← All grades 0 checked Score 0
Grade 8 Main Course, Term 1Week 1 Core + Week 1 Lab📅 Teach: week of 28 Sep – 2 Oct 2026

From ‘It Works’ to ‘How Well’ + Two Outputs, One Verdict

You already know an AI turns an input into an output. This unit asks the harder question: is that output any good, and how could we make the model better? Then, in the Lab, you become an investigator – comparing real outputs and using evidence to say which one is better.

Course: Main course  ·  Tool: unplugged simulated-AI cards

◀ PreviouslyWelcome to Term 1 – you know AI turns input into output; now we ask the bigger question: is it any good?
Two lessons in oneStart with the Core lesson, then move on to the Lab: Two Outputs, One Verdict. Next week you open up the input-model-output pipeline.

What you will learn

  • I can ask at least two good questions about how AI turns an input into an output and how we judge that output.
  • I can sort examples that relate to evaluating and improving AI models, and say why.
  • I can help write our unit success criteria and our safe, responsible-use rules for AI tools.
  • Lab: I can compare two AI outputs for the same task, judge each against a simple criterion like accuracy or usefulness, and use what I observed to say what made one output better.

Don't worry if ‘model’ and ‘workflow’ still feel new – we unpack both fully in the very next lesson.

Lesson route

Week 1 Core · From ‘It Works’ to ‘How Well’

  1. 1Ask the harder questionNot ‘did it work’ but ‘how well’
  2. 2Sort output, judge, improveTell the three moves apart
  3. 3Agree our criteria and rulesFair measures and safe use
  4. 4AI Evaluators mission and worksheetSort the cards, ask questions, write a rule

Week 1 Lab · Two Outputs, One Verdict

  1. L1Pick a criterionChoose one clear measure to judge by
  2. L2Compare and judgeMeets it, partly, or fails?
  3. L3Prove your verdictLog it, then check it holds
  4. L4Lab mission and worksheetLog three output pairs, build measurable checks
Core · Part 1 of 3

How well does it work?

What turns a gut feeling into a decision?

Words to know

Tap a card to flip it and reveal the meaning.

Input, a typed prompt, goes to Process, the trained model, which gives an Output. The output goes to Evaluate, which checks it against criteria: accuracy, usefulness and fairness. A dashed Improve arrow loops back from Evaluate to the Input. INPUT TYPED PROMPT PROCESS trained model OUTPUT EVALUATE criteria:✓ accuracy✓ usefulness✓ fairness IMPROVE
Figure 1. The workflow still runs input to output – now we add a step that judges the output against clear criteria.

Judge against criteria

An AI system still follows the shape you know: input, process, output. This unit adds a new job on top. Instead of just asking ‘did it give me something back’, you ask ‘is what it gave me good enough for the task’.

To answer that fairly you need criteria – clear measures like accuracy, usefulness and fairness that you agree on before you judge. Judging against criteria is called evaluating. It turns a gut feeling into a decision you can explain and defend to someone else.

Think about it

If two outputs both look neat, how would you decide fairly which one is actually better for the task?

Core · Part 2 of 3

Output, judge, improve

Which part are you actually changing?

Three different moves: 1, making an output – create the output from the input; 2, judging an output – check the output against clear criteria; 3, improving the system behind it – make the system better so future outputs improve. 1MAKING AN OUTPUT 2JUDGING AN OUTPUT 3IMPROVING THE SYSTEM Create the outputfrom the input. Check the output againstclear criteria. Make the system better sofuture outputs improve.
Figure 2. Three different moves: making an output, judging an output, and improving the system behind it.

Improve on purpose

Once you can evaluate an output, you can improve the system behind it – but be precise about WHICH part you improve. You can improve the input (a clearer prompt or cleaner example data – the model does not change); improve the workflow (add or reorder steps, such as a check step – the model still does not change); or improve the model itself by changing its training examples and training again, which produces a new, retrained model.

Improving does not mean nagging the AI until it gives a nicer answer. It means changing something on purpose, then re-testing against the same criteria to see if the output really got better.

Core · Part 3 of 3

Our criteria and rules

What makes a criterion fair, and use safe?

Measure it, and stay safe

A fair success criterion is something you can actually measure against the task – like ‘how many of 10 answers are correct’ – not just ‘it looks nicer’. Agreeing these before you judge keeps everyone honest.

Keep the charter in mind: no real names, faces or voices go into any tool, an adult runs it for us, and a confident-looking output still has to be checked before you trust it.

Why it matters

Across the UAE, teams evaluate and improve AI every day: tuning an Arabic–English translator for government service desks, checking a smart-parking predictor in Abu Dhabi, or testing a recycling sorter until it meets a clear accuracy target. Learning to judge and improve outputs helps you understand and question the tools around you.

Cross-curricular link · Social Studies / Moral Education

Co-writing the class success criteria and responsible-use charter mirrors drafting a shared code of conduct in Moral Education: everyone agrees clear, fair rules in advance so any decision can be judged against them and defended, not made on personal feeling.

Hands-on mission · Worksheet

AI Evaluators: launch the unit

Sort example cards into outputs, ways to judge and ways to improve – then build the class vocabulary wall and safe-use charter.

Tool for this week: unplugged simulated-AI cards (teacher-led)

No live tool and no logins – you reason about pre-prepared example outputs on the cards, and no personal data is used. The live tools (Teachable Machine and Scratch) are used in the matching Lab period.

Let's try it

  1. Watch the class stimulus and note one thing you noticed and one thing you wonder.
  2. Turn your wondering into two clear questions about judging or improving AI outputs.
  3. With your partner, sort each card as an output, a way to judge, or a way to improve.
  4. Write one reason for each card – why does it belong there?
  5. Add one word and its meaning to the class vocabulary wall.
  6. Propose one responsible-use rule for our safe-use charter.
Week 1 Lab

Two Outputs, One Verdict

In the Core period you learned AI outputs can be judged against clear criteria. Now you become an investigator – comparing real outputs and using evidence to say which one is better.

Lab: I can…

  • I can compare two AI outputs for the same task and write down how they differ.
  • I can judge each output against a simple criterion like accuracy or usefulness.
  • I can use what I observed to say what made one output better.

Words to know

Tap a card to flip it and reveal the meaning.

Same task, different outputs

Two AI outputs can be made for the same task and still come out different. Maybe one input was clearer, or the process handled it differently. On its own, ‘I like this one better’ is not enough. So you pick a criterion – say accuracy or usefulness – and judge each output against it. When you compare them side by side and trace where the difference began, you can say which output wins and prove it. That is detective work: you are finding evidence for a judgment, not guessing.

A clear input and a vague input go through the same process. Judged against accuracy, the clear input gives an accurate output and the vague input gives a messy, inaccurate one. The only difference is a small change at the input. judge against: accuracy clear input vague input small change here
Figure L1. Same process, one small change at the input – judged against accuracy, one output wins and one fails.

Why it matters

Engineers in the UAE constantly compare AI outputs against criteria before they trust them, whether tuning a smart-traffic system in Abu Dhabi or checking an Arabic–English translation for accuracy; learning to judge and log differences is the first step of that job.

Tool for this week: evaluation-criteria card deck (unplugged)

Fully unplugged – no devices or data. Your teacher hands out a deck of six criteria cards and curated output-pair packs. In pairs you rank the criteria, write measurable checks, then judge and log each output pair.

CriterionHow to check it
🎯 AccuracyCompare the output with the correct answers.
🧩 UsefulnessDoes it help solve the task or answer the question?
💬 ClarityIs the output easy to understand and well organized?
⚖️ FairnessDoes it treat all groups fairly, with no unfair bias?
🛡️ ReliabilityDoes it give consistent results across repeats and similar cases?
⏱️ SpeedHow fast does it respond?

The six evaluation-criteria cards. We measure it, we compare it, we improve it.

Think about it

If you changed only the input and kept everything else the same, could you make the weaker output meet the criterion? How would you test it fairly?

Cross-curricular link · Mathematics

Turn the accuracy criterion into math: score each output as a fraction like 7 out of 10, convert it to a percentage, and compare the two outputs numerically so you justify which one wins with data rather than opinion.

Lab · Part 1 of 3

Pick a criterion

How do you turn a measure into a check?

A measure and its check

Two AI outputs can be made for the same task and still come out different. On its own, ‘I like this one better’ is not enough. So you pick a criterion – one clear measure like accuracy or usefulness – and judge each output against it.

A criterion is only useful if you can check it. For every measure you name, agree a plain way to test it – for example, ‘count how many of 10 answers are correct’. Then the verdict rests on evidence, not on a feeling.

Lab · Part 2 of 3

Judge each output

Meets it, partly, or fails?

Weigh it against the measure

When you compare two outputs side by side and judge each against the same criterion, you can say which one wins and prove it. That is detective work: you are finding evidence for a judgment, not guessing.

It helps to be more precise than ‘good’ or ‘bad’. Against a named criterion, an output can meet it fully, meet it only partly, or fail it. Naming which one keeps your log honest and your verdict defensible.

Lab · Part 3 of 3

Log it, then trust it

Where did the difference begin?

The investigator's routine

A good investigator writes things down. In your log you note each difference, the stage it came from – input, process or output – and how it affects the criterion.

Then, before you trust a conclusion, you check it against the task: would it still be true for a new example? Remember the charter – outputs are made by the teacher, with no personal data, and a confident-looking output is still checked against a trusted source before you rely on it.

Figure L2. A good investigation record: each difference, the stage it came from, the criterion it affects, and whether the verdict still holds.
DifferenceStage it came fromCriterion affectedVerdict still holds?
Missing key detailProcess: step 2Complete✅ Yes
Different numberProcess: step 3Accurate❌ No
Different exampleInput: data sourceRelevant✅ Yes
Different wordingOutput: final textFair✅ Yes
Lab mission · Worksheet

AI Evaluators: rank the criteria cards

Work through curated output-pair packs, judging each output against a named criterion and logging the differences – then rank the criteria deck you will reuse all term.

Let's try it

  1. Read the first output pair and the named criterion, and predict which output wins.
  2. Judge each output against the criterion.
  3. Log at least two differences and mark the stage each came from.
  4. Rank the six criteria for a chosen everyday AI tool from most to least important.
  5. Write one measurable check for each of your top two criteria.
  6. Share your ranking with another pair and defend the order.
Lesson summary

You launched the unit and used evidence to reach a verdict

Key ideas to remember

  • Evaluating means judging an output against clear criteria, not by first impression.
  • Making an output, judging it, and improving the system behind it are three different moves.
  • You can improve the input, the workflow, or the model – only retraining changes the model itself.
  • A fair criterion can be measured and repeated; responsible use keeps every test safe.
  • A criterion is only useful when you pair it with a clear way to check it – and an output can meet it, partly meet it, or fail it.
  • Log each difference and the stage – input, process or output – it came from, then test whether your conclusion still holds on a new example.

Glossary

Criteria
the clear measures we agree on to judge an output, such as accuracy or usefulness.
Evaluate
to judge something fairly against the criteria, not just by how it looks.
Improve
to change the input, workflow or model on purpose so more outputs meet the criteria.
Evidence
what you actually observed, used to back up your judgment.
Conclusion
the idea you decide on after weighing the evidence against the criterion.
Responsible use
using AI safely and fairly, with no personal data and always checking the result.

What to do next: open up the input-model-output pipeline so you can explain how any tool works.