Machine Learning and Bias

Active learning works for this topic because students need to see bias in action, not just hear about it. When they manipulate biased datasets and analyze real cases, they move from abstract worry to concrete understanding. These hands-on experiences build lasting empathy for people affected by unfair AI systems.

• Year 8
• Computing
• 4 activities
• 30 min–45 min

Learning Objectives

1. Analyze a given dataset to identify potential sources of bias that could affect an AI model's predictions.
2. Explain how societal biases can be unintentionally encoded into machine learning algorithms through data selection and feature engineering.
3. Evaluate the fairness of an AI model's output in a specific scenario, citing evidence of disparate impact on different demographic groups.
4. Propose at least two strategies for mitigating bias in machine learning models, such as data augmentation or algorithmic fairness constraints.

Simulation Game

⏱ 30 min·Small Groups

Simulation Game: Biased Data Bags

Distribute bags of colored beads with uneven distributions to represent biased datasets. In small groups, students 'train' a partner to classify new beads by majority color patterns, then test on balanced bags and record failure rates. Groups debrief on how data imbalance caused poor predictions.

Prepare & details

If an AI makes a biased decision, who is responsible: the programmer or the data?

Facilitation Tip: During Biased Data Bags, pause after each round to ask groups to share which bead colors failed most often and why.

Setup: Flexible space for group stations

Materials: Role cards with goals/resources, Game currency or tokens, Round tracker

• Apply
• Analyze
• Evaluate
• Create
• Social Awareness
• Decision-Making

Socratic Seminar

⏱ 45 min·Small Groups

Case Study Carousel: Real AI Examples

Prepare stations with cases like biased hiring tools or facial recognition errors. Groups rotate, noting bias sources (data or code), impacts, and fixes. Each group presents one insight to the class for collective discussion.

Prepare & details

Explain how we can ensure that machine learning models are fair and transparent.

Facilitation Tip: For the Case Study Carousel, assign each pair a single case to analyze deeply before rotating to the next one.

Setup: Chairs arranged in two concentric circles

Materials: Discussion question/prompt (projected), Observation rubric for outer circle

• Analyze
• Evaluate
• Create
• Social Awareness
• Relationship Skills

Socratic Seminar

⏱ 35 min·Pairs

Debate Pairs: Who Bears Responsibility?

Assign pairs to argue for programmer or data as primary bias source, using evidence from prior activities. Pairs share arguments in a whole-class debate, voting on strongest points and reflecting on shared accountability.

Prepare & details

Critique the limitations of a machine's ability to learn compared to a human.

Facilitation Tip: In Debate Pairs, give students a visible timer so they practice concise arguments within a set limit.

Setup: Chairs arranged in two concentric circles

Materials: Discussion question/prompt (projected), Observation rubric for outer circle

• Analyze
• Evaluate
• Create
• Social Awareness
• Relationship Skills

Timeline Challenge

⏱ 40 min·Individual

Timeline Challenge: Build a Fair Dataset

Individually, students select a scenario like image labeling, brainstorm diverse data sources, and sketch a balanced dataset plan. Pairs review and refine plans, then share prototypes with the class for feedback.

Prepare & details

If an AI makes a biased decision, who is responsible: the programmer or the data?

Setup: Long wall or floor space for timeline construction

Materials: Event cards with dates and descriptions, Timeline base (tape or long paper), Connection arrows/string, Debate prompt cards

• Remember
• Understand
• Analyze
• Self-Management
• Relationship Skills

Teaching This Topic

Teachers should frame bias as a data problem first, not a coding problem, to avoid the misconception that fixing the algorithm solves everything. Use analogies like ‘garbage in, garbage out’ to make the concept stick. Avoid long lectures on ethics; instead, let students discover the issues through structured tasks where bias emerges naturally from their own choices.

What to Expect

Successful learning looks like students explaining how biased data leads to unfair outcomes, not just identifying bias in a case. They should cite evidence from simulations and debates when discussing responsibility and fairness. By the end, they can suggest practical steps to reduce bias in datasets.

Watch Out for These Misconceptions

Common Misconception

During Biased Data Bags, watch for students who assume the beads represent people directly.

What to Teach Instead

After the simulation, ask students to reflect in writing: ‘What real-world group might be like the underrepresented beads?’ This helps them connect the model to societal bias without conflating beads and people.

Common Misconception

During Case Study Carousel, watch for students who generalize that all AI bias comes from bad programmers.

What to Teach Instead

After analyzing cases, use a quick card sort where students match specific bias sources (e.g., ‘unrepresentative training data,’ ‘labeling errors,’ ‘narrow problem framing’) to the cases they studied. This makes the variety of bias sources visible.

Common Misconception

During Debate Pairs, watch for students who claim machines can ‘learn like humans’ when comparing AI and human judgment.

What to Teach Instead

After the debate, ask each pair to write one sentence explaining how their analogy between human and machine learning breaks down, using an example from their debate cards.

Assessment Ideas

Discussion Prompt

After the Job Candidate Scenario discussion, ask each group to record their final consensus on responsibility and one key reason. Collect these to assess how well they connect bias sources to responsibility.

Quick Check

During the Identifying Bias in Test Scores quick-check, collect student responses and sort them into categories: ‘Correctly identified bias source’ or ‘Misidentified source.’ Use this to plan mini-lessons on data representation.

Exit Ticket

After the Fairness in Decision-Making exit-ticket, review responses to see if students can name at least one fairness technique (e.g., dataset auditing) and one limitation of AI in ethical contexts, such as lack of context.

Extensions & Scaffolding

• Challenge: Ask early finishers to design a new biased dataset that would cause a hiring AI to favor candidates with a specific name origin. They must justify their choices in a short paragraph.
• Scaffolding: For students struggling with the Build a Fair Dataset task, provide a partially completed dataset with clear gaps, such as missing age groups or underrepresented genders.
• Deeper exploration: Have students research a real-world AI fairness initiative, like MIT’s IBM Fairness 360, and present its methods in a short video or infographic.

Key Vocabulary

Algorithm	A set of rules or instructions followed by a computer to solve a problem or perform a task. Machine learning algorithms learn from data to make decisions.
Dataset	A collection of data used to train and test machine learning models. The quality and representativeness of the dataset are crucial for model performance and fairness.
Bias (in AI)	Systematic errors in an AI system that result in unfair outcomes, often reflecting societal prejudices present in the training data.
Fairness (in AI)	The principle that AI systems should not produce discriminatory or prejudiced outcomes against individuals or groups based on protected characteristics.
Feature Engineering	The process of selecting, transforming, and creating variables (features) from raw data to improve the performance of machine learning models.