What is artificial intelligence: a plain guide

Everyone is talking about artificial intelligence: on the news, at work, even around the table at family celebrations. Some see it as a miracle that will solve every problem; others fear it will “put everyone out of work”. The reality is much simpler and more interesting. In this article we explain what artificial intelligence is without technical jargon: how AI, machine learning, deep learning and language models differ, how a computer “learns”, which AI tools you already use every day without noticing, and what generative AI can and cannot do. We then compare common myths with the facts, look at how to protect your personal data, suggest where to start learning about AI, and finish with a practice exercise and a checklist.

What is artificial intelligence: a simple definition

Artificial intelligence (AI for short) is the general name for computer systems that perform tasks which normally require human intelligence. These include recognising speech, identifying objects in a photo, translating text, answering questions or finding the shortest route.

To see how this differs from an ordinary program, take an example. In an ordinary program, the developer writes every rule: “if the user enters the wrong password three times, lock the account”. The computer simply follows that rule. Step-by-step instructions like this are explained in detail in our article on algorithmic thinking.

Some tasks, however, cannot be written down as rules. How would you write rules that answer the question “is there a cat in this photo?” Cats come in different colours and positions, in bright light and in the dark. This is where the AI approach comes in: instead of rules, the computer is given thousands of examples and “finds” the patterns itself.

An important point: today’s AI is a tool for narrow tasks. A translation system cannot drive a car, and a chess program cannot write a letter. Even the most powerful chatbots do not understand the world the way a person does and have no wishes of their own; they work on patterns in data.

AI, machine learning, deep learning and language models: what is the difference

These terms are often mixed up. Picture them as boxes nested inside one another.

• Artificial intelligence (AI) is the biggest box. Any computer system that seems “smart” belongs here, even one that runs on hand-written rules.
• Machine learning is a box inside AI. Here the computer works out the rules from examples by itself. For instance, a bank uses thousands of past loan applications to build a model that flags which new applications need extra checks.
• Deep learning is a box inside machine learning. It uses multi-layered “neural networks”. These networks are inspired by the human brain but are not a copy of it; in essence they are a very long chain of mathematical operations. Deep learning is especially strong at working with images, sound and text.
• Language models (LLMs, large language models) are models built with deep learning and trained on enormous amounts of text. They have learnt to continue text: in reply to your question they put together the “most likely” answer word by word. Chatbots run on models of this kind.

One more term is generative AI: systems that create new content such as text, images, audio or video. Language models are one type of generative AI. Where earlier AI mostly “recognised” things (is this spam or not?), generative AI “creates” (write me a letter on this topic).

In short: every language model is deep learning, all deep learning is machine learning, and all machine learning is AI. But not the other way round.

How a computer “learns”

The essence of machine learning is finding patterns in examples. Let us look at a real-life case.

Imagine Bahodir, who owns a small bakery in Samarkand and wants to know in advance how many loaves of bread to bake each day. He takes his records for a year: the day of the week, the weather, whether it was a holiday, whether there was a wedding in the neighbourhood, and how many loaves were sold that day. The example is illustrative, but the logic is real.

1. Collecting data. Each day is one example. The features of the day (day of the week, weather) are the input; the number of loaves sold is the result. A collection of examples like this is called a dataset.
2. Training. The program goes through the examples and looks for relationships: for instance, sales rise on Fridays and before holidays and fall on rainy days.
3. Testing. Part of the data is kept back from training, and the model is then checked on how accurately it “guesses” those days. It is like an exam: a pupil’s knowledge is tested on questions they have not seen before.
4. Using the model. The model gives a forecast for tomorrow. Bahodir compares it with his own experience and makes the decision.

Note that the model does not understand “why” more bread sells on Fridays. It only sees that the pattern exists.

Why data quality decides everything

Specialists have a well-known saying: “garbage in, garbage out”. A model learns only from the data it has seen, so:

• If the data is incomplete, the model does not know about some situations. If Bahodir’s records do not include the month of Ramadan, the model cannot predict how sales change during it.
• If the data contains errors, the model learns the errors too. If someone records one day’s sales twice, the model draws the wrong conclusion.
• If the data is one-sided, the model will be one-sided as well. For example, a face recognition system trained mostly on photos of people with one kind of appearance recognises others less well. This is called bias in data and is considered a serious problem.

That is why much of the work in AI is not building the model but collecting, cleaning and checking the data.

Artificial intelligence in everyday life

Many people use AI every day without realising it. Here are some familiar examples:

• Online translator. It translates text, or a menu through your camera, from one language to another. Modern translators run on neural networks and take the whole sentence into account.
• Finding a route on a map. A navigation app takes traffic into account, suggests the fastest way from one side of Tashkent to the other and estimates your arrival time.
• Spam filter. Your email automatically moves unwanted and fraudulent messages into a separate folder. The filter learnt what counts as spam from millions of emails.
• Recommendations. A video platform suggests the next video, and an online shop shows products you “might like”. The system finds patterns in your behaviour and that of users similar to you.
• Voice assistant. Say “set an alarm for seven tomorrow” to your phone, and it turns speech into text, works out the meaning and carries out the command. Voice input works better and better in Uzbek too, but quality varies by app and device.
• Your phone’s camera. The camera detects faces and focuses on them, brightens evening shots, and the gallery sorts photos into types such as “documents” or “food”. Unlocking your phone with your face also relies on AI.
• Banking app. Many banks use automated systems to spot suspicious transactions: if a purchase is unusual, the app may ask you to confirm it.

In all these examples AI acts as an assistant: it makes decisions easier, but the final word usually stays with a person.

Generative AI: what it can and cannot do

What has drawn the most attention in recent years is chatbots and image-generating services. They are genuinely useful, but you need to know their limits.

What it can do

• Write and edit text. A draft letter, the text of an announcement, a rewrite in a formal style, spotting typos.
• Explain. Put a complex topic into simple words, with examples. For instance, a pupil called Dilnoza might ask for a physics formula to be explained in different words.
• Summarise. Give the main points of a long text in a few bullet points.
• Suggest ideas. An event plan, the structure of a presentation, a list of questions.
• Translate and write code. Translate text, write simple code and explain an error.

How to use these tools at work is covered in detail in our article on AI tools at work.

What it cannot do, or does badly

• Hallucination. This is the most important limitation. A language model can give an answer that looks convincing but is wrong: a book that does not exist, an invented date, the wrong name of a law or an incorrect phone number. Instead of saying “I don’t know”, the model produces plausible-sounding text. So always check figures, dates, names, and legal and medical information against official sources.
• Out-of-date knowledge. A model is trained on data up to a certain date. It may know nothing about later events, new prices or rules that have changed. Some services can search the internet, but even that does not fully protect against mistakes.
• Uneven quality in Uzbek. Models are trained mostly on widely spoken languages such as English. Answers in Uzbek can contain odd word combinations, a mix of Latin and Cyrillic letters, or awkward phrases carried over from Russian or Turkish. Always read through and correct an official text yourself.
• Mistakes in calculations. A language model is not a calculator. Check complex calculations in a spreadsheet or on a calculator.
• Responsibility. The user remains responsible for an AI answer. “The bot wrote it” is no excuse for an employer or a teacher.

Myths and facts

Myth: “AI will take all the jobs.” Fact: AI automates certain tasks, such as sorting routine emails or drafting text. But most jobs are made up of many different tasks: dealing with people, making responsible decisions, manual work, reading a situation. The more likely change is that the content of work changes: an employee who can use AI tools works faster. That is why the most useful skill today is learning to work with AI properly.

Myth: “AI thinks and understands.” Fact: a language model picks the next word based on statistical patterns in text. It may sound very natural, but there are no feelings, intentions or human-like understanding behind it. A bot that says “I’m happy” is not actually happy.

Myth: “Computers don’t make mistakes.” Fact: AI repeats the errors and biases in its data and hallucinates. Results need to be checked with a critical eye.

Myth: “Only programmers understand AI.” Fact: you do not need to program to use AI tools. Anyone can learn the core ideas, just as you do not need to assemble an engine to drive a car.

Myth: “Text written by AI is always good.” Fact: it is often generic, repetitive and impersonal. The best results come when a person works on an AI draft.

Safety and personal data

Everything you type to a chatbot is sent to the service’s servers. Some services store conversations and may use them to improve their models. So here is a simple rule: do not write to a bot anything you would not tell a stranger.

Things not to send to a bot:

• passport details, your personal identification number (PINFL), bank card numbers, SMS codes and passwords;
• other people’s personal data: client lists, phone numbers, medical information;
• your employer’s confidential documents, contracts and internal reports.

If you need a document analysed, first replace names, numbers and addresses with placeholders such as “Client A” or “Employee B”.

Useful settings: many chat services have options such as Data Controls or Privacy under Settings. There you can delete your chat history or limit the use of your conversations for training the model. The names vary from service to service, so read the terms of use and the privacy policy. At work, first ask which tools your organisation allows.

Watch out for fraud. AI has made it easier to create fake voices, fake videos (deepfakes) and very convincing scam messages. If a “relative” calls unexpectedly and asks you to transfer money urgently, hang up and call them back yourself on a number you know. Detailed protection measures are in our article on online safety.

One more thing: do not install an “AI app” from an unknown source, even if it looks official. Download apps only from the official store and check the permissions they ask for.

Where to start learning about AI

Depending on your goal, there are two paths.

As a user

If you want to use AI effectively for work and study:

1. Choose one tool and use it every day for a week for a small task: a draft letter, shortening a text, explaining an unfamiliar term.
2. Learn to write precise requests. Give context (who it is for, what style, what length), provide an example and ask for corrections. Instead of “write a letter”, write “write a short, polite request to the mahalla chairman about a broken street light, no more than 5 sentences”.
3. Check every answer. Facts against official sources, calculations on a calculator.
4. Improve your English. Many guides, courses and tools are available in English.

As a specialist

If you want to build AI systems yourself, the path is longer but clear:

1. Working with data: spreadsheets, then SQL.
2. Programming: Python is the most widely used language in this field.
3. Maths basics: statistics, probability and linear algebra at an introductory level.
4. Machine learning: start with simple models, then move on to neural networks.
5. Projects: small projects with real data and a portfolio.

A step-by-step plan is set out in our Data Science roadmap.

Practice exercise and checklist

Do this exercise with any free chatbot. It takes 30–40 minutes. Do not use personal data.

1. Explanation. Ask the bot: “What is machine learning? Explain it as if to a 12-year-old, with one example.” Compare the answer with the definition in this article.
2. Spot a hallucination. Ask about a topic you know well: the history of your home town or a specific fact from your profession. Check every date and name in the answer. How many mistakes did you find?
3. Quality in Uzbek. Ask the same question in Uzbek and in English. Compare the quality of the answers: which one has more spelling mistakes and odd phrases?
4. A precise request. First type “Write a letter”, then repeat the request with full context (to whom, about what, in what style, how long). Compare the two results.
5. Everyday AI. Find three examples of AI on your phone: the camera, word suggestions on the keyboard, maps, a translator or search in the photo gallery.
6. Privacy. Open the settings of the bot you use and find out how your chat history is stored.

Checklist

• I can explain the difference between AI, machine learning, deep learning and language models in one sentence.
• I know that a model learns from examples and that data quality determines the result.
• I can name at least five examples of AI from my everyday life.
• I know what a hallucination is, and I check figures, dates and names against official sources.
• I do not give a bot passwords, card numbers, passport details or other people’s personal data.
• I do not immediately trust a “familiar voice” asking for an urgent money transfer; I call back and check.
• I read and correct AI-written text myself before sending it.
• I have chosen my next learning step: the user path or the specialist path.

Artificial intelligence is neither magic nor a threat; it is a new kind of tool. The better you understand it, the more safely and usefully you can use it.

If you would like to learn data skills, programming and machine learning systematically with a teacher, take a look at our association’s free Data Science programme and other training programmes. When you are ready, submit an application and our specialists will get in touch.