Artificial Intelligence (AI)

Why AI is needed

Before learning the formal meaning of AI, begin with an ordinary problem.

Imagine that your phone contains 10,000 photographs. You want to find every photograph containing a dog.

You could inspect the photographs one by one. That would take a long time. You could also try to write rules:

If the picture has two pointed ears...
and four legs...
and a tail...
then call it a dog.

But cats, toy animals, wolves, shadows, and unusual camera angles would break those simple rules. The real world contains too much variation.

This is one reason AI exists: to help computers perform tasks where the correct decision cannot always be described with a few simple instructions.

Build the idea one piece at a time

1. A computer receives something to examine, such as a photograph. This is the input.
2. A method examines the input. It might use rules, search, planning, or a learned model.
3. The computer produces a result, such as dog. This is the output.
4. People test whether the result is useful, correct, fair, and safe.

Photograph → AI system → "dog"
   input                    output

The words Artificial Intelligence describe the goal and field surrounding systems like this. They do not mean the machine has a human mind.

Key terms

• Intelligence: Here, the ability to perform tasks such as recognizing, predicting, planning, generating, or deciding.
• System: All the connected software and components working together.
• Model: A mathematical component that maps input to output.
• Prediction: A model’s best calculated answer, which may still be wrong.

Check your understanding

If a calculator follows an exact formula, is it automatically AI? No. It performs useful computation, but a simple exact calculation normally does not require AI.

If software recognizes many different ways a person might ask the same question, could AI help? Yes. Human language contains variation that is difficult to cover with a short list of fixed rules.

Imagine you hire two employees to sort your mail. The first one you hand a strict rulebook: “If the envelope is red, put it in the bin marked Urgent. If it has a stamp from another country, put it in the bin marked International.” That employee never thinks — they just follow your rules, forever, without getting better or worse at the job.

The second employee, you don’t give a rulebook. Instead, you show them a thousand examples of correctly sorted mail and say, “Learn the pattern yourself.” After a while, they start sorting mail you’ve never shown them before — and they get better the more mail they see.

That second employee is doing something close to what we call Artificial Intelligence.

The simple definition

At its core, Artificial Intelligence is the field of building machines that can perform tasks which normally require human intelligence — things like recognizing images, understanding language, making decisions, or spotting patterns in data.

That’s it. It’s not one algorithm, one product, or one company’s technology. It’s a whole field of computer science, the way “medicine” is a whole field rather than a single treatment. When people say “AI” casually, they’re usually pointing at one specific result of this field — a chatbot, a recommendation feed, a self-driving car’s vision system — but AI itself is the broader umbrella all of those live under.

Why does AI exist? What problem does it solve?

Ordinary software is extremely good at doing exactly what it’s told, exactly the same way, every time. That’s perfect for calculating a bank balance or rendering a spreadsheet. But a huge number of real-world tasks are hard to describe as a fixed set of rules:

• How do you write a rule for “this is a photo of a cat” that covers every possible cat, angle, lighting condition, and breed?
• How do you write a rule for “this sentence is sarcastic” that works across every way humans phrase sarcasm?
• How do you write a rule for “this transaction looks fraudulent” when fraud patterns change every week?

You could try. Early computer scientists did try, for decades. But the rulebook approach breaks down fast, because the real world is messy, full of exceptions, and constantly changing. AI exists to solve exactly this class of problem: tasks that are easy for a human to do intuitively but nearly impossible to write down as a complete, explicit set of instructions.

What would happen without it

Without AI, every one of those messy, judgment-based tasks would need to be handled either by a human directly, or by a programmer trying to anticipate every possible case in advance — an approach that’s slow, brittle, and constantly breaks the moment reality does something the rulebook didn’t predict. Spam filters would miss anything phrased slightly differently than before. Voice assistants would only understand a handful of rehearsed phrases. Fraud detection would always be one step behind fraudsters.

Rules vs. learning: the shift that defines AI

Here’s the technical distinction hiding inside the mail-sorting analogy above, and it’s the single most important idea in this entire field.

Traditional programming:

flowchart LR
    A[Rules written by a human] --> B[Program]
    C[Data / Input] --> B
    B --> D[Output]

A programmer writes the rules. The computer just executes them on whatever data comes in.

Machine intelligence (the modern core of AI):

flowchart LR
    A[Data / Examples] --> B[Learning Process]
    C[Correct Answers] --> B
    B --> D[Rules the system discovered itself]
    D --> E[Used later on new, unseen data]

Instead of a human writing the rules, the system is shown many examples (data) along with the correct answers, and it works out the underlying pattern on its own. Once it has learned that pattern, it can apply it to brand-new situations it has never seen before.

This shift — from “human writes the rules” to “system learns the rules from data” — is why AI can tackle problems that stumped rule-based programming for decades. You’ll see this idea again, in much more technical detail, when you get to Machine Learning, which is really the dominant modern method for building AI.

A quick but important nuance

Not everything historically called “AI” involves learning from data. Some of the earliest AI systems — like a chess program that searches through millions of possible future moves and picks the best one — used clever search and logic instead of learning. That’s still AI, because it performs a task requiring intelligence, even though no “learning from examples” happened. So think of AI as defined by the task (does it require intelligence?), not by one specific technique.

How AI actually shows up in the real world

You’ve almost certainly used AI today without thinking about it:

• Email spam filters — learned from millions of emails what “spam” tends to look like.
• Recommendation feeds (Netflix, YouTube, Spotify) — learned patterns in what people with similar tastes tend to enjoy next.
• Voice assistants (Siri, Google Assistant) — convert your speech to text, understand your intent, and generate a response.
• Fraud detection at banks — flags transactions whose pattern doesn’t look like your usual behavior.
• Photo apps that auto-tag faces — recognize visual patterns in pixels that correspond to specific people.
• Chatbots and tools like ChatGPT or Claude — a much more advanced and general form of AI, built on techniques you’ll meet soon in this glossary.

Notice the range: some of these are narrow and do one job well (a spam filter can’t hold a conversation). Others, like modern chat-based AI systems, are far more general. That range points to a useful way of categorizing AI.

Types of AI: narrow vs. general

This is one of the few genuinely important categorical distinctions in the field, so it’s worth being precise about it.

Narrow AI (also called “Weak AI”) is any AI system built to do one specific task, or a limited family of related tasks, and nothing outside that. A chess engine, a spam filter, and a medical image scanner that detects tumors are all narrow AI — each is excellent at its one job and useless outside it.

Every AI system that exists today, including the most advanced chatbots, falls into this category. They can appear broad because they handle many types of questions, but they are still fundamentally pattern-matching systems trained for specific kinds of tasks (like predicting text), not independently reasoning minds.

General AI (also called “Strong AI” or AGI — Artificial General Intelligence) refers to a hypothetical system with human-like flexibility: able to learn any intellectual task a person can, transfer knowledge between totally unrelated domains, and reason about the world the way a human does. This does not exist yet. It’s a major topic of research, debate, and speculation, but as of today, no AI system has this level of general capability — no matter how impressive its outputs sometimes look.

Keeping this distinction in mind will save you from a very common misconception, covered next.

The biggest misconception: “AI understands and thinks like a human”

When a chatbot gives you a thoughtful-sounding answer, it’s natural to assume something is “understanding” your question the way a person would. Under the hood, that’s not quite what’s happening.

Modern AI systems are, technically, extremely sophisticated pattern-recognition and pattern-generation engines. They’ve learned statistical relationships from enormous amounts of data — which words tend to follow which other words, which pixels tend to form which shapes, and so on. They don’t have beliefs, goals, consciousness, or a model of truth the way humans do. When an AI system gets something confidently wrong (sometimes called a “hallucination”), that’s not a bug where the “understanding” briefly failed — it’s the expected behavior of a system whose real job is producing statistically plausible output, not verified truth.

This isn’t a reason to dismiss AI — the results are often genuinely useful and impressive — but it’s the difference between “this system is intelligent-acting” and “this system is intelligent-being.” Keeping that distinction clear will make every concept later in this glossary click faster.

Advantages and limitations

AI is powerful where problems are pattern-heavy, data-rich, and repetitive at scale — things like sorting, classifying, predicting, and generating content based on learned patterns. It can operate 24/7, process far more data than any human could, and improve as it sees more examples.

It struggles where problems require genuine common-sense reasoning about a novel situation, verified factual accuracy, transparency about why a decision was made, or accountability for high-stakes mistakes. It also inherits and can amplify biases present in the data it learned from — if the training examples were skewed, the system’s “learned patterns” will be skewed too.

Recognizing which side of that line a task falls on is one of the most practical skills you’ll build as you go deeper into this field.

A mental model you can reuse

Think of an AI system as a goal-directed decision machine:

flowchart LR
    A[Goal: What should happen?] --> B[Input: What can the system observe?]
    B --> C[Method: rules, search, or a learned model]
    C --> D[Output: prediction, content, or action]
    D --> E[Check: Was the result useful and safe?]
    E -. feedback .-> C

The middle box is important. AI is a broad field, so an AI system does not always have to learn from data. A chess program can search possible moves. A planning system can compare possible routes. A modern assistant may combine a Large Language Model with search, memory, tools, and fixed safety rules.

Remember: AI describes the larger goal of making machines perform capabilities that appear intelligent. Machine Learning is one important way to build such machines, but it is not the whole field.

A simple example: sorting photos

Suppose a phone must group photos containing dogs:

1. Goal: Find dog photos.
2. Input: The pixels in each photograph.
3. AI method: A trained image model detects visual patterns associated with dogs.
4. Output: A label such as dog: 96% confidence.
5. Application: The gallery places the photograph in a searchable “Dogs” group.

The percentage is confidence, not a guarantee. A toy wolf, a blurry cat, or an unusual camera angle can still fool the model.

AI, ML, Deep Learning, and Generative AI

Artificial Intelligence
├── Rule-based systems, search, planning, and other techniques
└── Machine Learning
    ├── Classical Machine Learning
    └── Deep Learning
        └── Many modern Generative AI systems

• AI is the largest umbrella.
• Machine Learning learns useful patterns from examples.
• Deep Learning uses multi-layer neural networks.
• Generative AI produces new content such as text, images, audio, video, or code. Many current generative systems use Deep Learning.

These categories overlap, but they are not interchangeable names.

Production reality: the model is only one component

A real AI product normally surrounds the model with ordinary software:

User request
   ↓
Authentication → input checks → AI model or agent → safety checks
   ↓                                      ↓
Logs and monitoring ← application logic ← final response or action

For example, a customer-support assistant may retrieve company documents, ask a model to draft an answer, block unsafe output, record latency and cost, and send uncertain cases to a human. The complete product is the AI system; the model is one component inside it.

When AI is and is not a good choice

Consider AI when:

• The task involves language, images, speech, predictions, recommendations, or patterns that are difficult to express as complete rules.
• Some uncertainty is acceptable and results can be measured, reviewed, or corrected.
• You have suitable data, a capable model, or a reliable external AI service.

Prefer simpler software when:

• The correct behavior can be written as a small set of exact, stable rules.
• Every result must be perfectly repeatable and auditable.
• A database query, formula, or ordinary program solves the problem more cheaply and reliably.

A production checklist

Before calling a feature “AI-powered,” ask:

• What exact user problem are we solving?
• What inputs may the system use, and do we have permission to use them?
• How will we measure usefulness, errors, bias, safety, latency, and cost?
• What happens when the model is uncertain or wrong?
• Can a human review high-impact decisions?
• Is a non-AI baseline simpler or better?

Ordinary software, today’s AI, and AGI

System	What it can do	Example
Ordinary rule-based software	Follows rules written by people.	A calculator applies fixed arithmetic rules.
Narrow AI	Learns or applies patterns for a limited family of tasks.	A photo app identifies faces; GPT or Gemini answers across many language tasks but still has designed limits.
Artificial General Intelligence (AGI)	A proposed system able to learn and reason broadly like a capable human across unfamiliar domains.	No system has been universally demonstrated and accepted as AGI.

The word AI does not automatically mean human-like intelligence. Nearly every deployed AI system today is narrow AI, even when its interface can discuss many subjects.

Where this fits in what comes next

Artificial Intelligence is the umbrella term for this entire glossary — everything you’ll read about next lives inside it. The next major idea to understand is Machine Learning, which is the dominant modern approach used to actually build AI systems: instead of a human writing rules, the machine learns them from data, exactly as described above.

From there, you’ll move into more specific techniques like Deep Learning, and eventually into the systems powering today’s most talked-about AI, like Large Language Models and Generative AI. Each of those is best understood as a more specific, more powerful tool built on the general idea introduced here.

In one sentence

Artificial Intelligence is the broad field of making machines perform tasks that normally require human-like intelligence — and almost everything impressive you’ve heard about in AI recently is really a story about how that intelligence gets built, which is exactly where this glossary goes next.

Related Terms

• Machine Learning
• Deep Learning
• Neural Networks
• Generative AI
• Large Language Models (LLMs)