Collaborative reasoning

Reason jointly with an AI that is required to disagree when warranted, so "collaboration" doesn't collapse into the model doing the thinking while you nod.

When to use this: When you want to go deeper
Learning impact: Clear help, modest gain
Learning evidence: Credible with caveats
Where you are with the topic: Building understanding
How often to use it: Worth repeating often

When to use it

You want a thinking partner, not an answer machine.
A hard problem needs moves proposed and challenged in turns.
Past “collaboration” ended with the model doing the solution while you nodded.

What to ask AI

Type something like: “Collaborative reasoning on: [problem]. My current thinking: [ ]. If you disagree, you must say so and why.”

Require that you author the next claim after each challenge.

Forbid a full solution dump.

How it helps

Collaboration collapses when the partner does the thinking and you only approve.

Collaborative reasoning keeps joint work, with a mandate to disagree when warranted.

You stay the author of the claim; the AI proposes moves and challenges.

How to run it

State the problem and your current thinking.
Invite proposed moves, with mandatory disagreement when warranted.
After each challenge, you write the next claim or step.
Ban solution dumps; restart if one appears.
End by restarting the solution alone from memory.

Example exchange

You

Does this study design support causal claim X? My thinking: maybe, because they have a control group.

I disagree that a control group alone is enough, assignment wasn’t randomised in your summary. What’s your next move?

You

Then I need to check assignment and confounds before claiming cause.

Propose the next check as a claim, not a question to me.

Copyable prompt

Collaborative reasoning on: [PROBLEM]. My current thinking: [ ].
Rules: you may propose moves, but if you disagree you must say so and why;
I author every next claim; no full solution dumps.

The Tell

If the transcript is brilliant and you cannot restart the solution alone, collaboration became spectatorship.

If the AI never disagrees, the mandate failed, tighten the prompt.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

The underlying learning idea is rated moderate. Collaborative reasoning with mandated disagreement resembles productive peer discourse: joint problem-solving plus critique. Benefits depend on the learner still authoring claims.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

AI partners often over-agree or over-take. Your authorship rule is load-bearing. Delivery is speculative.