What Is Artificial Intelligence? A Machine That Guesses From Patterns
Artificial intelligence does not know anything. It has seen millions of examples, found the patterns in them, and makes a very good guess.
The whole idea, in one diagram
You show it a huge number of examples, like millions of pictures labelled cat.
It finds the patterns those examples share. Nobody writes the rules by hand.
Then it guesses about something new. A guess can be excellent, and it can be confidently wrong.
What problem did it solve?
Some jobs cannot be written as rules. Try writing exact instructions for recognising a cat, or for telling a happy voice from a sad one. People tried for decades and the rules always broke on the next example.
Instead of writing rules, show the machine thousands or millions of labelled examples and let it work out the patterns itself. That is called machine learning.
Computers can now read handwriting, understand speech, spot illness in a scan and hold a conversation. They do it by pattern and probability, never by understanding, and that difference matters enormously.
How it started
- 1950
Can a machine think?
👤 Alan Turing, England
He suggested a test: if you swap messages with something hidden and cannot tell whether it is a person or a machine, does the difference matter?
Why it mattered: He moved the question from what a machine IS to what a machine can DO. That is still the question we are arguing about.
- 1956
The name is invented
👤 John McCarthy and colleagues, at Dartmouth College, USA
A summer workshop was arranged to see whether machines could be made to use language and solve problems. They needed a name for the subject, and chose artificial intelligence.
Why it mattered: They expected serious progress in a summer. It took about sixty years, which is a useful lesson about confident predictions.
- 1966
The first chatbot, and a warning
👤 Joseph Weizenbaum, USA
ELIZA answered by turning your sentences into questions, with no understanding at all. People still told it their private worries, and some refused to believe it was a program.
Why it mattered: Weizenbaum was so troubled that he spent years warning about it. People trust something that sounds human, even when they have been told exactly how simple it is.
- 1997
A computer wins at chess
👤 IBM's Deep Blue against world champion Garry Kasparov
The machine beat the best human chess player in the world, mostly by checking an enormous number of possible moves very quickly.
Why it mattered: It was a real milestone and it was not thinking. It was searching. People had to learn the difference.
- 2009 to 2012
Data turns out to be the missing piece
👤 Fei-Fei Li, then Krizhevsky, Sutskever and Hinton
Fei-Fei Li built ImageNet, a collection of millions of labelled photographs. In 2012 a program trained on it recognised images far better than anything before.
Why it mattered: The big idea was not a cleverer program. It was enough examples. This is the moment modern AI actually starts.
- 2016
A machine plays a move nobody expected
👤 AlphaGo against Lee Sedol
In the ancient game of Go, a program beat one of the world's best players, and played one move so strange that commentators assumed it was a mistake. It won the game.
Why it mattered: Chess was solved by searching harder. Go needed something more like intuition learned from experience, which felt genuinely different.
- 2022 onwards
AI that writes
👤 Many companies
Programs trained on enormous amounts of text learned to predict what word is likely to come next, well enough to write essays, answer questions and explain ideas.
Why it mattered: Useful and unreliable in the same breath. Predicting a likely next word is not the same as knowing whether it is true, which is why these systems invent facts.
- 2024
The Nobel Prize
👤 John Hopfield and Geoffrey Hinton
They shared the 2024 Nobel Prize in Physics for the foundational work on the networks that learn from examples.
Why it mattered: Ideas dismissed as a dead end for decades turned out to underpin almost all modern AI. Being unfashionable is not the same as being wrong.
Who made it happen
Alan Turing
1912 to 1954 · England
Asked in 1950 whether a machine could think, and suggested judging it by what it can do rather than what it is.
John McCarthy
1927 to 2011 · USA
Gave the field its name in 1956 and helped organise the workshop where it began.
Fei-Fei Li
born 1976 · China and USA
Built ImageNet, millions of labelled photographs. It turned out that data, not cleverness, was the missing piece.
Geoffrey Hinton
born 1947 · England and Canada
Spent decades on networks that learn from examples while most people thought it hopeless. Shared the 2024 Nobel Prize in Physics.
Joseph Weizenbaum
1923 to 2008 · Germany and USA
Built the first chatbot in 1966, then spent years warning people how easily they trust something that merely sounds human.
Timnit Gebru
born 1983 · Ethiopia and USA
Showed that face recognition worked far worse on darker skin and on women, because of who was in the training examples.
How a computer learns to recognise something
- 1
Collect a huge number of examples
To learn what a cat looks like it might see millions of photographs, each one already labelled by a person as cat or not cat. The labelling is slow human work.
- 2
Let it find the patterns
Nobody tells it about whiskers or ears. It adjusts itself over and over until its answers on the examples are mostly right, and the patterns it finds are its own.
- 3
Test it on things it has never seen
The real test is a new picture. If it only did well on the pictures it studied, it memorised instead of learning, which is a mistake humans make in exams too.
- 4
It answers with a probability
It does not say cat. It says something closer to 94% cat. That number is a measure of pattern match, not of truth, and it can be high and wrong at the same time.
- 5
It inherits whatever was in the examples
If the examples were mostly one kind of person, one country or one point of view, the answers lean that way. The machine cannot notice this. Only people can.
Watch it happen, step by step
Tap through the animation. You can switch it to Urdu with the button.
How does a computer learn to recognize a cat?
AI has no idea what it is talking about. It predicts what usually comes next, based on what it has seen, which is why it can be brilliant and completely wrong in the same sentence.
Ask an AI helper, with a grown-up, about something you already know extremely well: your town, your favourite sport or a book you have read many times. Count how many things it gets right, and find the one thing it says confidently that is simply not true. Almost every time, there is one.
Key words on this page
- What is AI?AI کیا ہے؟
- AI means teaching a computer to be smart by showing it lots and lots of examples. Just like you learn to count by practising, the computer learns by looking at millions of examples. Then it can help us answer questions, draw pictures, and much more!
- Did you know? The word 'Artificial' means 'made by people, not nature', so AI is smartness made by people!
- AI is all around usAI ہمارے چاروں طرف ہے
- AI is hiding everywhere! When you ask Siri or Google a question, that is AI. When YouTube shows you a video you might like, that is AI too. Even the characters in your favourite video games use AI to move around!
- Did you know? There are more AI assistants in the world than there are people, billions of them!
- Robots and AIروبوٹ اور AI
- No! A robot is a machine that can move and do things with its body. But most simple robots just follow exact instructions, they cannot think or learn. AI is the clever, thinking part that some robots have, but not all of them do.
- Did you know? A robot vacuum cleaner uses AI to figure out where the walls are so it doesn't keep bumping into them!
- AI learns by practisingAI مشق کر کے سیکھتی ہے
- AI gets smarter the same way you do, by practising lots and lots! If you show an AI a million pictures of cats, it learns what a cat looks like. Every time it gets an answer wrong, it adjusts and tries to do better next time.
- Did you know? Some AIs have practised on more examples than a person could read in a thousand lifetimes!
- AI can make mistakesAI غلطیاں کر سکتی ہے
- Yes! AI is a very helpful tool, but it is not perfect. It can sometimes give a wrong answer and sound very sure about it! That is why it is always good to check important facts with a grown-up, a book, or your teacher.
- Did you know? Even the smartest AI in the world sometimes thinks a dog is a muffin because they can look the same in a photo!
- How a machine learnsمشین کیسے سیکھتی ہے
- Step one: give the AI lots of examples called data, like thousands of photos of dogs. Step two: the AI finds a pattern, the things that are the same in all the pictures. Step three: when it sees a new photo, it makes a smart guess using that pattern!
- Did you know? The more examples you give an AI, the better its guesses become, just like practising a sport!
- What is a chatbot?چیٹ بوٹ کیا ہے؟
- A chatbot is a computer programme you can have a conversation with by typing. You ask it something and it types back an answer. The very first chatbot was called ELIZA, made way back in 1966!
- Did you know? Modern chatbots are millions of times smarter than ELIZA, which could only follow very simple rules.
- Large Language Model (LLM)بڑا لسانی ماڈل
- A Large Language Model, or LLM, is an AI that has read a HUGE amount of writing, books, websites, and articles. It works a bit like super-smart autocomplete: it reads your question and guesses the best next word, then the next, and the next, until it has written a full answer!
- Did you know? Some LLMs have been trained on more words than all the books in the biggest library in the world, combined!
- ChatGPT and Claudeچیٹ جی پی ٹی اور کلاڈ
- ChatGPT is a friendly AI helper made by a company called OpenAI. Claude is another friendly AI helper, made by a company called Anthropic. Both of them are Large Language Models you can chat with, they can answer questions, tell stories, help with homework, and much more!
- Did you know? When ChatGPT first launched in 2022, it got one million users in just five days, faster than any app in history!
- Always double-check AIAI کو ہمیشہ جانچو
- AI can sound very confident even when it is wrong! It is guessing the best next word, but sometimes the guess is not accurate. Always check important facts with a teacher, a parent, or a trustworthy book, especially for school work or health questions.
- Did you know? Scientists have a word for when AI confidently says something wrong, they call it a 'hallucination'!
- How AI makes picturesAI تصویریں کیسے بناتی ہے
- It uses a clever trick called diffusion. First the AI starts with a screen that looks like random TV static, just random coloured dots everywhere. Then, step by step, it slowly cleans up those dots until they form a real picture that matches your words. It is like a sculpture slowly appearing from a block of marble!
- Did you know? Some AI image tools can make a beautiful, detailed picture in just a few seconds from a sentence you write!
- How AI makes videosAI ویڈیوز کیسے بناتی ہے
- A video is really just lots of pictures called frames that play one after another very fast, usually 24 pictures every second! AI makes each frame using the same diffusion trick, but it also makes sure each frame flows smoothly into the next one so that things look like they are really moving.
- Did you know? The very first AI-generated video was only a few seconds long, but now AI can make videos several minutes long!
- What is a neural network?نیورل نیٹ ورک کیا ہے؟
- Your brain has billions of tiny cells called neurons that are all connected to each other. When you learn something, those connections get stronger. A neural network is a computer system built to work in a similar way, millions of tiny connections all working together to recognise patterns and make smart decisions.
- Did you know? The neural network that helps AI recognise your face in a photo has more connections than a bee's brain!
- What is training?تربیت کیا ہے؟
- Training means showing the AI millions of examples so it can learn. You give it data, it makes guesses, it checks if those guesses are right or wrong, and then it adjusts its connections to do better. You repeat this process millions and millions of times until the AI is smart enough. Training a big AI can take weeks and costs a lot of electricity!
- Did you know? Training a large AI model uses as much electricity as flying a jet plane across the world hundreds of times!
- Using AI kindly and safelyAI کا مہربانی اور حفاظت سے استعمال
- There are three golden rules! First: never share personal secrets with AI, like your address, passwords, or family details. Second: be respectful, using AI to be mean to others is never okay. Third: always check important facts with a grown-up, a teacher, or a trusted book, especially for health or school questions.
- Did you know? Many countries around the world are now writing special rules and laws to make sure AI is used safely and fairly for everyone!
Quiz: how much did you remember?
Quiz time!
10 questions, picked at random from 30. No sign-in and no score is kept. Play as many times as you like, the questions change every time.
Questions about artificial intelligence
What is artificial intelligence in simple words?
AI is a computer program that has been shown a huge number of examples and has worked out the patterns in them, so it can make a good guess about something new. It is guessing from patterns, not knowing.
Is AI actually intelligent?
Not in the way a person is. It has no understanding, no experience of the world and no idea what its own words mean. It is extremely good at spotting patterns, which can look a lot like intelligence from the outside.
Why does AI sometimes say things that are completely wrong?
Because it predicts what is likely to come next rather than checking what is true. A confident and false answer costs it nothing, which is why you must check anything that matters.
What is machine learning?
It is the method behind most modern AI. Instead of a person writing the rules, the computer is shown many labelled examples and works the rules out for itself.
Can AI take my job one day?
It will change many jobs rather than simply removing them, and it will create new ones. The safest skills are the ones AI is worst at: judging whether an answer is right, understanding people, and deciding what is worth doing at all.
Why is AI sometimes unfair?
Because it copies its examples. If the pictures or the writing it learned from mostly showed one kind of person, it works less well for everyone else. Researchers have shown this happening in real face recognition systems.
Should I use AI for my homework?
Ask your teacher, and be honest about it. Using it to explain something you do not understand can be genuinely helpful. Using it to write the answer means you skipped the part where your brain does the growing, and it may well be wrong anyway.
Who invented AI?
No single person. Alan Turing asked the question in 1950, John McCarthy named the field in 1956, and the modern version arrived when Fei-Fei Li's huge image collection met the network ideas Geoffrey Hinton had worked on for decades.
Want lessons that are marked and saved?
The free courses run from playgroup to Grade 5, in order, with three difficulty levels.
See the free courses →