Lesson video

Hi, everyone, my name is Mrs. Jenkins.

Welcome to today's lesson.

Today's lesson is called, "When should AI be trusted?

" and it's taken from the unit, "The Digital Discovery Squad.

" Let's get started.

In today's lesson, our learning outcome is, "I can explain when AI is helpful and when it should not be trusted.

" In today's lesson, we have three keywords.

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

We have reliable, mistake, and trust.

Reliable is something that works correctly most of the time.

Mistake is something that is wrong or incorrect.

And trust is to believe that something is honest or correct.

So these three words, reliable, mistake, and trust, will appear all through the lesson so keep an eye out for them.

In today's lesson, "When should AI be trusted?

" we have two learning cycles.

The first is, "Recognize what AI can help with," and the second is, "Explain when AI should not be trusted.

" And to begin with, "Recognize what AI can help with.

" Artificial intelligence, AI, applications, or apps, compare patterns in data to give answers or suggestions.

AI applications are not suitable to help with all tasks, because some tasks depend on human understanding.

Andeep says, "I asked the AI app about an argument with my friend.

The answer it gave said my friend was being too sensitive, but I'm not sure.

" Jacob says, "AI apps do not understand the circumstances or human feelings.

" That's right, Jacob, well done.

It is up to humans to decide whether an AI app is reliable for a particular task.

Reliable means it's something that works correctly most of the time.

AI applications are more reliable when the goal is clearly defined, the task depends on patterns, and the AI application uses correct and complete data.

AI applications are less reliable when the task requires an understanding of feelings, involves personal values, and needs human judgment.

Let's have a little check.

Which example is most likely to be reliable when using an AI app?

A, predicting the next word in a sentence.

B, deciding how someone feels.

Or C, giving friendship advice for specific people.

What do you think?

Have a go.

How did you get on?

I asked, "which example is most likely to be reliable when using an AI app?

" A, predicting the next word in a sentence.

B, deciding how someone feels.

Or C, giving friendship advice for specific people.

The answer is A.

Predicting the next word in a sentence is the most likely to be reliable when using an AI app.

Great job, everyone.

Predicting words depends on patterns in language, So AI apps are usually quite reliable for this type of task.

So here we have toaster, cupboard, freezer.

And the sentence says, "Put the bread in the," so that optional words, "Put the bread in the toaster, cupboard, or freezer.

" All of these words are sensible suggestions about what a person may type in next.

Deciding how someone feels depends on human understanding and context.

So AI apps are not likely to be reliable for this type of task.

Jun says, "Fine," and he is smiling.

And in the second one, Jun says, "Fine," but he looks cross.

" Let's have a little check-in.

True or false?

AI apps are reliable for tasks that require understanding how someone feels.

What do you think?

Have a go.

How did you get on?

I asked you, "True or false?

AI apps are reliable for tasks that require understanding how someone feels.

" The answer is false.

The reason why, AI applications compare patterns in data.

Understanding feelings requires human judgment, so AI applications are not reliable for this type of task.

Great job, everyone.

Even when AI applications are helpful, they do not replace human judgment.

Humans decide when to rely on AI.

Okay, we are going to move on to task A now.

For each of the tasks, A to G, you are going to decide if an AI app could help you with the task.

Decide if the result would likely be reliable, and explain your decision.

We have, A, recognizing objects in pictures.

B, suggesting a route using traffic data.

C, choosing who to invite to a party.

D, deciding how someone feels from a photo.

E, recommending a film based on previous choices.

F, giving advice about a friendship problem.

And G, face-unlock recognizing a phone owner.

Have a go.

How did you get on?

I asked you to look at the tasks A to G and decide if an AI app could help you with the task, decide if the result would likely be reliable, and explain your decision.

The tasks were, recognizing objects in pictures, suggesting a route using traffic data, choosing who to invite to a party, deciding how someone feels from a photo, recommending a film based on previous choices, giving advice about a friendship problem, and face-unlock recognizing a phone owner.

So for the first one, decide if an AI application could help you with the task.

And the second, decide if the result would be reliable.

So A, recognizing objects in pictures.

AI could help and it is likely to be quite reliable data.

Suggesting a route using traffic data.

AI could help and it's likely to be a reliable result.

Choosing who to invite to a party.

AI couldn't help with that because they don't know your friends, they don't know who you'd want to invite, so it's unlikely to be a reliable result.

Deciding how someone feels from a photo.

Again, this isn't something AI could help with, and it will not give a reliable result.

E, recommending a film based on previous choices.

AI could help with this and it's likely to give a reliable result.

Giving advice about a friendship problem.

This isn't something that AI could help with and it's unlikely to give a reliable result.

And face-unlock recognizing a phone owner.

This is something AI could help with and it is likely to give you a reliable result.

Now explaining your decision.

Jacob says, "AI applications are reliable for clear pattern-based tasks.

Humans are needed when understanding and judgment are required.

Deciding how someone feels from a photo requires human understanding and context.

AI might not be reliable for this task.

" That's right, Jacob.

And Andeep says, "Suggesting a route using traffic data depends on patterns in the data.

AI can usually help and it's often reliable.

" That's right, Andeep, thank you.

Well done, everyone.

We are going to move on to learning cycle two now.

"Explain when AI should not be trusted.

" Reliable means something works correctly most of the time.

This does not mean it works correctly every time.

So here we have the weather forecast saying on Monday it will be sunny and they are stood in the sun and it is sunny.

Even reliable systems can make mistakes.

So now we have Monday, it is sunny, but actually it is raining.

A mistake is something that is wrong or incorrect.

AI applications follow instructions written by humans.

If the instructions are incomplete, the answer the application generates may contain mistakes.

Even when AI applications follow complete instructions, the system is comparing patterns in data.

If some data is missing or incorrect, the answer may contain mistakes and not be reliable.

So for example, if the system has only ever been given green apples and has also been given large red tomatoes, it may think a large red apple is a tomato.

Let's have a little check-in now.

Why might an AI application make a mistake?

A, it feels confused.

B, it uses incomplete or incorrect data.

C, it wants to trick people, or D, it does not fully understand the task.

What do you think?

Have a go.

How did you get on?

I asked, "Why might an AI application make a mistake?

A, it feels confused.

B, it uses incomplete or incorrect data.

C, it wants to trick people, or D, it does not fully understand the task.

" The answer is B.

An AI application might make a mistake because it uses incomplete or incorrect data.

Great job, everyone.

AI applications do not check if their outputs make sense because an app cannot understand meaning.

Aisha says, "The app returned, "Pigs can fly in the sky.

" That doesn't make sense.

" Okay, got another check-in here.

True or false?

If an output from an AI application sounds confident, it must be correct.

What do you think?

Have a go.

How did you get on?

I asked you, "True or false?

If an output from an AI application sounds confident, it must be correct.

" The answer is false.

If an output from an AI application sounds confident, it might not be correct.

The reason why, AI apps output mistakes and correct answers in the same confident way.

Well done, everyone.

Trust means to believe something is honest or correct.

Izzy says, "I might trust an AI app suggestion, but I still need to think about it.

" Trusting AI applications means deciding whether to rely on an answer or suggestion they generate.

Outputs from AI applications should not always be trusted, particularly when the decision affects safety, affects feelings, or requires human judgment.

Whether to trust the output from an AI application depends on the task you are doing and the possible results.

Predicting the next word in a sentence is a suitable task for an AI application.

Any mistakes would generally have low consequences.

So here we've got, "Put the bread in the," and the three optional words are toaster, cupboard, and shower.

So the AI is likely to suggest three words that make sense.

Asking an AI chatbot whether your friend is a good friend is not a suitable task.

This type of task requires human understanding and judgment.

Any mistakes could affect feelings and damage relationships, so you should not trust it.

Okay, we're going to move on to task B now.

I would like you to decide whether you should trust AI applications for each of the following examples.

So we have, predictive text suggests the next word in a message.

A map app suggests the fastest route to school.

A video platform recommends a new cartoon.

A voice assistant answers a maths question.

A chatbot gives advice about an argument with a friend.

And an AI tool guesses how someone is feeling from a photo.

Have a go.

How did you get on?

I'd asked you to decide whether you should trust AI applications for each of the following examples.

Predictive text suggests the next word in a message.

A map app suggests the fastest route to school.

A video platform recommends a new cartoon.

A voice assistant answers a maths question.

A chatbot gives advice about an argument with a friend.

And an AI tool guesses how someone is feeling from a photo.

So we should trust the predictive text suggesting the next word in a message, a map suggesting the fastest route to school, and a video platform recommending a new cartoon.

We should not trust a voice assistant answering a maths question, a chatbot giving advice about an argument with a friend, or an AI tool guessing how someone is feeling from a photo.

Well done.

For task two, I'd like you to choose two examples that you think should not be trusted.

Explain your reasoning.

Have a go.

How did you get on?

I asked you to choose two examples that you think should not be trusted and to explain your reasoning.

Sofia says, "I would not trust a chatbot giving advice about a friendship problem because friendships need understanding and human judgment.

AI applications only compare patterns so they do not understand feelings.

" Great job, Sofia.

Lucas says, "I would not fully trust a voice assistant answering a maths question because even though it is usually reliable, AI applications can still make mistakes.

If the answer is important, I wouldn't rely on it completely.

" Well done, Lucas, great job.

Your digital discovery squad member has earned their fourth badge.

You can now use judgment when deciding whether to trust AI applications.

Add your, "think first" badge and your character's rule to your logbook.

You have worked so hard in today's lesson.

Let's summarize our learning.

Reliable means something works correctly most of the time.

AI applications can make mistakes if the data or instructions are limited or incorrect.

AI applications should not be trusted for tasks that require human understanding.

Trusting AI applications depends on what the task is and the possible consequences.

Well done everyone.

You've worked really hard today.

I hope to learn with you again soon.