Computing · Secondary 3 · Impacts of Computing on Society · Semester 2

Bias in AI and Algorithmic Fairness

Students will investigate how biases can be embedded in AI systems and discuss strategies for promoting fairness and equity.

• MOE Syllabus Outcomes
• MOE: Ethics and Social Issues - S3

About This Topic

Bias in AI and Algorithmic Fairness examines how training data reflecting societal prejudices can produce discriminatory outcomes in systems like facial recognition or job recruitment tools. Secondary 3 students analyze cases where skewed datasets lead to unfair results, such as higher error rates for certain ethnic groups. They explore strategies including data diversification, algorithmic audits, and transparency requirements to ensure equitable AI deployment.

This topic aligns with the MOE Computing curriculum's Ethics and Social Issues standards in the Impacts of Computing on Society unit. Students justify auditing practices and design hypothetical scenarios of AI-driven injustice, such as biased policing tools in diverse communities like Singapore's. These activities cultivate critical thinking, ethical reasoning, and civic responsibility needed for informed participation in a tech-driven society.

Active learning benefits this topic by turning complex abstractions into relatable discussions. Role-playing stakeholders affected by biased AI or collaboratively auditing sample datasets helps students internalize fairness principles, empathize with impacts, and practice real-world problem-solving with peers.

Key Questions

1. Analyze how biases in training data can lead to discriminatory AI outcomes.
2. Justify the importance of auditing AI systems for fairness and transparency.
3. Design a hypothetical scenario where AI bias could lead to significant social injustice.

Learning Objectives

• Analyze how specific biases in training data, such as demographic underrepresentation, can lead to discriminatory outcomes in AI applications like facial recognition systems.
• Evaluate the effectiveness of different strategies, such as data augmentation and algorithmic debiasing, in mitigating AI bias.
• Design a hypothetical AI system, detailing its purpose, potential biases, and proposed fairness interventions.
• Justify the necessity of ongoing AI system audits and transparency mechanisms for ensuring equitable societal impact.

Before You Start

Introduction to Artificial Intelligence

Why: Students need a basic understanding of what AI is and how it learns from data before exploring the concept of bias within AI systems.

Data Representation and Analysis

Why: Understanding how data is collected, structured, and analyzed is crucial for identifying potential biases within datasets used for AI training.

Key Vocabulary

Algorithmic Bias	Systematic and repeatable errors in a computer system that create unfair outcomes, such as privileging one arbitrary group of users over others.
Training Data	The dataset used to train an AI model. Biases present in this data can be learned and perpetuated by the model.
Fairness Metrics	Quantitative measures used to assess whether an AI system's outcomes are equitable across different demographic groups.
Algorithmic Auditing	The process of examining an AI system's algorithms and data to identify and address potential biases and ensure fairness.
Transparency	The principle of making AI systems' decision-making processes understandable and accessible, allowing for scrutiny and accountability.

Watch Out for These Misconceptions

Common Misconception

AI systems are always unbiased because they use objective math.

What to Teach Instead

Algorithms amplify biases in training data from human sources. Group dissections of datasets reveal hidden prejudices, while peer discussions correct overconfidence in tech neutrality.

Common Misconception

Bias in AI can be fully eliminated with better code.

What to Teach Instead

Fairness requires ongoing mitigation like audits, not perfection. Scenario redesign activities show trade-offs, helping students appreciate iterative strategies over quick fixes.

Common Misconception

AI bias only affects distant countries, not Singapore.

What to Teach Instead

Local examples like hiring tools disadvantaging minorities exist. Role-plays with Singapore contexts build relevance, encouraging students to connect global issues to home.

Active Learning Ideas

Small Groups: Real-World Case Audit

Provide groups with a case study on biased AI, like facial recognition failures. Students identify bias sources in data, evaluate impacts, and propose three fairness fixes. Groups share audits in a class gallery walk.

45 min·Small Groups

Pairs: Bias Scenario Redesign

Pairs design a hypothetical AI system with embedded bias leading to social injustice, then redesign it for fairness using strategies like balanced datasets. They sketch system flows and present changes.

35 min·Pairs

Whole Class: Fairness Debate

Divide class into teams to debate mandatory AI audits: one side argues benefits for equity, the other potential innovation barriers. Use structured turns and vote on strongest points.

50 min·Whole Class

Individual: Personal Bias Checklist

Students create a checklist for auditing AI fairness, drawing from class learnings. Test it on a provided algorithm example and reflect on one improvement.

25 min·Individual

Real-World Connections

• In Singapore, the Land Transport Authority (LTA) could use AI for traffic management. If training data for pedestrian detection underrepresents certain skin tones or clothing types common during festivals, the system might be less effective, potentially impacting safety for specific groups.
• Global tech companies like Google and Microsoft face scrutiny over AI recruitment tools. If these tools are trained on historical hiring data that reflects past gender or racial biases, they might unfairly screen out qualified candidates from underrepresented backgrounds.
• Facial recognition technology used by law enforcement agencies worldwide has shown higher error rates for women and people of color. This can lead to misidentification and wrongful accusations, highlighting the critical need for fairness in such sensitive applications.

Assessment Ideas

Discussion Prompt

Present students with a scenario: An AI system is developed to help Singaporean banks approve loan applications. Ask them: 'What kinds of biases might be present in the training data? How could these biases lead to unfair loan rejections for certain communities in Singapore? What steps should the bank take to ensure fairness?'

Exit Ticket

Provide students with a short description of an AI application (e.g., an AI tutor, a content recommendation engine). Ask them to identify one potential source of bias in its training data and one specific strategy they would use to make the AI fairer. Collect these as students leave the class.

Quick Check

Display a list of AI fairness strategies (e.g., data diversification, bias detection tools, human oversight). Ask students to match each strategy to a brief description of how it helps mitigate AI bias. This can be done as a short quiz or a drag-and-drop activity.

Frequently Asked Questions

What are real examples of AI bias in everyday systems?

Facial recognition often misidentifies darker skin tones due to underrepresented training data, leading to wrongful arrests. Hiring algorithms from resume data favor male candidates from certain industries. Loan approval AI perpetuates racial disparities from historical records. These cases show how unchecked data skews outcomes, underscoring the need for diverse datasets and audits in Singapore's multicultural context.

How can teachers introduce strategies for algorithmic fairness?

Start with data cleaning techniques like oversampling underrepresented groups and fairness metrics such as equalized odds. Teach auditing via tools checking disparate impact. Encourage transparency through explainable AI models. Hands-on redesigns help students apply these, justifying choices against ethical standards in MOE curriculum.

How does active learning help students grasp AI bias concepts?

Activities like group case audits and stakeholder role-plays make biases tangible by simulating real impacts. Collaborative redesigns foster debate on fixes, building empathy and critical skills. Students retain more through peer teaching and reflection, turning passive knowledge into active ethical advocacy relevant to Singapore's tech ecosystem.

Why audit AI systems for fairness and transparency?

Audits detect discriminatory patterns early, preventing harm like unequal access to services. Transparency builds public trust, vital in regulated sectors like healthcare or finance. In Singapore, it supports inclusive innovation under Smart Nation goals. Students learn to justify audits by weighing societal equity against efficiency gains.