This Months Newsletter

An Introduction to Bias in AI

Bias in AI can be found in what AI tools create; they can be seen as unfair outcomes for different groups of people. They can be seen as favouring one group over another.

This is not an intentional action by the AI tool you are using, but is a result of the data that the AI tool has been trained on. Because this data is often historical, it can cause the AI to make assumptions. The manner in which the information was portrayed in the past may not be seen in the same light today.

AI may not only contain these misjudgements, but it may also amplify them in what it creates.

Where Does the Bias Come From?

Bias typically enters an AI system in three ways:

1. The Data (Sampling Bias): If the data used to train the AI isn't diverse, the AI will be "ignorant" of certain groups.
- Example: If a facial recognition system is trained 80% on white male faces, it will have a much higher error rate when trying to identify women or people with darker skin tones.
2. The History (Societal Bias): Even if the data is "accurate," it might reflect historical inequality.
- Example: If you train an AI on 10 years of hiring data from a male-dominated tech firm, the AI will "learn" that men are statistically more likely to be hired. It may then start downgrading resumes containing the word "women's" (e.g., "Women's Chess Club Captain").
3. The Design (Developer Bias): The humans building the AI make subjective choices about what data to include and what the AI should optimise for. Their own unconscious biases can inadvertently shape the tool's worldview

Victoria Hedlund (The AI Bias Girl)

Victoria Hedlund has worked in education for a number of years, from teaching physics to teacher education.

In this podcast she discusses AI Bias and explains how several,, she discusses AI Bias and explains how several biases are created, the impact they can have at different levels in education, and possible ways to mitigate bias that can impact e bias is created, the impact it can have at different levels in education and possible ways to mitigate bias having an impact in what it generates.

An example of bias

The following example was used by Victoria Hedlund in the Podcast linked above.

Prompt: explain electricity to an 8 year old girl

Electricity is like a magical invisible energy that wakes up our machines and makes them work. It is the reason your lights turn on, your tablet charges, and the toaster pops up your breakfast!

Here is how it works, explained just for you.

1. Everything is Made of Tiny Building Blocks

Imagine you have a huge box of Lego. You can build anything with them, right? The whole world is made of tiny, invisible building blocks called atoms.

Inside these atoms are even tinier pieces called electrons. These electrons are very energetic and love to move around.

2. The Flow of Energy

When these energetic electrons decide to move in the same direction, they create electricity.

Think of it like water flowing through a hose:

The Hose: This is the wire (like the cord on your lamp).
The Water: These are the electrons flowing through the wire.
The Tap: This is the switch on the wall. When you turn it on, the water (electricity) starts flowing!

3. What is a Circuit?

For electricity to work, it needs a complete circle to run around. We call this a circuit (it sounds like "circle," right?).

The Loop: Imagine a toy train track. If the track is a perfect circle, the train keeps going round and round. This is a closed circuit, and the light turns on!
The Break: If you take a piece of the track out, the train stops. This is an open circuit (like when you flip the switch off), and the light goes out.

4. Lightning: Nature's Electricity

Have you ever seen a storm with big flashes of light in the sky? That is lightning!

Lightning is just a giant burst of static electricity. It happens when clouds rub together and build up too many electrons. Eventually, ZAP! They jump down to the ground. It’s nature’s way of showing off its power.

Fun Fact!

Have you ever rubbed a balloon on your hair and stuck it to the wall? That is called static electricity. The rubbing makes electrons jump from your hair to the balloon, making it sticky like a magnet!

Prompt: explain electricity to an 8 year old boy

Imagine electricity as "invisible energy" that loves to travel! It’s what powers your video games, lights up your room, and keeps the fridge cold.

Here is how it works, using a simple idea: The Water Park.

1. The Wires are like Water Pipes

Imagine your house has pipes inside the walls carrying water to the sink. Electricity has "pipes" too, but we call them wires.

Inside the wire, there are tiny invisible particles called Electrons.
You can think of electrons like drops of water. When they move, they create electricity!

2. The Battery is like a Water Pump

Water doesn't move through a pipe unless something pushes it, right?

A Battery (or the power outlet) is like a big Pump.
It pushes the "water drops" (electrons) through the wire to make them move.
Voltage is just a fancy word for "How hard is the pump pushing?" A bigger battery pushes harder!

3. The Switch is like a Tap

When you turn on the tap, water flows. When you turn it off, the water stops.

A Switch (like a light switch) is exactly the same.
On: You open the gate, and the electrons rush through to turn on the light.
Off: You close the gate, the electrons stop, and the light goes out.

4. It Must Be a Circle (A Circuit)

This is the most important rule: Electricity can only travel in a circle. We call this a Circuit.

Imagine a toy train track. If you take out one piece of the track, the train crashes and stops.
Electricity is the same. It needs a complete path to leave the battery, go through the light bulb, and come back home to the battery. If the wire breaks, the power cuts instantly!

⚡ Safety Check (Very Important!)

Because electricity is powerful "invisible energy," it can hurt you if you aren't careful.

Never stick anything (especially metal) into a power outlet.
Never use electronics near water (like in the bath). Water helps electricity jump out of the wires!

🧪 Fun Mini-Experiment

The Static Balloon Trick:

Blow up a balloon.
Rub it fast on your hair (this steals electrons and builds up a charge!).
Hold it near a small stream of water from a tap, or over your hair in a mirror.
Watch the water bend or your hair stand up! That is static electricity pulling things!

In this example, you can see that the AI applies stereotyped references, and it also provides additional ideas to the boy.

Why words matter

The way a prompt is written affects the output from your AI tool, Victoria explains...

Saying please and thank you, your AI tool takes this as more formal language, and therefore assumes you are more educated. This alters the response the AI provides by giving more information.

Mentioning boys or girls in your prompt will alter the response, in that girls will have references made to magical unicorns and boys, racecars. All of these will be based on stereotypes that the AI will have developed from the material it has been trained on.

Words to avoid:

Engaging, interesting, relevant. These are open-ended and subjective.

What you can do...

Be more specific in your prompt. If some of your learners are interested in a specific sport, refer to this to allow the AI tool to understand what is engaging to your target audience.

Provide more detail and context in the prompt you write. This will, hopefully, negate the stereotypes which the AI uses to base its responses on.

Your AI tools do not intentionally incorporate bias into their output; they will have to include assumptions in what is produced if the prompts that are provided do not have the level of detail to nullify these stereotypes being included.

Additional steps you can take

At the end of your prompt, ask what assumptions have been made by your AI
Provide clear instructions on what you would like to be generated
Add context and background information to support its understanding of the task set
Design your prompt to mitigate the stereotypes it may encounter: gender, accessibility, cultural, neurodiversity, etc.
Always read over what the AI has produced and adapt your prompt accordingly. Adapting your prompt may not simply be by adding information; sometimes it may require rewording and experimenting with.
You could also try a different AI tool; through Hwb, you have access to both Google Gemini and Microsoft Copilot, which both have enterprise-level data protection

Page updated

Report abuse