Phase 1: Prepare
Step 1 Warm-up recall
List two systems in your organisation that you would call “AI”, and note what each one actually does.
Private: nothing you type here is sent or stored.
Step 2 Learning goal
By the end of this lesson you can:
- name the three families (Foundation);
- give a typical use and risk for each (Core);
- classify a proposal by family (Core).
Phase 2: Teach
Step 3 A familiar example
A vendor says their product “uses AI”. The word covers everything from a fixed set of if-then rules to a system that writes paragraphs. Which one it is changes the price, the risk and the evidence you need.
Step 4 Clear explanation
For decisions, it helps to sort AI into three families by how the system gets its behaviour.
| Family | What it does | Typical example | Typical risk |
|---|---|---|---|
| Rules-based automation | Follows fixed rules written by people | Route an invoice over a set value to a manager | Rules miss cases nobody foresaw |
| Predictive models | Learns patterns from past data to estimate an outcome | Estimate the chance a customer will leave | Learns the gaps and biases of its data |
| Generative AI | Produces new text, images or code from a request | Draft a first reply to a customer email | Fluent output that is wrong or invented |
Rules-based automation follows rules people wrote. It is predictable and easy to audit, but it handles only the cases it was written for. Predictive models learn patterns from past data, so they handle messy cases, but they reflect the gaps and biases of that data; the entry on machine learning gives background. Generative AI produces new content; it is flexible and fast, but its output must be checked because it can be fluent and wrong.
The figure shows the order of thinking: start with a problem worth solving, choose the family that fits, check data and ownership, and run a small pilot. Many products combine families, so ask the vendor which part does what. For a broader picture, read What Is Artificial Intelligence, Really?
Step 5 Worked example (4 marks)
Classify: “send any refund above £500 to a manager for approval”.
- The rule is written by a person. (1 mark)
- It follows a fixed condition. (1 mark)
- Family: rules-based automation, not AI learning from data. (1 mark)
- Implication: cheap, auditable, and no need for a model. (1 mark)
Phase 3: Practise
Step 6 Learner practice (instant-marked)
1. A system that follows fixed if-then rules written by people is:
The behaviour comes from written rules.
2. A system that learns patterns from past data to estimate an outcome is:
It learns from examples.
3. Which family carries the risk of fluent but wrong output?
Generated content must be checked.
4. A key weakness of rules-based automation is that it:
Rules cover only what was written.
Phase 4: Close
Step 7 Recap
- Three families: rules-based, predictive, generative.
- Each fits different problems and carries different risks.
- Ask which part of a product is which family.
Step 8 Exit ticket (ungraded)
Classify three systems in your organisation by family.
Private: nothing you type here is sent or stored.
Step 9 Go further (Extension)
Read the encyclopedia entry on artificial intelligence and note one kind of AI not covered by the three families here.
Common mistakes
| Mistake | Why it fails |
|---|---|
| Calling everything AI | Ask what produces the behaviour. |
| Buying a tool before naming the problem | The problem should come first. |
| Ignoring the family when judging risk | Risks differ by family. |
Limits. The three families are a simplification for decisions; technical taxonomies are finer.