Body 101 · Special Lesson · Lesson 15 · Ages 10 to 13

Artificial Intelligence How AI Learns

The map app that finds a way around traffic jams, the next video an app picks for you, and the chatbot that answers questions in whole paragraphs: all of these are artificial intelligence, or AI. In this lesson we open the box and look inside to see how it works, why scientists copied the idea from the nerve cells in our brains, why it is brilliant at some things yet gets others hilariously wrong, and how we can use it wisely and safely.

A smartphone, smart speaker, laptop and wristwatch on a desk at night, linked by glowing golden lines like a network, with city lights through the window

1 · Meet AIAI is already all around us

What is AI, and where can we find it?

An ordinary computer program follows rules that people have written, line by line. A calculator, for example, always adds numbers using the same formula. AI is a program where nobody writes all the rules. Instead, it works out the rules for itself from a huge number of examples. Tap to see where AI is hiding in everyday things.

A white self-driving car on a city street at night, blue sensor beams fanning out around it to scan the road, other cars and trees — **Self-driving cars**They use cameras, radar and lasers to look in many directions at once, then AI works out in a split second which things are people, cars or road markings. It is a very hard job because there is no room for mistakes.

A round smart speaker with a blue light ring on a kitchen counter at night, next to a steaming cup of tea and a bowl of oranges — **Smart speakers that follow voice commands**AI turns our speech into text, guesses what we want, then answers back out loud, all in less than two seconds.

The AI timeline

Slide through year by year

A robotic arm on a table placing a dog photo onto a pile of dog photos, with a pile of cat photos on the other side and more cat and dog photos scattered around

2 · LearningLearning from examples, not rules

Teaching a computer to tell cats from dogs

Imagine writing rules to tell a computer how cats differ from dogs. It is really hard, because there are big cats, small dogs, pointy ears and floppy ears. So scientists changed their approach: they give the machine lots of labelled examples and let it find the patterns itself. Have a go at being the teacher. Tap the board to add examples. The background colour shows what the machine would guess if it met a new animal in that spot.

SmallBody sizeBig

Vertical axis: the higher up, the pointier and more upright the ears

Choose what to teach, then tap the board

0

Examples taught

–

Test set guessed correctly

This is the heart of machine learning The more examples there are, and the more accurate they are, the better the machine guesses. But if you teach it wrongly, it will faithfully learn the wrong thing too. Scientists call this garbage in, garbage out. Real AI learns from millions to billions of examples and looks at thousands of features at once, not just two axes like this.

A real neuron with glowing branches on the left gradually turning into a network of glowing round nodes joined by straight lines on the right

3 · NeuronsCopied from the brain

Real neurons and artificial neurons

In Lesson 8 we saw that a neuron receives signals from many neighbours. Some signals excite it and some calm it down. If they add up to something strong enough, it fires a signal onwards. Scientists copied this idea to make the artificial neuron, which is just a small equation: multiply each input by a weight, add them together, then check whether the total is over a threshold. Try setting the weights yourself.

Should I go out to play?

It’s sunny · weight1.0

Homework is done · weight1.0

It’s raining · weight-2.0

Threshold to beat 1.5

Let the machine find the weights

This is what we call training an AI. We give it eight examples along with the right answers and let the machine guess. Every time it guesses wrong, it nudges the weights a little in the right direction, and it keeps going round until it gets them all right. Press one round at a time to watch the changes.

Sunny	Homework done	Raining	Right answer	Machine’s guess

A pink model of a human brain on the left beside a large glowing golden computer chip on the right, on a dark desk — **The brain and the computer chip**The brain has about 86 billion neurons, and each one can connect with thousands of others. Large AI models have hundreds of billions of weights to adjust, but they still work very differently from a real brain.

A long data-centre aisle lined with tall server racks blinking blue and gold, cool mist on the floor, no people — **Data centres: the home of AI**Large AI models are trained in buildings where tens of thousands of computers work together for months. They use huge amounts of electricity and need water to keep cool. This is a cost people often don’t see.

Our brain compared with AI

	Human brain	Today’s AI
Energy use	About 20 watts, like a small light bulb	Training uses as much electricity as thousands of homes use in a whole year
How many examples does it need?	A child can recognise a giraffe after seeing one just a few times	It needs to see thousands to millions of pictures
What is it good at?	Understanding feelings, adapting to new situations, and having a body to experience the real world	Reading huge amounts of information very fast, without getting tired or bored
Does it have feelings?	Yes: happiness, pain, fear	There’s no evidence it feels anything, even if it types that it’s happy

Did you know?The 2024 Nobel Prize in Physics went to John Hopfield and Geoffrey Hinton, pioneers of artificial neural networks. An idea that began with copying nerve cells has become the foundation of almost every AI today.

An open book on a wooden desk with small glowing tiles streaming out of its pages into a laptop screen, beside a lamp and a globe

4 · ChatbotsThe next-word guessing machine

How does a chatbot come up with its answers?

Chatbots such as ChatGPT or Claude are trained to do one thing over and over: guess what the next word is likely to be. They have read more text than everyone on Earth could read in a lifetime, so they guess very well. Try playing with this tiny word-guesser that learned from just a handful of sentences.

Pick a first word

Next words the machine is considering

The numbers are chances. The machine works them out by counting, in the sentences it has read, how often each word came after this one. Real chatbots don’t just look at the one word before. They can look back over pages of the conversation, but the idea is the same.

Why do chatbots sometimes make things up?

Because they are trained to put together words that sound likely. They weren’t built to open a book and check the facts every time. If we ask about something they don’t know, they may string together words that sound convincing but aren’t true, like inventing the title of a book that doesn’t exist. Researchers call this hallucination, which means seeing things that aren’t there.

Thai is harder for AI

Thai is written without spaces between words, so the machine first has to guess where each word begins and ends. For example, ‘takrom’ could be read as ‘ta klom’ (round eyes) or ‘tak lom’ (to sit out in the wind). There is also much less Thai text on the internet than English, so many AI models are less good at Thai. That’s why Thai researchers are building their own Thai-language AI too.

A painting of mountains and a lake on an easel, the left half painted and the right half breaking into small coloured square pixels

5 · ImagesAI sees and draws pictures

To AI, a picture is a grid of numbers

A camera doesn’t see a cat. It only sees millions of tiny dots, and each dot is a number that says how bright it is and what colour. AI learns which arrangements of numbers mean an edge, which mean an ear and which mean a cat’s face. Try drawing on the grid and see what the machine sees.

Tap or drag to draw

What the computer sees

A single phone photo has about 12 million dots, and each dot has three numbers for red, green and blue. That adds up to tens of millions of numbers per picture.

How does AI draw pictures?

Many image-making AIs start from a picture of messy random dots, like a TV screen with no signal, then gradually remove the mess step by step over dozens of rounds, guided by our instructions, until it becomes a picture that matches the description. Every illustration in this lesson was made this way, and a doctor checked each one before it was used.

A hand holding a white teacup on a table; count carefully and you will find six fingers lined up, the kind of mistake AI pictures make — **Spot the AI mistake**AI can draw beautiful pictures, but it still makes mistakes that children spot straight away. Try counting the fingers in this picture. The things AI often gets wrong are the number of fingers, writing you can’t read, and shadows falling the wrong way.

Convincing fake pictures and voicesAI can make pictures, voices and videos of real people saying or doing things they never did. These are called deepfakes. If you see a clip of a celebrity asking you to invest, asking for money or sharing shocking news, be suspicious first, and ask an adult before you believe it or share it.

A hand holding a magnifying glass over a jigsaw of Earth, the lens revealing one piece whose shape does not fit its gap

6 · MistakesAI can sound sure, even when it’s wrong

What kinds of mistakes can AI make?

AI doesn’t lie on purpose. It has no intentions at all. But it can be wrong in many ways, and it often answers in the same confident tone whether it’s right or wrong. Tap each one to find out.

A tablet on a wooden desk showing Earth with network lines, a glowing translucent blue shield in front and a golden padlock beside it

7 · Using it wellSmart and safe

How to use AI so you really get better

AI is a tool, like a calculator or a library. Used the right way, it helps us learn faster. Used the wrong way, our brain gets no practice at all. Remember Lesson 12? The brain gets stronger by doing its own thinking.

6 safety rules

Use it to help you learn, not to do the work for you

An easy test: close the screen and try explaining it to someone at home. If you can’t, it means the AI understands it, but you don’t yet.

Try it: upgrade your prompt

A short request gets a vague answer. A request that says who you are, what you want and in what form gets an answer you can actually use. Pick a task, then tick the ingredients one at a time and watch the prompt change.

Two flashlights on a desk in front of a bookshelf: the left one throws a wide, faint beam across the whole shelf, the right one a narrow, bright beam straight onto a single book — **A clear prompt is like a focused flashlight**A vague prompt is like a flashlight with a spread-out beam: it lights the whole shelf but you cannot see any one book clearly. A prompt that says who you are and what you want is like a beam aimed exactly at the book you need.

A child’s hand moving a knight on a chessboard while a white robotic hand on the other side waits its turn, under a desk lamp — **AI as a sparring partner, not a stand-in**Chess players improve by moving the pieces and thinking for themselves, not by watching someone else play for them. Let the AI wait its turn, set you challenges and tell you where you went wrong, so your own brain gets real practice.

8 ideas for using AI more wisely

For children and adults alike. Each idea comes with an example prompt you can type straight away.

A child’s hands writing in a notebook with a pencil beside a laptop showing Earth, with books, a globe, a telescope and a solar system model — **Write it yourself, think for yourself**Writing a summary by hand after chatting with AI helps the knowledge really move into your brain. It matches the self-testing idea from Lesson 12.

Game: which one is AI?

Read the name of each thing and choose whether it learns from examples, follows rules people wrote, or doesn’t use a computer at all. You have 45 seconds.

045 seconds

Press to start the game

Two hands holding a tablet in front of a city skyline at sunrise, golden network lines rising from it into a world map across the sky

8 · The futureAI, doctors and our jobs

AI in hospitals, and jobs of the future

Many hospitals have started using AI to help doctors, for example to look for bleeding in brain scans, to screen photos of the retina in people with diabetes, and to warn when blood test results look unusual. But the person who decides on treatment, explains it so the patient understands, and takes responsibility for the result is always the doctor.

A hand in a white coat pointing at a screen showing an axial brain scan, with a glowing orange outline around one spot flagged by AI — **AI helps doctors read scans**AI marks suspicious spots for the doctor to look at first, so patients who are very ill get care sooner. But AI can raise false alarms or miss small spots, so the doctor must check every image themselves.

An orange industrial robot arm assembling metal parts on a factory conveyor, golden sparks flying, no people — **Robots in factories**Jobs that repeat the same thing every day are starting to be done by machines and AI. At the same time, new jobs are appearing that never existed before, such as people who train AI, people who check that AI is safe, and people who design AI to be easy to use and fair.

Skills AI still finds hard to replace

Asking good questions, checking whether answers are right, understanding the feelings of the person in front of you, working as a team, and taking responsibility for what happens. You can practise these skills starting today, by reading, talking with real people, and making things with your own hands.

What do you need to be good at to work in AI?

Maths helps you understand how machines learn, and programming lets you build things yourself. But just as important is knowing about the area where AI will be used. Good medical AI needs doctors to help design it, and farming AI needs people who understand rice and soil.

RememberAI is getting better very quickly, but it isn’t here to replace our thinking. It is a tool that extends what we already have. People who can think for themselves, know how to check things, and use AI responsibly are the ones who will benefit from it most.

A human hand and a white robotic hand opening a glowing book together on a wooden desk, with a cell, a brain, Earth and the Sun rising from its pages under a lamp

9 · Behind the scenesThe advanced AI used to build this page

Which AI built this lesson?

Every lesson in Body 101 was made by one doctor working with several advanced AI systems. Each one has a different job, like a team in a hospital. The doctor leads the team, sets the content and checks every piece before it goes online. Tap each step to see which AI it used, what the AI did and what a person checked.

How it was built

17

lessons in two languages

160+

AI illustrations, all checked

344

questions in the arena

0

pieces of learners’ personal data collected

Rules the doctor set for every AI

Why a person always stays in the loop

While this series was being made, the AI really did make mistakes. The first heart picture had blood vessels joined to the wrong chambers. The cat-and-dog board in part 2 once guessed dog everywhere when there were only a few examples. The number boxes in several lessons were displayed badly without anyone noticing. Every one of these was caught because someone opened the real thing and pressed every button.

Researchers call this design human-in-the-loop: let AI do the fast, heavy work, but keep a person as the checker and decision-maker wherever mistakes are not allowed.

Where does the read-aloud voice come from? The read-aloud button uses the synthetic voice that comes built into the learner’s own phone or computer. The text is not sent anywhere, and it does not imitate the voice of any real person.

Five robotic arms along a conveyor building glowing books step by step, and at the end a human hand holding a magnifying glass to inspect the finished book — **An AI production line with a person checking at the end**Each AI handles a different step, linked together into a very fast production line. But before any book reaches a child, it always passes under the magnifying glass of a person at the end of the line. That is what human-in-the-loop means.

In shortAdvanced AI lets one doctor do work that used to need a whole team of writers, programmers, translators, illustrators and testers. But responsibility for what children learn always stays with people. That is exactly the way this lesson recommends everyone use AI.

A tall wooden bookshelf of glowing books rising like steps, green on the lower shelves, purple in the middle and red at the top, with a wooden ladder leaning against it

10 · GlossaryFrom beginner to upper-intermediate

32 AI terms worth knowing

When you read news or articles about AI, you’ll often come across these words. They are split into three levels. Level 1 words are used every day, Level 2 words explain how AI learns, and Level 3 words are the ones AI developers use with each other. Start with Level 1, then work your way up.

Self-test mode uses the same idea as Lesson 12. Read the word, try to explain it in your head first, then tap the card to check. It works much better than just rereading.

Review12 questions

Test yourself

Correct: 0 of 12

Fancy a game on this lesson?

Every question from this lesson is in the Body 101 arena. Race the clock, duel someone at home, play as a team, beat the bosses, or practise with no timer at all.

Play this topic The whole arena