Bias in AI

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Understanding
AI Bias

GRADES 6–12

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

How does AI bias occur?

By the end of the lesson, you will be able to answer this question:

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Define AI bias.

Understand how AI bias happens.

Reflect on ways to reduce AI bias.

By the end of the lesson, I will be able to:

1

2

3

Open Ended

Name one of the things you will be able to do by the end of this lesson.

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

When computer scientists create AI,
they use two different types of data:

training data and
testing data.

EXPLAIN

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Training Data

the information given to an AI to help it learn how to do a specific task.

Key Vocabulary

Multiple Choice

What is Training Data?

• Data used to check the accuracy of AI

• Information given to AI to help it learn a specific task

• Data used to test AI reliability

• Information about AI bias

Multiple Choice

What does Training Data help AI with?

• Making decisions based on biases

• Checking other AI systems

• Learning how to perform specific tasks

• Testing the AI reliability

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Testing Data

the information used to check whether the AI that was created is reliable and accurate.

Key Vocabulary

Multiple Choice

What is Testing Data?

• Data used to check the accuracy of AI

• Information given to AI to help it learn a specific task

• Data used to test AI reliability

• Information about AI bias

Multiple Choice

Why is Testing Data important?

• To help AI learn a specific task

• To check if AI systems are biased

• To ensure the AI created is reliable and accurate

• To gather information about AI actions

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Imagine we're computer scientists
and are in the process of creating an AI app. The purpose of the tool we're building is to identify different types of fruits. We have some training data to help us get started.

EXPLAIN

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

FRUIT IDENTIFIER

Training Data

You use the images on the right to help train your AI.

Orange

Apple

Apple

Apple

Grapes

Orange

Orange

Banana

Grapes

Open Ended

Based on the training data we
have, what types of fruits might our AI be able to identify?

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

FRUIT IDENTIFIER

Testing Data

The images on the right are the data you are using to check that the AI works as intended.

The labels under each image
are what the AI thinks each fruit is
called.

Apple

Grape

Orange

Grape

Apple

Orange

Apple

Orange

Banana

Open Ended

• Do you notice any mistakes?
  
  Why do you think the AI is making these mistakes?

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

AI Bias

when an AI tool makes a decision that is wrong or problematic because it learned from training data that didn't treat all people, places, and things accurately

Key Vocabulary

Multiple Choice

What is AI Bias?

• Data used to check the accuracy of AI

• Information given to AI to help it learn a specific task

• Data used to test AI reliability

• Decision-making issue due to improper learning from training data

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Apple

Apple

Apple

FRUIT IDENTIFIER: BIAS

In the training data, apples were the only example of a red fruit.

The testing data shows that the AI learned to identify anything red as an apple.

The AI we created has a bias toward thinking that every red fruit is an apple.

Apple

Apple

Apple

Testing Data

Training Data

Open Ended

What are some ways we could
reduce the AI bias of this fruit detector?

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

REFLECT

What are some ways we could reduce the AI bias of this fruit detector?

●

Show it more examples of things that are NOT fruits
so that it can become better at identifying what IS a fruit

●

Show it more examples of other kinds of fruits (e.g. mangoes, kiwis, watermelons, grapefruits)

●

Show it pictures of fruits at different stages (different size, ripe vs. unripe, etc.)

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

While it's almost impossible to completely eliminate AI bias from a tool, we can do our best to reduce it by coming up with as diverse and complete a set of training data as possible.

EXPLAIN

Open Ended

Imagine you're now asked to create an AI tool that can identify weather conditions, such as sunny, rainy, or cloudy.

Describe the kind of images you would include as a part of the training data. Your goal is to come up with as complete of a data set as possible so that you limit the chance for AI bias!

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

FOLLOW-UP

●

Sky conditions: blue skies, partly cloudy, mostly cloudy, overcast

●

Times of day: sunrise, morning light, afternoon life, sunset, night

●

Seasons: spring, summer, fall, winter

●

Geographic locations: mountains, beach, desert, cities, rural towns,

●

Rarer occurrences: rainbows, lightning, tornados, hurricanes,

●

Precipitation: mist, drizzle, downpour, snow, sleet, hail

●

What else?

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.

Remember that behind every AI tool are
humans making decisions on what training data the tool will use.

Understanding how AI bias occurs can help us think critically about its potential impacts.

CLOSING REFLECTION

Multiple Choice

What are the two different types of data used when computer scientists create AI?

• Input data and output data

• Primary data and secondary data

• Training data and testing data

• Raw data and processed data

Multiple Choice

What is training data in the context of AI?

• Data used to check the reliability and accuracy of the AI

• Data given to an AI to help it learn how to perform a specific task

• Data used to identify errors in the AI

• Data that determines the final output of the AI

Multiple Choice

What is testing data used for in AI development?

• To help the AI learn a specific task

• To validate the behavior and accuracy of the AI

• To train the AI with new information

• To store the AI's learning progress

Multiple Choice

How can AI bias occur?

• When the AI is trained on diverse datasets

• When the AI makes decisions based on facts

• When the AI does not learn accurately from its training data

• When the AI is designed to be completely objective

Multiple Choice

In the context of AI, what does bias refer to?

• An AI tool making incorrect or problematic decisions based on inaccurate training data

• An AI tool always producing consistent and accurate results

• An AI tool being completely unbiased in its decision-making process

• An AI tool having access to unlimited amounts of data

Multiple Choice

What does the term "AI bias" mean?

• When an AI tool shows favoritism towards certain individuals

• When an AI tool is intentionally designed to be flawed

• When an AI tool makes decisions based on objective facts only

• When an AI tool produces inaccurate or unfair results due to its training data

Multiple Choice

How can we reduce AI bias in a tool like a fruit detector?

• Show it more examples of non-fruits and other kinds of fruits

• Only train it with images of apples to avoid confusion

• Limit the variety of images it sees to reduce complexity

• Provide it with incomplete and biased training data

Multiple Choice

Why is it important to have diverse training data when developing AI?

• To ensure the AI always makes biased decisions

• To limit the AI's exposure to different scenarios

• To reduce the chances of AI bias and improve accuracy

• To confuse the AI and make its decisions less reliable

Multiple Choice

What is a potential consequence of an AI tool having bias?

• The AI tool always making accurate and fair decisions

• The AI tool failing to learn from its training data

• The AI tool making incorrect or unfair decisions

• The AI tool becoming too objective and losing functionality

Multiple Choice

What would be a strategy to reduce bias in an AI tool designed to identify weather conditions?

• Train it only on images of sunny weather

• Include a diverse range of weather conditions in the training data

• Avoid showing the AI any weather-related images

• Provide the AI with inaccurate weather data

commonsense.org/education
Shareable with attribution for noncommercial use. Remixing is permitted.