What AI really is (and what it isn't)
Clear definitions of AI, machine learning, deep learning and generative AI, how they fit together, and what today's AI can and cannot do.
🎯 By the end of this lesson you will be able to:
- Explain AI, machine learning, deep learning and generative AI in one sentence each
- Tell the difference between traditional software and software that learns
- Recognise what today's AI is good at, and where it needs a human
In this lesson
Why this matters for everyone
AI is now part of almost every software team's work. Developers use coding assistants, testers generate test ideas, analysts summarise interviews, and managers draft plans and reports. You do not need to become a data scientist, but you do need a clear, accurate picture of what AI is, so you can use it well, explain it to others and spot when it is being oversold.
This first lesson gives you that picture without maths or code. Everything later in the course builds on these ideas.
The four terms everyone mixes up
People use "AI", "machine learning" and "generative AI" as if they mean the same thing. They are related, but each one is a smaller part of the one before it.
- Artificial intelligence (AI): any technique that makes a computer do something we would normally call intelligent, such as recognising speech, recommending a film or playing chess. It is the broad goal, not one technology.
- Machine learning (ML): a way of building AI where the computer learns patterns from examples instead of following rules a person wrote by hand.
- Deep learning: machine learning using large neural networks with many layers. It powers modern image recognition, speech recognition and translation.
- Generative AI (GenAI): deep learning models that create new content, such as text, images, code or audio, rather than only sorting or predicting. ChatGPT, Claude and Gemini are generative AI.
Traditional software versus software that learns
In traditional software, a person writes the rules. For example: if the order total is over 100, apply free shipping. You give the program data, it applies your rules, and you get answers. This works brilliantly when the rules are clear.
Some problems have rules that are too complicated to write down. Try writing rules that recognise a cat in any photo, or decide whether a customer email is angry. Machine learning flips the approach: you give the computer lots of examples together with the right answers, and it works out the patterns, which we call a model.
Spam filtering, two ways
Rules approach: block emails containing "FREE MONEY", "WINNER" or more than five exclamation marks. Learning approach: show the model 100,000 emails already labelled spam or not spam.
Rules are easy to understand but spammers quickly change their wording. The learned model picks up thousands of subtle signals and adapts when it is retrained on new examples.
What today's AI is good at
| Strong at | Examples at work |
|---|---|
| Language tasks | Summarising, drafting, rewriting, translating, explaining |
| Finding patterns | Classifying tickets, spotting anomalies, recommending items |
| First drafts | Code, test cases, user stories, emails, reports |
| Working with lots of text quickly | Searching documents, comparing versions, extracting fields |
Where AI still needs a human
- Being reliably correct. Generative AI can sound confident while being wrong. We cover why in Lesson 4.
- Knowing your context. It does not know your company, your customers or yesterday's decision unless you tell it.
- Judgement and accountability. It can suggest, but people are responsible for decisions that affect customers, money, safety or careers.
- Tasks with no good examples. Learning needs data. Rare or brand-new situations are hard.
What this means for your role
Developer
You will use AI as a pair programmer and may build features on top of AI models. Knowing that models learn patterns, rather than follow your rules, explains why you test AI features differently from normal code.
Tester / QA
AI features do not always give the same output for the same input. Your job shifts from checking exact outputs to checking quality, safety and consistency across many examples.
Business Analyst / PO
Your skill in describing problems and acceptance criteria matters more, not less. Good AI features start with a clearly defined task and examples of good and bad results.
Tech Lead / Architect
You will be asked whether something should use AI at all. Often a simple rule is cheaper, faster and more reliable. Reach for machine learning when the rules are too complex to write.
Project / Delivery Manager
AI work has more uncertainty than normal features: you learn whether something works by trying it. Plan for experiments and checkpoints, not only a fixed scope.
Scrum Master / Agile
Teams adopting AI need space to experiment and share what works. Short learning spikes and retrospectives on AI use help the whole team improve together.
Try it: sort five tasks
- Write down five tasks from your own week, for example "write release notes" or "triage bug reports".
- For each task, decide whether it is best done by a fixed rule, by AI with a human checking, or by a human alone.
- For one AI task, write one sentence describing what a good result looks like. You will use this skill in every later lesson.
Key takeaways
- AI is the broad goal; machine learning, deep learning and generative AI are progressively narrower parts of it.
- Traditional software follows rules people write; machine learning learns patterns from examples.
- Generative AI is excellent at drafts, summaries and language tasks, but can be confidently wrong.
- People stay responsible for decisions, so treat AI as an assistant, not an authority.
Key terms
- Model
- What a machine learning system produces after learning from data: a function that takes an input and gives a prediction or output.
- Generative AI
- AI that creates new content such as text, images or code.
- Deep learning
- Machine learning using neural networks with many layers.
Check your understanding
Answer all 5 questions. Score 3 or more to complete the lesson.
1. Which statement best describes machine learning?
2. How do AI, machine learning, deep learning and generative AI relate?
3. When is a simple rule usually better than machine learning?
4. Which task is today's generative AI weakest at without human checking?
5. What is a model in machine learning?