Module 1 of 6 · 20 minutes

What AI is — and isn't

AI is already part of your working day — in your inbox, your maps app and your bank. This module gives you a clear mental model of what AI is, what it isn't, and why that matters.

Learning objectives

  • Define AI in plain English and explain how it differs from rule-based software
  • Distinguish AI, machine learning, deep learning, generative AI and AI agents
  • Describe typical strengths and weaknesses of today's AI
  • Challenge common myths about AI

1.1 A working definition

Artificial intelligence (AI) is a broad term for computer systems that perform tasks we normally associate with human intelligence — recognising images, understanding language, making predictions, generating text or making recommendations.

A useful, practical definition comes from the EU AI Act, which describes an AI system as a machine-based system that infers from the inputs it receives how to generate outputs — predictions, content, recommendations or decisions. The key word is infers: instead of following rules a person wrote line by line, an AI system has learned patterns from data and applies them to new situations.

Traditional software

A person writes explicit rules: "If the invoice is over £10,000, send it to a manager." The software does exactly that, every time, and can explain itself by pointing at the rule.

AI system

The system learns from examples: thousands of past invoices labelled "fraud" or "not fraud". It then estimates how likely a new invoice is to be fraudulent — even for patterns no one wrote down.

Traditional softwareInvoiceRule: over £10k?Send to managerAI system10,000 past invoices(fraud / not fraud)Learned modelNew invoice → 87%likely fraudlearnsinfers
Rule-based software does exactly what it was told. An AI system learns patterns from examples, then infers an output — often a likelihood — for cases nobody wrote a rule for.

1.2 AI, machine learning and generative AI

You'll hear several terms used almost interchangeably. They are nested, like Russian dolls:

TermWhat it meansEveryday example
Artificial intelligenceThe whole field of making machines perform "intelligent" tasks.Route planning in a maps app
Machine learning (ML)The main way modern AI is built: systems learn patterns from data rather than being programmed with rules.Your bank flagging an unusual card payment
Deep learningMachine learning using large neural networks with many layers. Behind most recent breakthroughs.Face unlock on your phone
Generative AIDeep learning models that create new content — text, images, audio, video or code.ChatGPT, Claude, Gemini or Microsoft Copilot drafting an email
AI agentsSystems, usually built on generative AI, that can plan and take actions — such as searching, filling forms or running code — towards a goal.An assistant that books a meeting room and sends the invites
Artificial intelligenceroute planning, spam filtersMachine learninglearns from dataDeep learninglarge neural networksGenerative AIChatGPT, Claude, GeminiAI agentsGenAI + tools + actions
The terms are nested: generative AI is a kind of deep learning, which is a kind of machine learning, which is one approach to AI. Agents are built on generative AI and add the ability to act.

1.3 What AI is good at — and what it isn't

Today's AI is narrow: each system is very good at particular kinds of task, and can fail in surprising ways outside them. Even the most capable generative AI models are not "thinking" the way people do, and don't have human understanding, goals or common sense.

Typically strong at

  • Spotting patterns in large amounts of data
  • Drafting, summarising, translating and rewording text
  • Brainstorming options quickly
  • Classifying and extracting information from documents
  • Explaining concepts at different levels

Typically weak at

  • Guaranteeing facts are true — it can invent plausible answers
  • Knowing about events after its training data ends (unless it can search)
  • Taking responsibility or exercising accountable judgement
  • Understanding your organisation's unwritten context
  • Precise calculation, unless it uses a tool such as code
The "jagged frontier"

AI capability is uneven. A model may write a polished legal-style summary but miscount the words in a sentence. You can't assume that because AI handles one hard task, it will handle an easier-looking one. That's why this course keeps coming back to checking outputs.

1.4 Common myths

  • "AI is always objective." AI learns from data created by people, so it can reflect and amplify human biases.
  • "If it sounds confident, it's right." Generative AI is designed to produce fluent text. Fluency is not accuracy.
  • "AI will do my whole job." In practice, AI mostly changes tasks within jobs. People who learn to direct and check AI tend to get the most value.
  • "AI is only for technical people." The most widely used AI tools today are designed for everyone, in plain language.
Try it: spot the AI

List five tools you used this week (email, maps, streaming, banking, search…). For each, ask: does it make predictions, recommendations or generate content from patterns in data? You'll probably find AI in at least three.

Key takeaways

  • AI systems learn patterns from data and infer outputs, rather than following hand-written rules.
  • Generative AI is one branch of AI; agents add the ability to take actions.
  • AI capability is uneven — fluent output is not the same as reliable output.
  • AI changes tasks within jobs; people who can direct and check it gain the most.