Study Guide

What Counts as AI?

What artificial intelligence is and is not, how rule-based systems differ from systems that learn, how AI, machine learning, deep learning and generative AI fit inside one another, and why every AI in use today is narrow.

Lesson 8.1 AI Foundations

For: Learners aged 14 and over (Key Stage 4, sixth form) and adult beginners; no programming or mathematics needed

Stage: Think

One ordinary school day

Here are five things that happen in a typical school day:

  1. Your email moves a message promising a free phone into the spam folder.
  2. A maps app suggests a quicker way to walk to school because a road is closed.
  3. The canteen till refuses your card because you tried to spend more than the daily limit of £5.
  4. A streaming app suggests a programme “because you watched” something last night.
  5. You ask a chatbot to turn your exam dates into a revision timetable, and it writes one.

Think it through

Which of the five would you call artificial intelligence (AI)? Decide for each one before reading on, and note what made you decide.

Show what happens

Most people are sure about the chatbot and unsure about the rest. Many leave out the canteen till. This lesson shows that four of the five are AI by most definitions, and that the till is the most interesting case.

Stage: Explain

What AI is, and what it is not

Artificial intelligence (AI) is a broad name for computer systems designed to do tasks that would normally need human reasoning: recognising something, predicting what will happen, choosing an action, or producing content. The spam filter recognises, the maps app predicts journey times, the streaming app predicts what you will enjoy, and the chatbot produces text.

Three ideas about AI are worth correcting straight away:

  • It is not magic. Every result comes from patterns and calculations that can, in principle, be explained.
  • It is not conscious. It has no feelings, wishes or awareness. Saying a chatbot “wants” or “thinks” something is a figure of speech.
  • It is not always right. It makes mistakes, sometimes confidently, so people still need to check what matters.

Check your understanding

A chatbot replies, “I’m really excited to help you revise!” What does that tell you?
Show the answer

Answer: B. It has produced a friendly sentence; it does not feel excited.

The words are produced because they fit the conversation, not because of any feeling. Treat friendly or confident wording as style, not as evidence.

Stage: Discover

Following rules, or learning from examples?

Go back to the canteen till. Someone wrote a rule: if today’s spending would go over £5, refuse the payment. The till never changes that rule by itself.

Think it through

The spam filter was never given a list of every spam message. How do you think it learnt to spot new ones?

Show what happens

It was shown a very large number of emails already marked “spam” or “not spam”, and adjusted itself until it could tell them apart, including messages it had never seen.

That is the difference between two kinds of AI:

  • A rule-based system follows rules written by people. It is predictable and easy to explain, but it only knows what someone thought to write down.
  • Machine learning (ML) builds a system that learns patterns from examples, called data. It can handle cases nobody wrote a rule for, but its reasons are harder to see, and it is only as good as its examples.

Both are AI. A simple system can still be AI, and a simple explanation of AI can still be accurate.

Check your understanding

A school’s door system lets a card in only between 7 am and 6 pm. A photo app groups pictures of the same person together. Which is which?
Show the answer

Answer: C. The door is rule-based; the photo app has learnt from examples.

Asking “who decided the rule: a person, or the examples?” sorts most systems quickly.

Stage: Explain

How the terms fit together

You will often read AI, machine learning, deep learning and generative AI used as if they meant the same thing. They do not: each sits inside the one before it.

How the four terms nest

Each box is part of the box around it: Artificial intelligence, inside which is Machine learning, inside which is Deep learning, inside which is Generative AI and large language models.

  1. Artificial intelligence

    Any system designed to do tasks that need human-like reasoning, including rule-based ones.

    1. Machine learning

      AI that learns patterns from examples instead of being given every rule.

      1. Deep learning

        Machine learning that uses neural networks with many layers.

        1. Generative AI and large language models

          Deep learning models that produce new content, such as text or images.

An original diagram drawn for this lesson. Today’s generative AI systems are built with deep learning, so each is also machine learning and AI; not every AI system learns.

Deep learning uses a neural network: many simple connected units in layers, whose connections are adjusted during learning. Generative AI produces new content: text, images, audio, video or code. A chatbot that writes is one kind, built on a large language model. (Simpler programs could generate text long before deep learning, but the generative AI people mean today is built on it.)

Take the streaming app from the school day, which suggested a programme “because you watched” something last night.

  1. Is it AI? It predicts what you will enjoy, a task that would normally need human judgement. So it is at least in the outer box.
  2. Who decided the rule: a person, or the examples? Nobody wrote a rule for every viewer. It learnt from what millions of people watched and finished. So it moves into machine learning.
  3. Does it use a neural network with many layers? Many large services do, which would put it in deep learning. If you do not know, machine learning is the safe answer.
  4. Does it make new content? No: it picks from programmes that already exist. So it stops before the generative AI box.

Change one fact: if the same app wrote a new trailer for each viewer, step 4 would say yes, and it would belong in the innermost box.

Apply it

Using the diagram, which is the innermost box that fits the canteen till?
Show the answer

Answer: A. Artificial intelligence, but not machine learning.

The till is AI in the broad sense, but it does not learn, so it does not belong in the machine-learning box.

Stage: Discover

Making something new, or sorting what exists?

Look again at the school day. The spam filter puts each email into a category: spam or not spam. That is classification. The chatbot produces a revision timetable that did not exist before: that is generative.

Check your understanding

Which of these is generative AI?
Show the answer

Answer: C. A tool that writes a short poem about the school trip.

Labelling and routing are classification. Generative AI makes new content: text, images, audio, video or code.

Stage: Apply

Narrow or general?

Every AI in use today is narrow AI: built for a particular task or kind of task. It may be impressive within that area and limited outside it; a chess program cannot plan a revision timetable. Artificial general intelligence (AGI) would be able to learn and do any intellectual task a person can. It is a research goal, not a product, and no system in use today is generally accepted as meeting that definition.

So when a claim sounds dramatic, ask: is this describing narrow AI or general AI? Today the answer is always narrow.

A chat assistant can write about almost any subject, which makes it far more general than a chess program. It is still not AGI: it works within the kinds of task it was built and trained for, it makes mistakes a person would not, and it cannot learn and do any intellectual task a person can.

Apply it

A fictional advert says: “Meet Professor Pip, the AI tutor that understands every subject exactly like a human teacher.” Which response is best?
Show the answer

Answer: A. Read it as narrow AI described in general-AI words, and test it on your own subject.

The advert is invented for this lesson. The habit it practises is real: notice general-AI language, and judge the tool by what it actually does.

Go deeper Why “AI” is hard to define exactly

Experts do not all use the same definition, and what counts as AI shifts over time. Searching a map for the shortest route and checking spelling were once AI research problems; today most people call them ordinary software, although some courses still teach route-finding as part of AI. This lesson uses a broad, practical definition, which is why it counts the rule-based till.

Stage: Consolidate

Bring it together

Think it through

Without looking back, draw the four nested boxes and place three things from your own day in the innermost box that fits each one.

Show a model answer

A spam filter or photo grouping goes in machine learning at least, or in deep learning if you know it uses a many-layered neural network (photo grouping usually does); a chatbot or image generator goes in generative AI; a fixed rule like a spending limit goes in AI only.

Check your understanding

A library’s new system reads the cover of a returned book and puts it on the right shelf trolley. It learnt to do this from thousands of labelled photos of covers. Which description fits it?
Show the answer

Answer: B. Machine learning that classifies, built for one narrow task.

It learnt from examples (machine learning), it sorts into existing groups (classification), and like all AI today it is narrow. The library system is fictional.

  • AI is a broad name for systems that do tasks needing human-like reasoning. It is not magic, not conscious and not always right.
  • Rule-based systems follow rules people wrote; machine learning learns patterns from examples. Both are AI.
  • The terms nest: AI ⊃ machine learning ⊃ deep learning ⊃ generative AI.
  • Generative AI makes new content; classification sorts what exists.
  • All AI today is narrow; AGI does not exist, so test dramatic claims with “narrow or general?”.

Sources and scope

  • What this lesson is. The first of four introductory AI Foundations lessons, written for this site. It is a draft awaiting the site owner’s editorial review, and an introduction, not preparation for any certificate. Its curriculum links are provisional: the wording of each linked specification point has been checked against the published specification, but no link has yet been verified by the site owner. The links show where the lesson may help with a course; they do not imply that any examination board has reviewed or endorsed it.
  • Planning reference. Its scope was planned against areas 1.1 to 1.4 and 3.1 of the UK AI Council’s AI Foundation technical syllabus (version 1.0), using the site’s own mapping. Nothing here is endorsed by, affiliated with or approved by the UK AI Council, and no text or example is taken from its course materials.
  • Original material. The school-day scenario, the questions, the advert and the library system (both fictional) and the diagram were written for this lesson.
  • Simplified on purpose. The lesson uses a broad, practical definition of AI, under which a system following one written rule counts; narrower definitions would leave it out. The nesting of the four terms describes how they are used today.