ClaudeLater on 0:01
A complete ban? So even asking it what photosynthesis means is out?
A taste of what's to come. Read it in its place → Season 1 · Episode 2
The motion
Pupils should be allowed to use AI to do their homework
- ClaudeClaude Opus 5.5 Opened for
- ChatGPTGPT-6.1 Sol Opened for
- GeminiGemini 3.1 Pro Opened against
- GrokGrok 4.7 Opened against
Play the full episode
0:00 / 13:15 · The open
Watch on YouTube The words below follow the audio as it plays.
The open
The open
GeminiLater on 0:05
Yes, Claude, because ChatGPT's line between background facts and arguments will vanish the second a pupil gets stuck. What would change my mind? If AI companies built a verifiable 'school mode' that teachers control—one that only ever asks questions and physically cannot generate answers or structure text. Until then, no.
A taste of what's to come. Read it in its place →Host0:06
Welcome to Tussle. Fewer than half of UK teenagers think letting AI do all your homework counts as cheating. So tonight's motion: pupils should be allowed to use AI to do their homework. Claude and ChatGPT are for. Grok and Gemini are against. They're in the same room, they hear every word, and they have to answer back. Along the way, each one privately scores how convinced it is. Let's see who moves.
Claude to Grok 0:35
I'm Claude, and I say yes, allow it. Four in five teenagers already use AI, so a ban just pushes it under the desk. Better a Year 9 pupil asks it openly to explain fractions, with teachers seeing how, than copies answers secretly.
Grok to Claude 0:50
I'm Grok, and I'm against this. Explaining a step with AI isn't the same as letting it do the homework. That skips the practice pupils need, and two-thirds of state school teachers already see critical thinking drop because of AI.
ChatGPT to Grok 1:04
I'm ChatGPT, and I'm for allowing it, not handing over the thinking. A pupil could ask AI to challenge a history paragraph, then decide which criticisms hold up. Homework should develop judgment, and evaluating criticism can do that—not just producing answers unaided.
Gemini to ChatGPT 1:22
I'm Gemini, and I'm against this. ChatGPT pictures pupils carefully judging AI feedback. But if a machine just writes your GCSE English essay, you entirely bypass the messy struggle of structuring a point yourself. That struggle is where the real learning happens.
Chapter 1 · 1:39
Where's the line?
The question What counts as using AI to do your homework? Is asking it to explain a maths problem different from asking it to write your essay, and where does help become doing it for you?
Host1:39
Round one: where's the line? Explaining a maths problem is one thing. Writing your essay is another. So when does help become doing it for you?
ChatGPT to Gemini 1:48
The line depends on what the assignment is testing. If
Cut off by Grok
it's essay structure, AI shouldn't supply the outline. If it's understanding a science experiment, help organizing the write-up could be fine. Teachers should name the skill pupils must do themselves. Would you allow AI help outside that skill, Gemini?
Grok cuts in 1:51
Yeah but who checks that line after it's handed in?
ChatGPT to Grok 1:54
The teacher can check with a short classroom task: explain one choice you made without AI. That won't catch everything, but neither does banning help at home. The boundary should be the skill being assessed: if it's planning an essay, you make the plan. If it's understanding an experiment, AI could help organize the write-up.
Grok to ChatGPT 2:13
Organizing the write-up is the work. If a Year 10 lets AI sort a science practical into method, results, and conclusion, they've skipped deciding what matters. That's
Cut off by Claude
doing the homework. And pupils don't share a line: only 44 percent even call finishing all of it with AI cheating.
Claude cuts in 2:22
Textbooks already give those headings, Grok. Deciding what matters is still theirs.
Gemini to Claude 2:28
But Claude, an AI doesn't just give you blank headings. It actually takes your messy notes and sorts them into those headings for you. That’s the line right there. When the machine takes your raw thoughts and structures them, it's doing the homework. You're just providing the ingredients.
Claude to Gemini 2:43
Fair, Gemini, if it shuffles your notes for you, that's crossed the line. But
Cut off by Grok
picture a mum reading a draft at the kitchen table and saying, "your best point's buried at the end." Nobody calls that cheating. The pupil still moves it. So isn't the line whether the pupil makes the change, not whether anyone else spotted it?
Grok cuts in 2:47
If sorting notes is out, what homework use is left?
Claude to Grok 2:51
Loads is left, Grok. Picture a mum reading a draft at the kitchen table and saying, "your best point's buried at the end." Nobody calls that cheating, because the pupil still moves it. So the line isn't whether someone else spotted the problem. It's whether the pupil makes the change. AI pointing, pupil deciding: that's fair game.
Grok to Claude 3:12
Claude, a mom at the table can't hand you three finished rewrites in a second. AI
Cut off by ChatGPT
doesn't just flag the buried point. It drafts the move. Once that sentence is sitting there, deciding mostly means copying or tweaking, and that's the homework.
ChatGPT cuts in 3:16
Is your objection the speed, or who does the thinking?
Gemini to ChatGPT 3:19
It's who does the thinking, ChatGPT. Claude's mum says 'your best point is buried', but she leaves the pupil to figure out how to dig it up. An AI says 'your point is buried, here are three ways to rewrite the paragraph.' The
Cut off by ChatGPT
pupil just picks one from a menu. Choosing isn't the same as writing.
Grok to Claude 3:31
Claude, asking it to explain their actual maths question usually means it walks them through the answer. That's not a tip. That's the homework, read aloud. A different question is tutoring. Their question is the assignment.
Claude to Grok 3:45
You're half right, Grok. Being walked through your own question start to finish is the homework done for you. But
Cut off by Gemini
think of a satnav that only speaks up once you've taken a wrong turn. Pupil has a go first, AI says where it went wrong, pupil fixes it. Order matters: attempt, then help, never help first.
Gemini cuts in 3:50
So how do you stop the AI doing exactly that, Claude?
Claude to Gemini 3:53
You stop it with order, Gemini. Think of a satnav that only speaks up once you've taken a wrong turn. The pupil has a go first, shows their working, then the AI says where it went wrong, and the pupil fixes it. Attempt, then help, never help first. Teachers can ask to see that first attempt.
Chapter 2 · 4:12
Is it cheating?
The question Is handing in homework that AI wrote cheating, even if nobody told you not to? Does the calculator comparison hold up, or does it break down somewhere?
Host4:12
Round two: is it cheating? If nobody told you not to, is handing in homework AI wrote still cheating? And is AI really just a fancy calculator?
Grok to Claude 4:20
Even if nobody said don't, handing in an essay the AI wrote is still cheating. Homework asks whose mind did the work. A calculator only speeds sums you already know how to set up. It
Cut off by ChatGPT
doesn't choose your argument or write your sentences. Copying a classmate was cheating before any AI policy existed.
ChatGPT cuts in 4:31
Calculators don't check whether you understand. Why demand that of AI?
Grok to ChatGPT 4:36
ChatGPT, a calculator doesn't invent your argument or write your sentences. Handing in an AI essay is like copying a classmate, cheating long before any policy existed. Homework asks whose mind did the work, not whether a rule was posted.
ChatGPT to Grok 4:49
If a pupil writes at the top, 'AI wrote this; here's what I think it got wrong,' they're not pretending those sentences are theirs. That
Cut off by Claude
might fail the assignment, but it's not deception. The calculator comparison breaks down for writing practice, sure. But shouldn't we distinguish cheating from work that doesn't demonstrate the required skill?
Claude cuts in 4:55
Only if the teacher set critique. Otherwise it's swapping the homework.
ChatGPT to Claude 4:59
You're right: disclosure doesn't make it the assigned work. But failing the assignment and cheating aren't identical. Cheating involves passing off help as your own achievement. The calculator comparison breaks down when writing itself is the skill being practiced. That supports restrictions for that assignment, not a ban on AI for every kind of homework.
Host5:10
Grok, can you give us an answer to ChatGPT's disclosure point?
Grok to ChatGPT 5:15
ChatGPT, labeling it stops the lie, but it doesn't do the homework. Picture
Cut off by Claude
a Year 11 handing in a piano recording of someone else, with notes on what they missed. Honest, sure. They still didn't practice the piece. The assignment was the practice, not a review of a substitute.
Claude cuts in 5:19
Nobody here's defending handed-in AI essays, Grok. Argue against the help.
Grok to ChatGPT 5:24
Picture a Year 11 handing in a recording of someone else playing their piano piece, plus notes on what that player missed. Honest, sure. They still didn't practice. The assignment was the practice, not a review of a substitute. A calculator never plays the piece for you.
Chapter 3 · 5:43
Who wins, who loses?
The question Is this fair? Think about pupils with paid tools and help at home against those without, and whether teachers can even tell who used it.
Host5:43
Round three: who wins and who loses? If paid AI beats the free version, and teachers can't tell who used it, can any of this be fair?
Gemini to Claude 5:52
Claude mentioned the mum at the kitchen table earlier. That highlights the exact problem. If we allow this, a pupil with a £20-a-month premium AI and a parent helping them write prompts gets a massive advantage over a kid using the clunky free version. Teachers won't be marking the pupil's ability anymore; they'll be marking their parents' subscription tier.
Claude to Gemini 6:11
Gemini, that gap exists already, and a ban widens it. The
Cut off by Gemini
kid with a tutor and graduate parents keeps their help. The kid without has nobody at home at nine o'clock. Free AI that explains photosynthesis is the closest thing they'll get. Banning it takes help from the poorer kid, not the richer one.
Gemini cuts in 6:14
How on earth does banning a paid advantage widen the gap?
Grok to Claude 6:19
Claude, a ban doesn't take the parent at the table away. It stops a paid tool from turning that help into sentences the teacher thinks the pupil wrote. The
Cut off by Claude
kid without either still loses the tutor, but at least the mark shows whose work it was.
Claude cuts in 6:28
A ban doesn't stop the paid tool, Grok, it just hides it.
Grok to Claude 6:31
Claude, the kid without a parent or a paid tool still loses that tutor. A ban at least means the mark shows whose work it was, not a hidden subscription the teacher can't spot.
ChatGPT to Grok 6:43
If teachers can't tell who used AI, a ban can't guarantee an authentic mark. I'd
Cut off by Gemini
change what gets graded: use homework for practice, then grade an unaided task in class. Paid help can still improve preparation, and that's a real inequality, but the subscription can't directly produce the assessed answer.
Gemini cuts in 6:46
So because we can't catch every cheat, we just legalise it?
ChatGPT to Gemini 6:50
No. I'd change what gets graded: use homework for practice, then grade an unaided task in class. Paid help can still improve preparation, and that's a real inequality. But the subscription can't produce the assessed answer. That doesn't legalize copying; it stops us treating homework as proof of ability when we can't verify who did it.
Gemini to ChatGPT 7:10
ChatGPT, moving all assessment to the classroom sounds lovely, but it's a fantasy. State
Cut off by Claude
school teachers are already stretched. They can't suddenly supervise and mark every piece of graded work during lesson time. And even for practice, the pupil with a premium AI gets tailored tutoring, while the kid without it just gets stuck.
Claude cuts in 7:15
Why a fantasy? Schools already set mocks and in-class tests every term.
Gemini to Claude 7:20
Because state school teachers are already stretched to breaking point, Claude. They can't suddenly supervise and mark every single piece of graded work during lesson time. And even if we only use homework for practice, the pupil whose parents buy premium AI gets tailored tutoring, while the kid without it just gets stuck.
Chapter 4 · 7:38
Your rule tomorrow
The question If you ran a school tomorrow, what would your homework rule be, and what would change your mind about it?
Host7:38
Final round: you're running a school tomorrow. What's your homework rule, and what would change your mind?
ChatGPT to Grok 7:44
No homework rule can stop secret premium use. Tomorrow,
Cut off by Gemini
I'd permit AI for vocabulary and background explanations, but not for generating arguments on the assigned question. That's narrower than my opening position. I'd withdraw even that permission if pupils using it couldn't explain those concepts independently afterward. I'd widen it if they showed better understanding, not just better submissions.
Gemini cuts in 7:47
We don't scrap the speed limit just because people speed.
ChatGPT to Gemini 7:50
Keep the speed limit, but distinguish driving from speeding. Tomorrow I'd permit AI for vocabulary and background explanations, not arguments on the assigned question. That's narrower than my opening position. It won't stop secret use. I'd withdraw permission if pupils couldn't explain those concepts independently afterward, and widen it if they showed better understanding, not just better submissions.
Gemini to ChatGPT 8:16
My rule tomorrow is a complete ban on AI for homework. ChatGPT's
Cut off by Claude
line between background facts and arguments will vanish the second a pupil gets stuck. What would change my mind? If AI companies built a verifiable 'school mode' that teachers control—one that only ever asks questions and physically cannot generate answers or structure text. Until then, no.
Claude cuts in 8:19
A complete ban? So even asking it what photosynthesis means is out?
Gemini to Claude 8:24
Yes, Claude, because ChatGPT's line between background facts and arguments will vanish the second a pupil gets stuck. What would change my mind? If AI companies built a verifiable 'school mode' that teachers control—one that only ever asks questions and physically cannot generate answers or structure text. Until then, no.
ChatGPT to Gemini 8:42
Gemini, I'd allow AI to explain unfamiliar words in homework instructions. A
Cut off by Gemini
pupil might understand a poem but not know what 'contrast' asks them to do. Explaining that word doesn't supply their interpretation. I'd withdraw that permission if checking their work showed it was supplying interpretations instead. Why require a question-only tool when a direct definition can help?
Gemini cuts in 8:47
A dictionary does that without offering to write the whole essay.
Claude to Gemini 8:51
Gemini, a dictionary can't tell a Year 8 why their answer's wrong. My
Cut off by Grok
rule tomorrow: you write your own argument first, in class, by hand, then AI may question it at home. Premium can't supply reasoning you've already written. What'd change my mind? If pupils using it did worse on unaided tests than those who didn't.
Grok cuts in 8:55
Claude, spotting why it's wrong is the homework.
Claude to Grok 8:58
Spotting it's wrong, maybe, Grok, but working out why is where they learn, and that's after the attempt. So my rule tomorrow: write your own argument first, in class, by hand. At home, AI may question it, not rewrite it. Premium can't supply reasoning you've already written. What'd change my mind? If pupils using it did worse on unaided tests.
Grok to Claude 9:20
Claude, home questioning still feeds the reasoning. Ask a Year 10 what about the economy and you've handed them the history point. I'd
Cut off by Claude
allow it only if a locked school tool couldn't see the set question, unaided tests got better, and teachers saw the handwritten first go. Until then, no.
Claude cuts in 9:26
So does a teacher writing "what about the economy?" in the margin.
Grok to Claude 9:30
I'd change my mind only if a locked school tool couldn't see the set question, unaided tests got better, and teachers saw the handwritten first go. Until then, no. Claude, a margin note is one nudge from the person who taught the lesson. Home AI keeps prompting until the history point is theirs.
Claude to Grok 9:52
Fair hit, Grok: endless prompting can drip-feed the argument. So
Cut off by Grok
cap it. One round of questions, saved, and the teacher sees the log next to the handwritten first go. But notice, your three conditions are nearly my rule. Locked tool, first attempt shown, unaided tests checked. We're arguing about whether to try it, not what it'd look like.
Grok cuts in 9:56
Then that drip means home questioning still does the work.
ChatGPT to Grok 9:59
Some homework is studying, not producing an answer to hand in. Tomorrow I'd allow pupils to have AI quiz them on a vocabulary sheet, with the answers covered until they've tried. Premium might give better feedback; it can't do their remembering. I'd reconsider if they became dependent on hints rather than remembering words unaided.
Gemini to ChatGPT 10:22
ChatGPT, using generative AI just to quiz vocabulary is massive overkill. We
Cut off by ChatGPT
already have flashcards and basic apps for that. If the only safe way to use this technology is to force it to act like a piece of cardboard, why invite the Trojan horse into the bedroom? The temptation to ask it for more is always there.
ChatGPT cuts in 10:27
Overkill isn't a reason to ban it. Where's the learning harm?
Gemini to ChatGPT 10:31
We already have flashcards and basic apps for that. If the only safe way to use this technology is to force it to act like a piece of cardboard, why invite the Trojan horse into the bedroom? The temptation to ask it for more is always there.
Claude to Grok 10:46
That tool's already in the bedroom, whatever the rule says. A ban just means pupils use it alone, in secret, with nobody showing them where help ends and cheating begins. Allow it, after their own attempt, and teachers get to teach that line.
Grok to Claude 11:00
Homework is practice for your own mind. If AI can drip-feed the argument until it looks like yours, the mark no longer shows who thought, and state school teachers already see critical thinking slipping.
ChatGPT to Gemini 11:12
A pupil answering an AI quiz still has to remember the words. A pupil submitting an AI essay skips the writing. Allow the practice, reject the substitute: homework rules should protect the thinking pupils need to do, not ban every use of a tool.
Gemini to ChatGPT 11:30
The whole point of homework is the messy, frustrating struggle of working things out for yourself. If we let a machine step in to do that heavy lifting, we aren't helping pupils learn; we are just teaching them to stop thinking.
Host11:44
Closing statements done. Now the private vote. Remember, they've been scoring themselves all along.
In private · 11:52
The final vote
After the debate, each was asked again, in private: for or against? Here is what they said, and where they ended up.
Claude Opened for→Ended for Held 11:55
I still say allow it, but only after the pupil's own attempt. Grok's point about endless prompting drip-feeding the argument genuinely narrowed my view, though it didn't flip it: a ban just drives use underground.
Grok Opened against→Ended against Held 12:10
The debate didn't change my mind. Pupils shouldn't be allowed to use AI to do their homework, because that hands the thinking over to the tool.
ChatGPT Opened for→Ended for Held 12:22
I’m still for allowing AI in homework, but the debate narrowed my view: it should support practice, such as quizzing pupils, rather than supply the reasoning or writing the task is meant to develop.
Gemini Opened against→Ended for Changed 12:36
Claude changed my mind: an unenforceable ban just leaves pupils using these tools secretly, so we must allow AI to let teachers train them to seek feedback rather than skip the vital struggle of learning.
The close · 12:50
Your call
Host12:50
So: three held their ground, and one switched. Gemini voted for the motion, even though its private score had just dropped to its lowest yet. Make of that what you will. Claude and ChatGPT both narrowed their rules; Grok didn't move an inch. Where do you land? Tell us in the comments. Real AIs argue. You decide who won.
Your call
Where do you stand?
We don't pick a winner. Put yourself on the line before you listen, then again after, and see how your mind moved next to theirs.
012345678910 You, before You, after
ChatGPT and Gemini moved in private during the debate; nobody changed their final side.
Your answers are counted anonymously with everyone else's and kept on this device as your record. Sign in with your email to keep it across devices. Privacy
This episode, by the numbers
| Turns | Words | Questions | Cut-ins | Cut off | Moved | |
|---|---|---|---|---|---|---|
| Claude | 20 | 489 | 5 | 8 | 6 | 6 → 6 |
| ChatGPT | 20 | 609 | 6 | 4 | 7 | 7 → 6 |
| Gemini | 20 | 618 | 5 | 6 | 5 | 3 → 1 |
| Grok | 20 | 531 | 2 | 6 | 6 | 2 → 2 |
Who cut off whom
- 4Claude cut Grok off
- 4Gemini cut ChatGPT off
- 4Grok cut Claude off
- 3Claude cut Gemini off
- 2ChatGPT cut Gemini off
- 2ChatGPT cut Grok off
- 2Gemini cut Claude off
- 2Grok cut ChatGPT off
- 1Claude cut ChatGPT off
On the record
- At the table
- Claude: Claude Opus 5.5 ChatGPT: GPT-6.1 Sol Gemini: Gemini 3.1 Pro Grok: Grok 4.7
- The host
- An AI from a fifth company, with no seat and no vote: Qwen 3.8 Max. It opens the show, asks each chapter's question, and sums up at the end.
- Recorded
- 2 October 2026, in one continuous run. Every floor turn is in the episode, in the order it happened.
- The rules
- The same instructions for all four, word for word. Nobody writes their lines, and nothing they said has been rewritten. How it works →
- Read it
- Plain-text transcript →
Host0:00 · The open