Lesson video

Hello, everyone.

Welcome to today's lesson.

My name is Mrs. Jenkins.

Today's lesson is called "Fairness and checking AI applications.

" It's taken from the unit, "The digital discovery squad.

" Let's get started.

Our learning outcome in today's lesson is "I can explain why outputs from AI applications should be verified using other sources.

" In our lesson today, we have three keywords.

These words will appear throughout the lesson, so it's really important we know what they mean before we get started.

We have fair, bias, and verify.

Fair means treating people equally.

Bias is something that unfairly supports one side or group.

And verify is to make sure something is true or correct.

So those three words are fair, bias, and verify.

Keep an eye out for them as we move through today's lesson.

In our lesson today, "Fairness and checking AI applications," we have two learning cycles.

First is "consider whether outputs from AI are fair," and the second is "identify ways to verify AI answers.

" Let's get started with "consider whether outputs from AI are fair.

" Fair means treating people equally and giving them what they need to have the same chance of success.

Sometimes AI apps give different results to different people.

Is that fair?

Fair does not always mean the same.

Fair means people get what they need to have an equal chance.

Jacob says, "If everyone got the same size shoes, they would have the same, but it would not be fair because some people's shoes would not fit.

" We have an image here as well, which is a great example.

If we gave everyone the same size box to see over a fence, they still would have the problem that they couldn't see.

Some people would not be able to see, and that would not be fair.

But rather than giving children, giving people what they need.

So the smallest person needed two boxes to see over, the middle-height person needed one, and the tallest person didn't need a box at all.

Now everyone has an equal chance.

AI applications use data to make decisions.

If the data is not balanced, the results may not be fair.

The quality and balance of data will affect the results.

Bias means something that unfairly supports one side or group.

Bias can happen when an AI application's training data is incomplete or reflects patterns from the past.

AI applications identify patterns in data, so they may repeat this bias in their results.

Jacob and Izzy search for scientist using an AI application.

This is the output from the AI application.

We have an image with three different men all in white coats being scientists.

Why might this happen?

The AI application has identified patterns from images in its training data.

If most images in the training data show men as scientists, the AI application may repeat the pattern.

This does not mean that the AI application is intentionally unfair, but the result can still be biased.

Unfair outcomes can cause problems.

People may be left out, misunderstood, or treated differently.

Bias can affect opportunities and representation.

Let's have a little check now.

Which statement best describes bias?

a, AI applications are always correct, b, AI applications can unfairly work better for one group than another, c, AI application outputs are random, or d, AI applications have feelings.

What do you think?

Have a go.

How did you get on?

I asked which statement best describes bias?

a, AI applications are always correct, b, AI applications can unfairly work better for one group than another, c, AI application outputs are random, or d, AI applications have feelings.

The answer is b, AI applications can unfairly work better for one group than another.

Great job, everyone.

If we notice unfair results from an AI application, the problem may come from the data used to train it.

Sometimes data reflects patterns from the past.

If the data is incomplete or unbalanced, the AI application may repeat those patterns.

This means bias can appear even when nobody intended it.

Even when an AI application gives the same answer to everyone, it may still be unfair.

If the answer ignores important differences between people, the outcome may still contain bias.

Let's have another little check.

True or false?

If AI gives the same answer to everyone, it is always fair.

What do you think?

Have a go.

How did you get on?

I asked you true or false?

If AI gives the same answer to everyone, it is always fair.

The answer is false.

The reason why, an answer can be the same but still ignore important differences or contain bias.

Great job, everyone.

Giving everyone the same answer can still be unfair if the answer does not work well for everyone.

Fair systems must work equally well for different groups.

We're going to move on to Task A now.

I would like you to look at two AI search results for the same question, "What jobs can people do?

" We've got result A and result B.

Which result seems fairer?

And explain how bias may appear in the result.

Have a go.

How did you get on?

I asked you to look at two AI search results for the same question, "What jobs can people do?

" Result A and result B, we had two different pictures.

Which results seemed fairer?

Explain how bias may appear in the results.

A fair result should show a wide range of options.

If the AI application shows mostly one type of job, the training data may not represent everyone equally.

Bias can appear when the data used to train AI is limited.

Let's move on to learning cycle 2, "Identify ways to verify AI answers.

" Verify means to check that something is true or correct.

When an AI application gives an output, we should not always accept it immediately.

AI applications sometimes give answers that sound confident, even when the information is incomplete or incorrect.

So we should verify important answers.

Verify means to check that something is true or correct.

The output from AI applications should not be automatically accepted.

It is important to check it using another source.

A source is where information comes from.

When we verify information, we check the answer using another reliable source.

Examples of sources include books or encyclopedias, trusted educational websites, teachers or knowledgeable adults.

Checking more than one source helps us decide if information is correct.

If an AI application gives you information, you can verify it by checking another trusted website, asking a teacher, comparing with a textbook, or looking for multiple sources.

We have a simple verification routine.

So it's pause, check, and decide.

Pause means don't share it just yet.

Check, use another source to make sure you find the answer.

And decide, is it reliable enough?

These steps help prevent mistakes spreading online.

Sometimes AI applications produce incorrect answers.

The application might miss important details, use patterns in data to extend an answer, use information that is incomplete.

Let's have a little check.

Which action is an example of verifying?

a, sharing the first answer straightaway, b, checking a book or trusted site to see if the answer matches, c, asking AI again and again until you like the answer, or d, choosing the funniest answer.

What do you think?

Have a go.

How did you get on?

I asked which action is an example of verifying?

a, sharing the first answer straight away, b, checking a book or trusted site to see if the answer matches, c, asking AI again and again until you like the answer, or d, choosing the funniest answer.

The answer is b.

An example of verifying is checking a book or trusted site to see if the answer matches.

Great job, everyone.

Verifying matters most when it affects safety, feelings, and decisions.

If an AI application outputs "A rabbit is a reptile," what should you do?

Izzy says, "I would verify this by checking another trusted source to see if the information is correct.

" Rabbits are mammals, so the AI answer is incorrect.

Well done, Izzy.

That's a great way of checking.

If an AI application outputs "This plant is safe to eat," what should you do?

Jacob says, "I would not trust this answer immediately.

I would ask a teacher or check a trusted guide.

" That's a great response, Jacob.

Information about safety should always be verified carefully.

Why should the outputs from AI applications be verified?

a, AI apps always lie, b, AI apps may use incomplete or incorrect data, c, AI apps have feelings, or d, AI apps cannot read.

What do you think?

Have a go.

Okay, how did we get on?

Why should the outputs from AI applications be verified?

a, AI apps always lie, b, AI apps may use incomplete or incorrect data, c, AI apps have feelings, or d, AI apps cannot read.

The outputs from AI applications should be verified because AI apps may use incomplete or incorrect data.

The answer is b.

Great job, everyone.

Verifying information helps us make better decisions.

It also prevents incorrect information from being shared with others.

We're going to move on to Task B now.

I would like you to read the three AI answers.

For each one, explain how you would verify it and name a source you would use.

So the first one is the Great Wall of China can be seen from space.

The second, sharks are mammals.

And the third, you should drink eight liters of water every day.

Have a go.

How did you get on?

I asked you to read the three AI answers.

For each one, explain how you would verify it and name a source you would use.

So we had the Great Wall of China can be seen from space, sharks are mammals, and you should drink eight liters of water every day.

The first one, the Great Wall of China can be seen from space.

How would you verify it?

You could check a trusted science or history website.

So the source, a space agency website or encyclopedia.

But the answer: The Great Wall is not visible from space without special equipment.

Well done.

Number two, sharks are mammals.

How could you verify it?

You could check a reliable science source about animal groups.

So the sources you could use, a science book, encyclopedia, or trusted educational website.

The answer: Sharks are fish, not mammals.

AI apps sometimes produce incorrect answers.

Well done, everyone.

Number three, you should drink eight liters of water every day.

How could you verify it?

You could check a health website or ask a trusted adult.

So the source would be a health website, a doctor, a teacher.

The answer: Eight liters is far more than most people need to drink in a day.

Health advice can vary depending on age and activity, so it is important to check health information carefully before believing or sharing it.

Great job, everyone.

Well done.

Your digital discovery squad member has earned their fifth badge.

You can now explain why AI answers should be verified.

Add your Expert verifier badge and your character's rule to your logbook.

Well done.

Well done, everyone.

You have worked extremely hard today in today's lesson.

Let's summarize what we have learned.

AI applications can produce different results for different groups.

Bias can occur when training data is unbalanced.

Even identical answers can sometimes be unfair.

The outputs from AI applications should be verified using other trusted sources.

And checking information helps ensure fairness and accuracy.

Well done, everyone.

You have worked really hard today.

I hope to learn with you again soon.