Module 5 · About 25 minutes

AI in the classroom

For four modules we used artificial intelligence on the teacher's desk. Now the student joins the table. You will see an AI that answers next to an AI that asks, mark where learning happens inside a real student dialogue, split one learning objective into three levels, find the gap in a set of class results and build a 40-minute lesson minute by minute.

You can also complete this module by listening

The audio summary of this module was produced with the Madlen Podcast tool.

0:00 0:00

The pedagogical basis

Learning lives in the effort, not in the output.

When AI enters the classroom the biggest risk is not technical but pedagogical: handing the effort from the student to the model. Three established frameworks tell you where that risk sits.

The ICAP framework

Chi and Wylie, 2014. Learning gains rise in order: passive listening, active annotating, constructive explaining, interactive discussing. Whichever rung the AI puts the student on is how much learning you get.

Bloom's two sigma problem

Bloom, 1984. A student taught one to one moves markedly above the class average. The problem was always scale: nobody could reach thirty students one at a time.

The zone of proximal development

Vygotsky. Learning happens in the band between what a student can do alone and what they can do with support. Good AI use provides that support; bad AI use takes the whole job.

This module's criterion

Whose effort?

One question is enough to judge any activity: who is doing the cognitive work? If the student explains, compares, finds the error and builds a justification, learning is happening. If the model explains and the student copies, there is a product but no learning. The rest of this module turns that criterion into practice.

Interactive · Side by side

Same topic, two different AIs.

Same student, same assignment, same model. The only difference is in the prompt. Click the tabs and watch where each dialogue goes.

The student's prompt

Write a paragraph explaining the difference between photosynthesis and respiration.

Model: Photosynthesis is the process in which plants use sunlight to convert carbon dioxide and water into glucose and oxygen. Respiration is the process in which glucose is broken down with oxygen to release energy. Photosynthesis stores energy, respiration spends it.

Result: There is a flawless paragraph on the screen and nothing in the student's head. On the ICAP scale this is the passive rung: the text did not pass through the student, it passed by them. The assignment gets submitted, the topic does not get learned.

The prompt the teacher set up

You are a science teaching assistant working with a Year 8 student. We are studying photosynthesis and respiration. Your rule: never give me the answer directly. Ask exactly one question at a time, wait for my answer, find what is missing or wrong in it, and ask a new question that makes me think about that part. When I say "I understand, let me summarise", listen to my summary and correct only the parts that are wrong.

Model: What might a plant be doing at night, in the dark? Do you think it produces energy, or spends it?
Student: Nothing at night, kind of like it is asleep.
Model: If the plant keeps growing overnight, where does the energy for that growth come from?
Student: Hmm. From what it stored during the day?
Model: Spending what it stored has a name. Do both of these processes happen in the plant, or does one only happen in daylight?

Result: Same model, same amount of time. But the student forms a guess, meets their own error and repairs the concept in their own words. On the ICAP scale this is the interactive rung. The difference is not in the knowledge, it is in one rule inside the prompt: "never give me the answer directly."

That single line is the most valuable technical thing you can teach your students. If a student learns to write that rule in their own account, they have turned an assignment machine into a tutor.

Classroom-ready patterns

Five patterns, five ready prompts.

You met the idea behind these patterns in the task map in Module 3. These are the classroom-ready versions, complete with the rule, the boundary and the exit ticket. Copy them and adapt them to your own objective.

1 Socratic tutor · individual work ›

When? Revision, the run-up to a test, or whenever you want a misconception to surface. Ten to fifteen minutes, followed by sharing in class.

You are a teaching assistant working with a [year group] student on [topic]. Never give me the answer directly; ask exactly one question per turn. Find the gap in my answer and ask a question that makes me think about it. Every five turns, ask me to summarise what we have covered so far. If I drift off topic, bring me back gently.

Exit ticket: Before closing the chat, the student writes a three-sentence summary in their exercise book by hand. What you assess is not the chat, it is those three sentences.

2 Debate opponent · one screen, whole class ›

When? Whenever you are working on argument building around a contested claim. Split the class in two, let the model defend the other side, and have students produce replies.

I am running a debate with my [year group] students on "[claim]". You argue the opposing side. Give one short argument at a time, then wait for the students' reply. Do not be aggressive, do not use logical fallacies, and explicitly acknowledge the strong point in a student's argument. If they give a convincing reply, say "you are right on that point" and concede it.

Exit ticket: Each group writes the model's strongest argument and their own best reply to it on a single slide.

3 Error hunt · group work ›

When? Whenever you want students to experience hallucination from Module 1 first hand and to move from consumer to auditor. One of the best-liked activities, because catching the model out feels like a game.

Write a 200-word text about [topic] at [year group] level. Deliberately plant three factual errors in it: one date or figure error, one conceptual error, and one plausible-looking but fabricated source. Do not tell me where the errors are, just give me the text.

Exit ticket: Groups find the errors and write one line for each on how they verified it. The verification method matters more than the error found.

4 Peer assessment rehearsal · individual ›

When? Before students assess each other's work, so they learn to use the rubric. The model supplies a neutral text to rehearse on, which avoids anyone's real work being the practice material.

Write a mid-quality example [text type] at [year group] level. Deliberately make some aspects strong and others weak. After giving me the text, make no comment at all. I will mark it against a rubric, then show you my marking and ask what I overlooked.

Exit ticket: The student's own marking and their justification. The model's score is never the standard, only material for discussion.

5 Language practice partner · individual or in pairs ›

When? Speaking and writing practice in language lessons. The model is a patient, non-judging partner with unlimited repetitions, which is especially valuable for students who are self-conscious.

You are doing conversation practice with a [year group] student learning [language]. My level is [level]. Use only vocabulary and structures suitable for that level. After each of my replies, first continue the conversation, then correct at most two of my mistakes briefly and show the correct form. Do not overwhelm me and do not launch into long grammar explanations.

Exit ticket: The student notes three of the corrections they received and uses those forms correctly in the next lesson.

Interactive · Learning audit

Did learning happen in this dialogue?

Below is a shortened version of a real conversation, six turns long. For each student turn, decide: did the student think here, or hand the work to the model? Then check.

Student

Can you write the causes of the Industrial Revolution, I need it for homework.

Handing over. The whole task goes to the model and the student contributes nothing. On the ICAP scale this turn sits on the passive rung.

Model

Three main causes are usually listed: agricultural change freeing up labour, access to coal and capital, and expanding trade networks.

Student

I do not get the third one. Why would trade networks make machines appear? Trade existed long before that.

Learning. The student is not satisfied with receiving an answer; they are interrogating the causal link. That is the mark of the constructive rung: they are trying to build their own meaning.

Model

Because a wider market makes producing at scale worth the investment, and scale is what machines are for.

Student

So you are saying trade did not cause machines, it made them worth paying for. Have I got that right?

Learning, and the most valuable kind. The student restates the idea in their own words and asks for confirmation. This is exactly the interactive rung; as a teacher, this is the sentence you want to hear.

Model

Yes, your summary is right. If you like, we can add the role of capital availability to that picture.

Student

No need, just turn this into 200 words in essay format. And add some sources.

Handing over, and doubly risky. The writing passes to the model and the requested sources will most likely be fabricated. The rule from Module 4: every source is verified one by one.

Model

Here is a 200-word text in essay format, with a bibliography...

Student

I searched the library catalogue for the first book in that bibliography and it does not exist. I am cutting it and citing the chapter in our textbook instead.

Learning. The student is auditing and correcting the output. This behaviour is worth more than any single right answer, because the skill transfers to every subject and to life after school.

Running this exercise with your students is a good lesson in itself. Take an anonymous extract from their own conversations, put it on the board and ask the same question: who did the thinking here?

Interactive · Differentiation

One objective, three different students.

Here is today's answer to the scale problem Bloom pointed at: preparing the same objective at three levels now costs one prompt. Fill in the fields and generate the prompt for your own class.

Year group

Subject

Main need

Output type

Generated prompt

Select all four fields and your prompt appears here.

Notice that no student's name appears anywhere in the prompt. The rule from Module 4 still holds: describe the need, not the person.

Interactive · Class data

Where is the gap?

Differentiation starts with seeing who needs what. Below are the anonymised results of an end-of-unit quiz: five students, four objectives, scored out of four. Click a cell to see the suggested action. If a whole column drifts into the red, it has stopped being an individual problem and become a class one.

1 · not yet secure 2 · partially 3 · largely 4 · secure

Start by clicking a cell. The red marker in the column headers shows where the class average is lowest.

Interactive · Lesson flow

How do you build the 40 minutes?

A lesson with AI in it does not work by opening a screen and saying "off you go". Choose your setup and your goal, and a minute-by-minute flow appears. Click the coloured sections to read what happens in each stage.

Setup

Goal

0 min10203040 min

The risk

Convenience can be the natural enemy of learning.

The most important warning in this module comes at the end: AI is extremely good at reducing cognitive effort. For a teacher that is a gain; for a student it is usually a loss. Click the cards to flip them.

Myth

"If students work with AI they learn faster."

Click to flip

Fact

Work finished quickly is not a topic learned.

The assignment gets done sooner, but to the extent no cognitive effort was spent, retention falls. The measure is not time taken, it is whether the student can explain the topic in their own words.

Myth

"Students already know AI better than we do."

Click to flip

Fact

Familiarity with a tool is not skill with it.

Students pick up the interface fast, but they do not know how to write a good prompt, audit an output or recognise where to stop. All three are taught, and you are the one who teaches them.

Rule 1

Do not hand the hard part to the model. Finding the idea, writing the first draft and building the argument stay with the student; formatting, translating and proofreading can go to the model.

Rule 2

Every AI activity closes with a non-AI exit ticket. Three sentences written by hand are worth more than everything left on the screen.

Rule 3

Teach students to interrogate the model. "Are you sure, what is the source, where might this be wrong" outlasts any answer the model gives.

Bonus · the admin load

Four ready prompts for the work on your desk.

Cutting the load outside the classroom grows the time inside it. This section is not the main subject of the module, but it is the part that saves the most time. Use no real names or personal data in any of them.

1 Parent information letter ›

Write an information letter to [year group] parents about [topic]. Keep the tone respectful and plain, use no blaming language, stay under 180 words, and end with two bullet points on what the parent should do.

2 Department meeting agenda and minutes ›

Turn the scattered notes below into department meeting minutes. Headings: decisions taken, owners, dates, items carried to the next meeting. Add nothing that is not in the notes and mark anything missing as "not stated".

3 Cover lesson plan ›

I will be absent tomorrow. Write a 40-minute plan for [year group] [subject] that a non-specialist cover teacher can deliver with no extra materials. Give minute-by-minute instructions and list separately what should go on the board.

4 Feedback drafts, at scale ›

Using the rubric below, produce three feedback drafts: one for work that fully meets the objective, one that partially meets it and one that does not. Each should contain two sentences of praise and two concrete next steps. I will adapt them to each student's actual work.

Remember the ranking from Module 4: a feedback draft can be written by the model, the grade and the judgement come from a human.

Check yourself

Fill in the blanks.

In the ICAP framework the rung with the highest learning gain is the one. The critical rule in a Socratic tutor prompt is . If a whole column of an objective is low in your class data, the problem is now . Every AI activity should close with a non-AI .

A student rewrites a paragraph they received from AI in their own words and asks "is this what it means?". Which rung does this behaviour match?

Restating in your own words is constructive; opening that restatement to correction is interactive. This is exactly the behaviour you want to see in a student: not finishing the assignment, but testing their understanding.

Almost there.

For the certificate you need to engage with at least three activities in each module. No exam, no score.

Try it in Madlen All modules

Your progress is stored in this browser.

AI in the classroom

Learning lives in the effort, not in the output.

Same topic, two different AIs.

Five patterns, five ready prompts.

Did learning happen in this dialogue?

One objective, three different students.

Where is the gap?

How do you build the 40 minutes?

Convenience can be the natural enemy of learning.

Four ready prompts for the work on your desk.

Fill in the blanks.

Almost there.

Subscribe to Our Mailing List!