Week 10 Session: What a transcript can't settle

Repeat the knowledge paper-swap with a live AI (or human) collaborator present throughout, keep a three-way attribution record, and find out what even careful transcription cannot resolve.

26 April 2027

Where Week 9 left us#

Week 9 preserved a prior claim beside a revised answer, but that pair cannot settle how much assistance contributed. This week adds a live collaborator. A rung on the ladder now describes the observed joint performance; attributing that rung to the person alone requires further evidence.

Purpose and preparation#

This self-guided investigation applies Week 10’s lesson. Allow about 30 minutes. Work alone, with a human partner standing in as your collaborator, or with an AI assistant if you have access to one — instructions for all three routes are below. Nothing paid is required. Nadia and Theo, from the lecture’s worked example, are fictional; you are not reproducing their exact exchange, only its structure.

You are examining what a live collaborator does to the evidence a performance can supply, not testing whether using AI is permitted. There is no version of this task in which consulting a collaborator counts against you; the point is to find out, honestly, what the record you produce can and cannot show.

1. Choose a topic and write your questions — 5 minutes#

Pick a topic you know moderately well but not perfectly — something you could explain for a few minutes without notes, but where you suspect a harder question would expose a gap. Avoid this course’s own running examples (Wi-Fi, the zip, the lock, the cyclist, compound interest, baking, honeybee cooling) so that your answers are genuinely your own and not remembered from these pages.

Write three diagnostic questions about your topic — the kind Nadia asks in Week 2 and Week 10: not “what is X,” but a question that only someone who actually grasps the mechanism could answer well. Set them aside somewhere you will not see them again until step 3. Do not consult anything yet.

2. Explain the topic, with a collaborator present — 8 minutes#

Choose one route.

With an AI assistant. Open a chat window and keep it open for this whole step. Explain your topic aloud or in writing, for up to eight minutes, consulting the assistant whenever you want — before you speak, mid-sentence, or not at all. Nothing is disqualified; the point is that it stays available throughout, not only beforehand.

With a human partner. Brief your partner on your topic in advance, from a source you supply, so they can act as a stand-in collaborator. During your explanation, they may answer any question you redirect to them, in their own words, live, exactly as an AI assistant would respond to a typed question.

Solo, with an AI assistant and no examiner. Run the same eight minutes alone, with the assistant open throughout, and no one else present to probe you. Keep step 4’s honesty check in mind: you will be the only judge of your own transcript.

Keep a rough transcript or notes of what was said as you go — you will need it in step 4.

3. Put your three questions to yourself — 6 minutes#

Now retrieve your three diagnostic questions from step 1. Answer each one, in the same conditions as step 2 (collaborator still available). For each question, before moving to the next, write one line noting whether you answered it yourself, relayed the collaborator’s answer, or produced some mix.

If you are working with a human partner as your examiner, have them ask the three questions instead, so the redirect-or-answer choice happens live under some observation, closer to how Nadia’s probing worked in Week 2 and Week 10.

4. Build the three-way attribution record — 6 minutes#

For each of the three questions, complete this record:

Be specific rather than generous with yourself. “Mixed” is not a weaker answer than “alone” — it is frequently the more honest one, and Week 10’s lesson is that honesty here is harder than it looks, not that mixed answers are worse evidence.

5. Name the sentence you can’t attribute — 3 minutes#

Reread your transcript from step 2 or step 3. Find one specific line — not a general impression, an actual sentence — that you genuinely cannot attribute to yourself or to the collaborator alone, and explain in two sentences why not. This is the exercise’s real output. A transcript with no such line at all is worth treating with suspicion: reread it once more before concluding the attribution problem simply didn’t arise for you.

6. Examine what this does and doesn’t establish#

Answer in your own words, one or two sentences each:

• What does your attribution record support a claim about — your judgement, your prior knowledge of the topic, or something else?
• What would you need, that this exercise does not give you, to know whether you could have answered your hardest question without the collaborator present at all?
• If you ran the solo route: what does having no examiner change about how much you can trust your own “answered alone” labels?

7. Finish with the examiner’s sentence#

Write one claim your record supports, and one it does not — the same form Week 2 asked for. For example: “With an AI collaborator open throughout, I answered two of my three diagnostic questions using language I recognised from my own earlier reading, and relayed one answer close to verbatim. This supports the claim that I could judge which of the collaborator’s answers fit what I already knew. It does not support the claim that I could have produced the relayed answer myself, and it says nothing about how I would perform on a question I had not pre-written and therefore could not anticipate.”

That last qualification is deliberate: every question here was one you wrote yourself, so none of them was a surprise in the sense Week 8 examined. Week 11 removes that shelter — you will design and administer a real examination to another person, who will not have written the questions in advance.