Week 6: What are you allowed to use?
Follow a fictional cyclist across a city using turn-by-turn navigation, and test whether needing a resource ever tells us, by itself, that someone does not understand.
29 March 2027
Where Week 5 left us#
Week 5 built a seven-rung ladder, but Recall and Reconstruction assume that removing a source makes the evidence stronger. This week tests that assumption by varying permitted resources. We move from the lock to a fictional courier; the examination question continues even though the setting changes.
Does needing support show that you do not understand?
This is a self-guided lesson. Both these notes and the slide deck contain the explanations you need; no presenter is assumed. Allow about 30 minutes, with another 15–20 minutes if you complete the investigation. Use paper or a text document for your answers. No navigation app, bicycle or AI account is required. The people and incidents below are fictional.
What you will learn#
By the end, you should be able to explain why removing a resource is not automatically a fairer test, name a concrete set of conditions under which an external resource functions as part of a person’s own reasoning rather than merely assisting it, and design a task that tests engagement with a resource instead of memory of one specific trip.
1. A courier and a phone#
Elena works as a bicycle courier. Her phone gives turn-by-turn directions to each delivery address. Today’s delivery is three suburbs from her usual route, somewhere she has never ridden before. She follows the directions, arrives on time, and hands over the parcel.
Did Elena demonstrate that she understands the area she just rode through? The question is not whether she followed the directions correctly — she did — but what following them correctly is evidence of. An examiner watching only the successful delivery cannot yet tell whether Elena engaged with the route at all, or simply executed instructions the way a machine would.
2. The intuitive move, and why it is not enough#
The obvious next step is to take the phone away and ask Elena to do it again. If she can, the reasoning goes, she really knew the route; if she cannot, she was only following the tool. Week 5’s ladder makes judgements like this systematic by naming stages a performance can reach — but it carries a hidden assumption worth testing directly: that performing without a resource is automatically stronger evidence than performing with one.
That assumption has a cost. Suppose Elena is handed a blank map immediately after the delivery and asked to draw the route she just rode. She might manage a rough sketch — not because she understands the area, but because a single recent trip sits briefly in short-term memory regardless of whether she understood anything about it. Conversely, she might fail to reproduce it, not because she lacks navigational competence, but because nobody remembers a one-off route from a single ride. Either outcome would be read as evidence about her navigation, when it is really evidence about recall of one episode. Removing support can reveal dependence, but it can just as easily test memory instead of the ability we actually meant to examine.
3. A resource that thinks with you, not just for you#
Clark and Chalmers’ The Extended Mind argues that an external resource is not automatically outside the boundary of someone’s own thinking. Their example is Otto, a man with memory loss who carries a notebook and looks up addresses in it the way another person would recall them from memory. They ask: if Otto’s notebook plays exactly the functional role that biological memory plays for someone else, on what principled basis do we say Otto’s belief is not really his?
Their answer is not “any resource counts.” In §4 they give four conditions under which an external resource is doing the same job as a memory rather than merely being consulted like a book: the resource is a constant presence the person turns to whenever it is relevant, rather than an occasional aid; the information is available without difficulty, not buried behind search or negotiation; retrieved information is automatically endorsed, taken as reliable without a separate act of checking each time; and it was placed there by the person’s own past endorsement — it is there because they put it there, not because someone else’s judgement sits inside it. They treat the fourth condition as more arguable than the first three, but hold that all four matter.
Clark & Chalmers (1998), §4, "From Cognition to Mind." A theoretical argument, not an empirical finding — it changes what question we should ask about a resource, not what any single case proves.
4. Applying the test to Elena’s phone#
Run Elena’s phone against the same four conditions. Is it a constant presence — does she check it for most unfamiliar routes, or only sometimes? Is it available without difficulty — instant, or does she have to stop, unlock the phone and search? Does she automatically endorse what it says, the way you would trust your own sense of a familiar street, or does she treat every instruction as a claim to be checked? And did she place the information there herself — did she choose this app and these settings — or is she being handed someone else’s judgement with no history of her own endorsement behind it?
For a courier who works the app all shift, every day, checking it instantly and trusting it without a second thought because she chose and configured it herself, the first three conditions are clearly met. That does not settle whether Elena “really” understands the city. It reframes the question: instead of asking whether the phone disqualifies her performance, we ask whether her use of the phone is the kind of use that counts as part of her own competence, or the kind that does not.
5. Pause and write: sort three cases#
Allow four minutes. For each case below, decide whether the resource looks more like Otto’s notebook (integrated, counts toward the person’s own competence) or more like a stranger’s advice consulted once (external, doesn’t). Give one sentence of reasoning for each, using the four conditions.
- A tourist checks a paper map for the first time, at an intersection, having never seen it before this trip.
- A pharmacist looks up an unfamiliar drug interaction in a reference she consults dozens of times a shift, trusts without re-verifying, and chose to keep on her desk.
- A student pastes an essay question into an AI tool they opened for the first time five minutes ago and copies out whatever it returns.
Compare your answer with a worked response
1. The tourist and the paper map fails constancy and past endorsement: this is a first use, not a habitual one, and the map was handed to them rather than assembled from their own prior choices. This looks like consulting a stranger, not extending one’s own memory.
2. The pharmacist and her reference meets all three of the strong conditions: constant, instantly available, automatically trusted because years of use have earned that trust, and it is a resource she selected and maintains. This is closer to Otto’s notebook — though it says nothing about whether she could diagnose the interaction unaided, which is a separate question a different task would have to ask.
3. The student and the AI tool fails constancy and past endorsement in the same way as the tourist, and adds a further problem: the output was not checked before being copied out, so even automatic endorsement is doing the wrong job here — trusting an unfamiliar source without verification, not trusting a well-tested one.
6. What the sorting exercise does and does not show#
Sorting these three cases does not tell us that the pharmacist understands pharmacology, that the tourist cannot read a map, or that the student understands nothing about the essay topic. It tells us which of the three resources is doing a job similar to memory, and which is closer to consulting an outside party for the first time. That distinction changes what a supported performance is evidence of — it does not replace the need to examine the performance itself.
This is why Week 5’s assumption needed testing rather than simply accepting it: “no resource” and “resource present” are not two clean categories with understanding on one side. A habitual, self-chosen, trusted resource can function as part of someone’s own reasoning; a fresh, unfamiliar one, even if technically permitted, mostly is not.
7. A better test than removing the phone#
If taking Elena’s phone away risks testing recall of one trip instead of navigational competence, what would test the thing we actually want to know?
Change the content of the support rather than its presence. Suppose the phone instructs Elena to turn right onto a street that a visible sign marks as one-way, entering against traffic. Does she follow the instruction anyway, or does she notice the conflict and choose a different way to the same address?
This does not remove the resource. It tests whether Elena is engaging with the route as she rides — reading the street the phone is describing — or treating the instructions as commands to execute without reference to anything in front of her. A courier who stops and reroutes has shown something about her own judgement that a successful, uneventful delivery could not show, whether or not she ever consults the phone again after this shift.
8. Pause and write: design the better test#
Allow five minutes. Pick a task you know something about that ordinarily involves a resource — a recipe app used while cooking a familiar dish, a calculator used during a familiar kind of problem, a reference chord chart used while playing a familiar song. Design one change to the content of the resource, not its presence or absence, that would let you tell whether someone is engaging with the task or just executing instructions. State what a response to your change would demonstrate, and what it still would not.
Worked response: a comparable example
A calculator used for a familiar kind of problem, e.g. finding a percentage discount, is a constant, instantly available, automatically trusted, self-chosen tool for most adults — it passes the four conditions and its presence alone tells us little. A content change that tests engagement: ask for the discount on a price already including tax, when the discount only applies pre-tax. Blind entry of the sticker price produces a plausible-looking wrong answer; a person tracking what the numbers represent notices the mismatch before reaching for the keys. This demonstrates whether the person is reasoning about what the calculation represents, not merely whether they can operate a calculator. It does not demonstrate that they could solve an unfamiliar kind of percentage problem with no worked structure at all — that would need a different, harder change.
9. Keep the record honest#
As in Week 1, record the condition rather than treating a supported success as disqualified or an unsupported failure as proof of ignorance:
| Field | Example |
|---|---|
| Resource | What was available, and how habitual is it for this person? |
| Fit against the four conditions | Constant? Available without difficulty? Automatically endorsed? Self-placed? |
| What was actually tested | Presence/absence of the resource, or engagement with its content? |
| Limit | What would still require a different task to examine? |
A record that only says “used a phone” or “used no phone” is not useful on its own. A record that says which of the four conditions the resource met, and whether the task tested content or presence, is inspectable in the way this course keeps asking for.
10. Apply the lesson yourself#
Use the Week 6 investigation to design and record your own content-change test, and to keep a comparison between a presence/absence version and a content version of the same task.
Before you finish, explain these distinctions in your own words:
- Why an unsupported performance is not automatically stronger evidence than a supported one.
- The difference between a resource that is a constant, trusted, self-chosen part of someone’s routine and one consulted for the first time under examination.
- The difference between testing whether a resource is present and testing whether someone is engaging with what it says.
Reading with a purpose#
Primary argument: Clark, A., and Chalmers, D. (1998). “The Extended Mind.” Analysis, 58(1), 7–19. Read the authors’ text, §4, “From Cognition to Mind,” and §5, “Beyond the Outer Limits.” Ask: which of the four conditions is doing the most work in distinguishing Otto’s notebook from a resource you would not want to count? We use this paper because it gives an explicit, arguable test for when a resource is part of someone’s own thinking, rather than leaving “using a resource” as a single undifferentiated category.
What this leaves unresolved#
Sorting resources into “integrated” and “not yet integrated” tells us how to read a supported performance; it does not, by itself, tell us whether the underlying reasoning is any good. Next week keeps the resource question open and asks a different one: if someone can present the same idea in an unfamiliar form — a new audience, a new representation, a task they were not drilled on — is that better evidence than either a supported or an unsupported repetition of the same rehearsed version.