AI (artificial intelligence) is software that learned from a huge amount of human writing, pictures and sound, so it can now write, draw, talk and even carry out tasks when you ask it in plain words. You do not need any technical knowledge to use it well.
This page explains the whole thing with everyday comparisons: no maths, no jargon left unexplained.
- What it is: think of a very well-read assistant that has read more than any person could in a lifetime, and learned the patterns of how people write and talk.
- How it works: it was not programmed with answers. It learned by example, and it writes its reply one word at a time, a bit like the word suggestions on your phone, only far bigger.
- Its big weakness: it can say something false with total confidence. Check anything that matters.
- How to use it well: talk to it like a smart new colleague on their first day: say what you want, give the background, and say what the answer should look like.
- Staying safe: treat what you type like a postcard, not a sealed letter. Do not share passwords or private details.
What is AI, in one minute?
AI stands for artificial intelligence. Today, when people say “AI”, they usually mean tools like ChatGPT, Claude or Gemini: you type a question or a request in normal words, and it answers in normal words.
Prefer paper? Print the one-page cheat sheet.
The comparison that helps most: imagine a student who has spent years in the biggest library in the world, reading everything: books, websites, manuals, conversations. That student never memorised the books word for word. Instead they got very good at the patterns: how a recipe is laid out, how a polite email sounds, how a legal letter is worded, how a joke usually lands. Ask them for any of those and they can produce one on the spot.
That is roughly what a modern AI is. It is not a person, it has no feelings, and it is not looking anything up in a hidden encyclopaedia unless it has been given a search tool. It is producing the kind of answer that fits your request, based on patterns it learned.
AI is a pattern expert, not a fact database. That one idea explains most of what it is good and bad at.
AI you already use every day
The plain point: chatbots are the new face of AI, but you have been using quieter kinds for years.
- Your spam filter. It learned what junk mail looks like from emails people had already marked as spam. Google uses spam as its teaching example of this.1
- Your maps app. Google Maps mixes past traffic patterns with live traffic and uses machine learning to predict how long your trip will take.2
- “Because you watched…” Streaming apps guess what you will enjoy from what you and people with similar taste watched (a recommendation algorithm). Netflix says its algorithms estimate the likelihood you will enjoy each title.3
- Voice assistants. “Hey Siri” is spotted by a neural network running on your phone itself, and IBM lists Siri and Alexa as everyday examples of AI built for specific jobs.45
If an app sorts, predicts or recommends something for you, there is a good chance a small, single-job AI is behind it.
Narrow AI vs general AI
Every AI you can use today is narrow AI: built for one job or a set of jobs, even when the job is as broad as chatting. General AI (AGI) would match or beat people across almost any task; it is a goal and a debate, not a product, and what AGI means explains why the word is contested.5
How does it learn?
It learns by example, not by rules. Nobody sat down and typed in the answers. Anthropic, which makes Claude, says so plainly: “Language models like Claude aren’t programmed directly by humans—instead, they’re trained on large amounts of data.”6
The comparison: think of how a child learns to speak. No one hands a toddler a grammar book. They hear thousands of conversations and slowly pick up what words mean and how sentences fit together. AI learning, called training, works in a similar spirit, but with an enormous amount of text and far more computer power.
Then it writes one word at a time. When you ask a question, the AI does not pull up a finished answer. It picks the most fitting next word, then the next, then the next, until the reply is done.6
The comparison: it is like the word suggestions above your phone keyboard, which guess your next word. A modern AI is that idea grown enormously, so good at choosing the next word that whole essays, poems and computer programs come out.
Training is the years in the library. Using it is asking that well-read student a question. The answer is written fresh each time, which is why asking the same thing twice can give two slightly different replies.
AI vs machine learning vs deep learning vs generative AI
The plain point: these four words fit inside each other like nesting dolls. AI is the biggest; generative AI is the newest and smallest doll inside.
| Word | What it means | Everyday example |
|---|---|---|
| AI | Means: any computer system doing thinking-type tasks. | Like: all of the examples below, plus older rule-based systems. |
| Machine learning | Means: AI that learns from examples instead of following rules a person wrote. | Like: the spam filter that learned from marked emails. |
| Deep learning | Means: machine learning that uses large, many-layered neural networks. | Like: the network that hears “Hey Siri”. |
| Generative AI | Means: deep learning that creates new text, pictures, video or sound when you ask. | Like: a chatbot writing an email, or an app drawing a picture. |
IBM describes the same nesting: machine learning sits directly under AI, deep learning is a kind of machine learning, and generative AI is deep learning that creates original content.5 The chatbots most people mean by “AI” today sit in the last row: how ChatGPT works explains the engine inside them.
The main kinds of AI you will meet
“AI” is a family, not one thing. Here are the members you are most likely to use, each with a job comparison.
| Kind | Think of it as | What you use it for |
|---|---|---|
| ChatbotsChatGPT, Claude, Gemini | Think of it as: the writer and explainer | Use it for: Questions, emails, summaries, ideas, homework help, planning. |
| Image makersMidjourney, Nano Banana, FLUX | Think of it as: the illustrator | Use it for: Pictures from a description: posters, product shots, art. |
| Video makersVeo, Kling, Seedance | Think of it as: the film crew | Use it for: Short clips from a description or a photo. |
| Voice and musicElevenLabs, Suno | Think of it as: the voice actor and the band | Use it for: Read text aloud in a natural voice, or make a song. |
| AgentsMeta Muse, Gemini Spark, Grok Bot | Think of it as: the assistant with the keys | Use it for: Not just answering but doing: booking, buying, filling forms, using apps for you. |
The last row is the big change of 2026. A chatbot gives you advice, like a friend on the phone. An agent can act, like an assistant you have handed your house keys to. That is useful, and it is also why the safety section below matters. Understanding AI model types goes deeper on each kind.
Advice you can ignore. Actions have already happened. Be more careful with agents than with chatbots.
Why does AI sometimes get things wrong?
The short answer: it was trained to always give an answer, and a confident guess often scores better than saying “I don’t know”. People call a confident wrong answer a hallucination.
The comparison: picture a student taking a multiple-choice test where a blank answer scores zero. Guessing can only help their score, so they learn to always guess. Researchers found AI ends up in the same position: the way it is trained and tested rewards a guess over admitting doubt.7
So the AI is not lying on purpose. It is doing what it was shaped to do: produce a fitting answer. When it does not actually know, the fitting answer can still sound perfectly sure of itself. Made-up book titles, wrong dates and invented quotes are the classic examples.
A second habit to know about: AI tends to agree with you. Push back on a correct answer and it may fold. Think of an eager new employee who does not want to contradict the boss. How AI influences you back shows what studies measured.
Use AI like a clever friend who is sometimes confidently wrong: great for drafts, ideas and explanations, but check names, numbers, dates, quotes and anything medical, legal or financial before you rely on it. Why does AI hallucinate? explains the full reason.
Ten words you will hear, explained
You will meet these terms everywhere. Each comes with a one-line comparison.
| Word | In plain words | Everyday comparison |
|---|---|---|
| Prompt | Means: What you type to ask the AI for something. | Like: the order you give a waiter: the clearer it is, the better the meal. |
| Model | Means: The AI “brain” itself, such as GPT-6 or Claude Opus 5.5. | Like: the engine; the app you type into is the car around it. |
| Training | Means: The learning phase, before you ever use it. | Like: the years in the library, before the student takes questions. |
| Token | Means: A small chunk of text, a word or part of a word, that AI reads and writes in. | Like: Lego bricks: every sentence is built from them, and usage is often counted in them. |
| Context window | Means: How much text the AI can keep in mind at once in one conversation. | Like: the size of its desk: once the desk is full, the oldest papers fall off. |
| Hallucination | Means: A confident answer that is wrong or made up. | Like: the test-taker who guesses rather than leave a blank. |
| Parameters | Means: The internal settings the AI adjusted while it learned. | Like: a huge number of tiny dials, tuned during training, that shape every answer. |
| Open weights | Means: The model’s files are published, so anyone can download and run it on their own computer. | Like: a restaurant publishing its recipe, so you can cook the dish at home. |
| Agent | Means: An AI that can take actions, not only answer. | Like: an assistant with your keys and your card, not just your phone number. |
| Fine-tuning | Means: Extra training to make a general model good at one job. | Like: a general doctor doing extra years to become a specialist. |
Want more? The AI glossary explains more than a hundred terms, and AI terms people get wrong clears up the common mix-ups.
How to get good answers
The comparison that works best: treat the AI like a very smart new colleague on their first day. They are capable, but they know nothing about you, your project or what “good” looks like to you. The more you brief them, the better the work.
- Say what you want. Not “write about dogs” but “write a short, friendly post for my dog-walking business”.
- Give the background. Who is it for? What do they already know? What should it avoid?
- Say what the answer should look like. A list, a table, three options, a short email, plain words.
- Improve instead of starting over. If the first answer is mostly right, say what to change: “Keep everything, but make it shorter and less formal.”
- Ask it to check itself. “What in your answer might be wrong?” often catches mistakes.
How to write AI prompts and what makes a good prompt take this further, step by step.
Vague request, vague answer. A two-line brief beats a one-word question every time you try it.
Staying safe
Treat what you type like a postcard, not a sealed letter. What you type is sent to the company that runs the AI, and some products remember it by default. For example, Anthropic says memory is switched on by default in Claude.8 So:
- Do not share passwords, bank details, ID numbers, or private information about other people.
- Look in the settings for memory and “use my chats for training” options, and switch off what you do not want. Is Meta Muse safe to use? shows how on one popular app.
- Give agents the smallest set of keys. Let an agent ask before it pays, sends or deletes anything. Agent guardrails explains the four lines worth setting.
- Be suspicious of urgent voices and videos. AI can now copy a voice or a face. If “your boss” or “a family member” urgently asks for money, hang up and call them back on a number you already have. How AI fraud works shows the common tricks.
- Know when not to use it. When not to use AI lists the cases where it makes things worse.
AI is a tool, like a car. Most of the risk comes from how it is used: what you let it see, what you let it do, and whether you check its work.
Questions people ask
Is ChatGPT AI?
Yes. ChatGPT is a chatbot built on a large language model, which is a type of generative AI; OpenAI’s own researchers describe it that way.9 See what a large language model is.
Does AI think?
Not the way you do. It builds each answer by predicting what should come next, one word at a time, and even the companies that make these models say they do not fully understand how the models do most of what they do.10 Does AI understand anything? sets out both sides.
Is AI dangerous?
It can be, depending on how it is used: the US standards agency NIST says AI can bring real benefits and also real risks to people, organisations and society. Most everyday risk comes from trusting wrong answers, oversharing, scams and agents acting without a check; is AI dangerous? goes through the risks one by one.11
Is AI the same as a robot?
No. A robot is a machine with a body; most AI, like a chatbot or a spam filter, is software with no body at all. Some robots use AI to see or move, and industrial robotics is one of the fields where AI is put to work.12
Where to go next
Pick the line that sounds like you.
- “I want to write better requests.” Start with how to write AI prompts (Prompting 101).
- “Which AI should I use?” Open the Models hub, then press a button for each company.
- “I want to make pictures or videos.” Go to the Filmmaking hub and the cheat sheets.
- “I keep hearing about Chinese AI.” Read Chinese AI models, explained.
- “How is AI actually made?” See how AI is made, step by step.
- “How does ChatGPT actually work?” Read how ChatGPT works.
- “Should I be worried?” Read is AI dangerous?
- “Will AI affect my job?” Try the Living with AI hub.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. Why does AI sometimes give a wrong answer that sounds completely sure?
- The way it is trained and tested rewards a guess over saying it does not know
- It looks every answer up in a hidden encyclopaedia that is out of date
- It is lying on purpose to keep you chatting for longer
Show the answer
Answer: The way it is trained and tested rewards a guess over saying it does not know. The page explains that AI was shaped to always give a fitting answer, so a confident guess can come out even when it does not know.
2. What does the page mean by treating what you type like a postcard?
- Keep every message short, like a few lines on a card
- What you type goes to the company running the AI, so leave out private details
- Write to the AI in a friendly, personal tone
Show the answer
Answer: What you type goes to the company running the AI, so leave out private details. The page says what you type is sent to the company and may be remembered, so do not share passwords, bank details or other private information.
3. What is the main difference between a chatbot and an agent?
- A chatbot is always free, while an agent always costs money
- A chatbot can see pictures, while an agent can only read text
- A chatbot gives advice, while an agent can take actions for you
Show the answer
Answer: A chatbot gives advice, while an agent can take actions for you. The page compares a chatbot to a friend on the phone and an agent to an assistant you have handed your keys to.
Related on this site: AI for Older Adults · AI for People With Disabilities · How to Learn a Language With AI · How to Make a Meal Plan With AI · How to Plan a Trip With AI · Pros and Cons of AI · Types of AI, Explained · AI literacy meaning · How to Use ChatGPT · How to Use Google Gemini · Starter Prompts for Everyday Tasks, by Job · Learn AI Basics in 7 Steps · Free MIT AI Courses
- 1Google, Supervised Learning, updated 25 Aug 2025, as cited on this site’s machine learning page, read at source 1 Oct 2026“Supervised learning uses labeled data to train models that predict outcomes for new, unseen data.” … “like identifying spam or predicting precipitation.” ↑
- 2Google, Google Maps 101: How AI helps predict traffic and determine routes, 3 Sep 2020, read at source 5 Oct 2026“We then combine this database of historical traffic patterns with live traffic conditions, using machine learning to generate predictions based on both sets of data.” ↑
- 3Netflix Help Center, How Netflix’s Recommendations System Works, undated help article, read at source 5 Oct 2026“We estimate the likelihood that you will enjoy a particular title in our catalog based on a number of factors including” “your interactions with our service (such as your viewing history and how you rated other titles),” “other members with similar tastes and preferences on our service”; “All of these signals are used as some of the inputs that we process in our algorithms.” ↑
- 4Apple Machine Learning Research, Hey Siri: An On-device DNN-powered Voice Trigger for Apple’s Personal Assistant, October 2017, read at source 5 Oct 2026“The ‘Hey Siri’ detector uses a Deep Neural Network (DNN) to convert the acoustic pattern of your voice at each instant into a probability distribution over speech sounds.” ↑
- 5IBM, What Is Artificial Intelligence (AI)?, updated 2 Oct 2026, read at source 5 Oct 2026weak AI examples include “Amazon’s Alexa, Apple’s Siri, a social media chatbot or the autonomous vehicles promised by Tesla.” ↑a“Weak AI: Also known as ‘narrow AI,’ defines AI systems designed to perform a specific task or a set of tasks.” … “Strong AI: Also known as ‘artificial general intelligence’ (AGI) or ‘general AI,’ possess the ability to understand, learn and apply knowledge across a wide range of tasks at a level equal to or surpassing human intelligence.” ↑b“Directly underneath AI, we have machine learning, which involves creating models by training an algorithm” … “Deep learning is a subset of machine learning that uses multilayered neural networks” … “Generative AI, sometimes called ‘gen AI’, refers to deep learning models that can create complex original content such as long-form text, high-quality images, realistic video or audio and more in response to a user’s prompt or request.” ↑c
- 6Anthropic, Tracing the thoughts of a large language model, 27 Mar 2025, read at source 1 Oct 2026“Language models like Claude aren’t programmed directly by humans—instead, they’re trained on large amounts of data.” ↑a“Claude writes text one word at a time.” ↑b
- 7Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 1 Oct 2026“language models hallucinate because the training and evaluation procedures reward guessing over acknowledging uncertainty”. ↑
- 8Anthropic, Claude’s memory works everywhere, and you decide what’s in it, 25 Aug 2026, read at source 16 Sep 2026“Memory is on by default”. ↑
- 9OpenAI (Chatterji et al.), How People Use ChatGPT, 15 Sep 2025, read at source 5 Oct 2026“ChatGPT is based on a Large Language Model (LLM), a type of Artificial Intelligence (AI) developed over the last decade and generally considered to represent an acceleration in AI capabilities.” ↑
- 10Anthropic, Tracing the thoughts of a large language model, 27 Mar 2025, read at source 5 Oct 2026“Claude writes text one word at a time.” … “They arrive inscrutable to us, the model’s developers. This means that we don’t understand how models do most of the things they do.” ↑
- 11NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, January 2023, read at source 5 Oct 2026“Artificial intelligence (AI) technologies have significant potential to transform society and people’s lives” … “AI technologies, however, also pose risks that can negatively impact individuals, groups, organizations, communities, society, the environment, and the planet.” ↑
- 12European Commission Joint Research Centre (Delipetrev, Tsinaraki & Kostić), AI Watch: Historical Evolution of Artificial Intelligence, EUR 30221 EN, 2020, read at source 5 Oct 2026AI has provided solutions in “different application domains including data mining, industrial robotics, logistics, business intelligence, banking software, medical diagnosis, recommendation systems, and search engines.” ↑
Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 1 Oct 2026.
The comparisons on this page (library, phone keyboard, test-taker, nesting dolls, new colleague, postcard, house keys) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK