities shape AI development?
- Source and authority: Which named source sets requirements, and which decisions remain with affected people?
- Transparency: How would people understand and challenge AI decisions?
- Accountability: Who would be responsible for AI impacts and how?
Step 3: Justice Implementation (4 minutes)
- What safeguards would prevent bias?
- How would benefits be distributed fairly?
- What ongoing monitoring would ensure equity?
- How would the system address past harms?
Step 4: Authority and consent (3 minutes)
- What data or knowledge may the proposal use, and under whose permission?
- Who may approve, change, challenge, or stop the system?
- Which design claim is source-backed, and what remains for consultation?
Solution Presentations (5 minutes):
Each group presents their redesigned AI system in 1 minute, focusing on:
- The key change that makes their AI more just
- How its source, permission, and governance requirements are visible
- What makes it different from current systems
- Who would benefit most from these changes
🌅 Whakamutunga — Reflection & Closing
Algorithmic Justice Commitment (5 minutes)
Personal AI Ethics Reflection:
Students complete individual reflection:
- What is one example of AI bias that you now recognise affects your life or community?
- Which named source or justice framework most changed your evaluation, and what is its scope?
- What is one evidence-bound next step you could take within your role after kaiako review?
- How has understanding AI bias changed your perspective on technology and justice?
Closing Circle — Justice Wisdom Sharing:
Students share one insight about AI justice, followed by teacher reflection:
"Kia tōtika ai te taiao"
AI systems are not neutral — they reflect the values and biases of their creators and the data they're trained on. Understanding this gives us power to demand better systems that serve justice rather than perpetuate inequality. Your critical analysis helps build a more equitable digital future.
📊 Assessment & Next Steps
Formative Assessment — Today's Evidence:
- Bias Recognition: Accuracy in identifying and explaining AI bias in case studies
- Justice Analysis: Quality of source use, scope limits, authority mapping, and additional justice-framework analysis
- Design Thinking: Creativity and feasibility of proposed AI justice solutions
- Critical Reflection: Depth of personal and systemic understanding of AI impacts
Preparation for Lesson 3:
- AI Audit: Find one AI system you use regularly and analyse it for potential bias
- Source research: Find one public source from an organisation or authority affected by the selected AI system
- Solution Thinking: Research one organisation working on AI ethics or algorithmic justice
🛠️ Teacher Resources & Adaptations
AI Bias Case Study Resources:
- Algorithmic Justice League: Real-world bias examples and research
- MIT Technology Review: Regular AI bias reporting and analysis
- AI Now Institute: Academic research on AI social impacts
- Te Mana Raraunga: Published Māori data-sovereignty principles
Authority and source support:
- Named public sources: Use material published or approved by the relevant Māori authority
- Te Mana Raraunga: National starting point for Māori data-sovereignty analysis
- Local application: Seek guidance from the relevant authority; do not ask Māori ākonga to authenticate the task
- Justice frameworks: Keep each framework’s evidence, definitions, and limits visible
Differentiation Strategies:
- Technical Levels: Adjust case study complexity for different tech familiarity
- Cultural Connections: Include bias examples relevant to diverse student backgrounds
- Justice Frameworks: Offer choice in which additional framework to apply
- Solution Complexity: Allow varying levels of technical detail in design solutions
Kaiako Planning Snapshot
Digital Technologies / Social Studies — AI Ethics & Te Ao Māori — Years 9–10
Learning context: Digital Technologies and ethical reasoning through a te ao Māori lens (Years 9–10).
Ngā Whāinga Akoranga — Learning Intentions
- Explain how AI systems work at a conceptual level and identify their ethical dimensions
- Analyse AI bias and algorithmic justice through a te ao Māori and tikanga framework
- Evaluate the implications of AI for tino rangatiratanga, Māori data sovereignty, and digital futures
Differentiation & Proximal Guidance
Extension: investigate a real AI system for bias and propose a tikanga-aligned redesign. Scaffold: AI ethics checklist with guided prompts and vocabulary support. On-level: structured case study analysis. Entry: visual explainer of AI concepts with relatable everyday examples.
Inclusion & Accessibility
ESOL/ELL: pre-teach AI vocabulary using visual analogies and familiar examples. Neurodiverse learners: offer unplugged alternatives to all digital tasks. Accessibility: all interactive digital activities have non-screen analogue alternatives.
Audited curriculum connections
- Te Mātaiaho (2025) · Technology · Phase 4 (Years 9–10) · Knowledge: “Technological outcomes can raise ethical and legal issues such as privacy, data security, bias, and fairness, which affect individuals and society.”