Session 39: What Is AI? Machines That Learn
At a glance: pupils learn that AI finds patterns in examples instead of following written rules, spot AI they already use, train a robot with good and bad examples, and explain why AI makes mistakes and can be unfair. · Time: 90 minutes (adjust the timings if your period is shorter) · Workbook pages: 6
Learning outcomes
By the end of the session, pupils can:
- Explain in their own words that AI learns patterns from examples (training data), unlike a normal program that follows rules people wrote.
- Give everyday examples of AI that recognises, predicts and creates.
- Describe how training data affects AI, and explain "garbage in, garbage out" and bias with an example.
Before the session
- Test the Teach the Robot lab on the projector PC and a pupil PC. It needs no internet and no AI account; keep the offline pack on USB.
- Bring 4–6 real plantains (or pictures) at different stages of ripeness for the opening, if you can.
- Have a smartphone ready to show predictive text on the projector (or just hold it up) for pupils with no phone at home.
- Draw the rules vs examples diagram (Figure 39.0) on the board.
Running order
| Time | Activity | Workbook | What the mentor does |
|---|---|---|---|
| 0–10 min | Typing warm-up | – | Pupils open the typing trainer (Typing in the menu, or the warm-up banner on the session page) and do the 10-minute warm-up. Walk round checking posture and home-row fingers. |
| 10–19 min | What do you already think AI is? | Page 1 | Collect ideas on the board without correcting them yet ("robots", "ChatGPT", "it will take jobs"). Then read the plain-talk definition together. |
| 19–28 min | Rules vs examples | Page 1 | Hold up the plantains. "Which rule would tell a computer this black one is ripe?" Link to the IF blocks from Sessions 36–37. |
| 28–37 min | Three jobs and AI around you | Page 2 | Pupils tick the table in pairs and add R, P or C. Show predictive text on a phone. |
| 37–45 min | Collect, train, test and the plantain puzzle | Page 3 | Pupils are "the machine": they solve the puzzle, then explain the pattern they used. |
| 45–57 min | Teach the Robot lab | Page 5 | Demonstrate Mission 1 on the projector, then pairs do all three missions and fill in the table. |
| 57–70 min | Garbage in, garbage out and bias | Page 4 | Use the pairs' Round 3 results as the hook. Tell Joy Buolamwini's story. Discuss the handwriting question. |
| 70–78 min | Spot AI at home | Page 5 | Explain the home task; do it now on your phone for anyone without one at home. |
| 78–86 min | True or false and review questions | Page 6 | Take a show of hands for each statement. |
| 86–90 min | Self-assessment and sign-off | Page 6 | Sign off using the evidence below. |
Key ideas and how to explain them
- AI is a pattern-guesser: a computer program that learned patterns from a huge number of examples, so it can guess about something new. It is not alive, does not feel, and has no opinions.
- Rules vs examples: in Scratch the pupils wrote every rule. With AI, people give examples and the computer finds the pattern. The ripe plantain is the anchor: nobody learned ripeness from a rule book.
- Recognise, predict, create: face unlock and spam filters recognise; phone keyboards and video suggestions predict; chatbots and picture makers create ("generative AI").
- Training data: the labelled examples. Collect → train → test. More, varied and correct examples usually mean better guesses.
- Garbage in, garbage out: wrong labels, too few examples or one-sided data teach the wrong pattern.
- Bias: one-sided data can make AI unfair to some groups of people. People must stay in charge of important decisions.
The 3D lab
Teach the Robot shows learning from examples happening. On the projector, do Mission 1 (teach only 2 examples, then press Let the robot sort) and ask "Why did it get those wrong?" Pairs then do Mission 2 (at least 2 examples of every kind, then 4 out of 5 right) and Mission 3 (one wrong example on purpose, then catch the mistake it learned). The memory shelf shows the robot's training data, and when it is wrong it tells you which example it thought looked most similar. That is a real (simple) machine-learning method, so you can honestly say "this is how some real AI works". The table on Page 5 records the three rounds.
Common mistakes and quick fixes
| Pupils often… | Try this |
|---|---|
| Think AI is a robot or is alive | "Is your phone keyboard alive?" AI is mostly invisible software, and it only guesses from patterns. |
| Think AI "knows" the right answer | Point to Round 3 of the lab: it repeated the wrong example you gave it without noticing. |
| Say "AI = ChatGPT" | Go back to the Page 2 table: most AI they meet is not a chatbot. |
| Teach the robot 10 examples of one kind only | Good! Ask them to test it, then ask why it fails on the other kinds: one-sided data. |
| Mix up Predict and Create | Predict guesses what comes next; create makes something new, like a whole paragraph or picture. |
When things go wrong
- No internet: nothing today needs the internet. The lab runs from the offline pack.
- Not enough PCs: threes, one pupil per mission at the mouse, others record the scores. Or run the lab on the projector, with the class shouting the labels.
- A PC freezes or the power goes: play the robot unplugged. One pupil is "the robot" and leaves the room; the class sorts 4 labelled objects (a bottle, a sachet, a newspaper, a can) on a desk. The "robot" returns and sorts 3 new objects by finding the most similar labelled one. Then swap one label to show garbage in, garbage out.
Answer key
Where have you seen AI? (Page 2): keyboard next word: P. Face unlock: R. "Videos you might like": P. Google Translate: C (it writes a new sentence; accept P with a good reason, since it predicts the most likely translation). Spam folder: R (it recognises scam messages). Chatbot answer: C. Any reasonable Ghanaian idea is fine, for example spotting crop diseases from a phone photo, reading X-rays, predicting rain for farmers, translating local languages, or catching MoMo fraud.
Plantain puzzle (Page 3): Mystery 1: Ripe. Mystery 2: Not ready. Mystery 3: Ripe. The pattern: soft (or yellow, brown or black skin) means ripe; hard and green means not ready. The best answers notice that softness is the most reliable clue, because colour alone fails for the black plantain if you only saw yellow ones.
Think about it (Page 4): it might mark pupils with messy, small or unusual handwriting (or left-handed pupils) as wrong even when their answers are right. Fix: train it on handwriting from many different pupils, and have a teacher check its marks.
Lab table (Page 5): scores vary. Round 1 usually scores lowest, Round 2 usually highest, and Round 3 shows the robot repeating the wrong label on items that look like the one taught wrongly.
Spot AI at home: any three suggested words.
True or False (Page 6)
- AI is alive and thinks and feels like a person: False.
- AI learns patterns from examples called training data: True.
- If an AI sounds very sure, the answer must be correct: False.
- An AI trained mostly on one group can be unfair to other groups: True.
- The keyboard suggesting the next word is AI predicting: True.
Written Review Questions
- A normal program follows exact rules a person wrote (IF this THEN that). An AI learns its own pattern from many examples and then guesses about new things.
- If you train an AI on bad data (wrong, too little or one-sided), it learns a bad pattern and gives bad answers. Example: a bottle labelled "paper" makes the robot put bottles in the paper bin.
Support and stretch
- Needs more support: use real objects for every idea. Let them be "the robot" in the unplugged game before the 3D lab, and pair them with a confident reader for Page 4.
- Finished early: in the lab, try to make the robot score 5 out of 5 with as few examples as possible, then explain which examples mattered most. Or list three jobs in their community where AI could help, and one where a person must always decide.
Evidence for sign-off
The pupil explains rules vs examples in their own words, solves the plantain puzzle, completes all three Teach the Robot missions with the table filled in, and gets at least 4 of the 5 true-or-false statements right. They can explain "garbage in, garbage out" with an example from the lab.
Safety and wellbeing
Some pupils may have heard frightening things about AI ("robots will take over", "AI will take all the jobs"). Take worries seriously and keep the tone calm and practical: AI is a tool made and controlled by people, and learning about it is how they stay in charge. The home task only uses a parent's phone to look at keyboard suggestions: pupils must not send messages or open other apps. Real AI chatbots come in Session 40, and only with supervision.