How Does AI Actually ‘Think’?
Quick 5-minute starter: understand that AI predicts patterns rather than "thinking", and why hallucinations occur.
- Lesson time
- 5 min
- Fits
- Lesson Starter (5 min)
- Format
- Discussion
- Level
- Beginner
- Downloads
- 32
- Views
- 217
32 downloads 217 views
Before you teach
Get ready
- Open the PDF (or the SlideForge deck) on the board before the class arrives.
- Arrange students into pairs.
- Optional: have ChatGPT or a similar tool ready to show one answer live.
Teacher notes
Students may anthropomorphise AI β gently correct 'it thinks' to 'it predicts'. Avoid being dismissive of AI's usefulness while explaining its limitations. Good analogy: AI is like a very sophisticated autocomplete, not a thinking being. This understanding helps students use AI more critically and effectively.
Learning objectives
- Understand that AI predicts patterns rather than ‘thinking’
- Recognise the difference between pattern prediction and understanding
- Learn why AI ‘hallucinations’ occur
The lesson
-
Starts at 0:00
Have the title up as the class comes in.
Teacher tip
Frame it as curious exploration: how does a tool they already use actually work?
-
Starts at 0:15
Ask the question, give pairs 60 seconds, then take the vote.
Teacher tip
Note who says AI "thinks" for gentle correction later. "Somewhere in between" is the most interesting answer to unpick.
-
Starts at 1:15
Walk through the four rows: what a language model does, against what understanding would need.
Teacher tip
This slide is the whole starter. If you only have two minutes, show the question and then this.
-
Starts at 2:00
Read the three figures: nearly everyone uses it, and half already know it makes things up.
Teacher tip
A hallucination is not a rare glitch. It is what prediction does when the pattern runs out.
-
Starts at 2:45
Go through the four answer cards, matching them to what pairs said.
Teacher tip
Do not skip the last card: explaining the limits should not tip into "AI is rubbish".
-
Starts at 3:45
Leave the two definitions up for pupils to copy.
-
Starts at 4:05
Go through the three habits and invite pupils to name one they will change.
Teacher tip
Emphasise "Verify". If you are short of time, end here.
-
Starts at 4:35
Read the takeaway, then stop.
-
Debate it (optional β now or next lesson): read the motion and take a first vote β for, against or not sure yet.
Teacher tip
Challenge card: What if someone followed an AI's health advice and it was wrong?
-
Split the class for and against. Each side builds its case from the three points, then hear two speakers a side and vote again.
Teacher tip
Ask who changed their mind, and what changed it.
-
Starts at 4:50
Say, out loud, who pupils can go to if anything today affected them, and leave the support slides up as the room moves on.
Teacher tip
Name the actual person β tutor, head of year or safeguarding lead β rather than leaving it to the slide.
Big question
When you ask ChatGPT a question, do you think it 'understands' you the way a human would?
Adapting the lesson
Support
For pupils who need more help
Provide a simplified definition sheet. Pair weaker students with stronger partners.
Stretch
For pupils ready to go further
Ask students to find a real example of an AI hallucination and explain why it happened.
SEND
For additional needs
Pre-teach key vocabulary before the lesson. Provide visual aids for abstract concepts.
In every lesson
Debate it
For the National AI Conversation
Turn it into a motion: reword the big question as "This house believesβ¦". Split the class for and against, give sides five minutes to prepare, hear two speakers each, then take a free vote.
National AI Conversation
Set up the debate
The same topic, pitched for each age group. Pick your pupils' pathway: the motion, a prompt to get them going, and three points each way to start their preparation.
Primary Β· Years 5 and 6
The motion
Would you trust an AI helper more than a book?
Sentence starter: I would trust ___ more because ___.
For
- AI answers quickly, and you can ask it anything.
- It can explain things in a way that suits you.
- Books can be old and out of date.
Against
- AI can make up answers that sound true.
- Books are checked by people before they are printed.
- AI does not really understand what it is saying.
- In class
- Pupils move to a corner of the room or vote with a show of hands, then tell a talk partner why.
- As a full debate
- Shorter speeches, with more time for questions from the floor and the final vote.
Secondary Β· Years 7 to 11
The motion
It does not matter whether AI understands, as long as its answers are useful.
Challenge card: What if someone followed an AI's health advice and it was wrong?
For
- We use calculators and maps without them understanding anything.
- What matters is whether an answer is correct, and we can check that.
- AI already helps people learn, write and solve problems every day.
Against
- Without understanding, AI cannot tell when its own answer is false.
- Hallucinations in health, legal or safety advice can cause real harm.
- People trust AI more when they think it understands, so the difference changes how we use it.
- In class
- Small groups with talking roles, Builder, Challenger and Summariser, then the challenge card tests their view. Shorter for Key Stage 3.
- As a full debate
- Three-minute opening cases, two-minute rebuttals, and questions from the floor.
Post-16 Β· Sixth form and college
The motion
This house would ban AI chatbots from giving medical or legal advice.
The tension: access vs accuracy. Research: what do UK health and legal regulators say about AI tools giving advice to the public?
For
- Pattern prediction is not professional judgement, and confident errors here can do serious harm.
- Doctors and lawyers are accountable and regulated; a chatbot is neither.
- Vulnerable people are the most likely to rely on free advice without checking it.
Against
- Many people cannot afford or quickly reach a doctor or lawyer, and general information helps them.
- A ban would push people towards worse sources, such as anonymous forums.
- Clear signposting and safety rules would reduce harm without removing a useful service.
- In class
- A formal debate, with points of information offered during the opening speeches.
- As a full debate
- Longer opening cases, and a show of hands or a digital vote at the end.
Key words
- Large Language Model (LLM)
AI trained on billions of texts to predict and generate human-like language. Works by predicting the most likely next word based on patterns, not by understanding meaning.
- Hallucination
When AI confidently generates false or made-up information. Happens because AI predicts 'likely' text, not verified facts.
Take it further
- Next lesson Research one real-world AI hallucination and present it in 2 minutes next lesson