The Map You Stopped Building

You've been to a city several times — Paris, maybe, or Chicago, or wherever you took that work trip two years in a row. You navigated it confidently each time, blue dot on a screen, turn-by-turn in your earbuds. Now someone asks you to draw a rough map, or describe the walk between two places you visited on different days. And you find you can't. The city never assembled itself into anything coherent in your head.

This is not a memory failure. It's a consequence of how GPS works.

A 2020 study from McGill University found that people with greater lifetime GPS experience have measurably worse spatial memory during self-guided navigation — not just in unfamiliar places, but in cities they'd visited repeatedly. Crucially, those who used GPS more didn't do so because they had a bad sense of direction to begin with. The GPS use came first. The decline followed.

Eleanor Maguire had documented the other side of this twenty years earlier. London taxi drivers must memorize something called "the Knowledge" before they can get a cab license: 25,000 streets within a six-mile radius of Charing Cross, 20,000 landmarks, 320 established routes. The training takes two to four years. MRI scans of qualified cabbies show significantly larger posterior hippocampi than people of similar age and education who don't drive taxis — and the volume correlates directly with years of experience.

The cab drivers didn't just navigate better. Something about the navigation changed their brains.

Lookup and map

The difference between the GPS user and the London cabbie isn't how much they traveled. It's the cognitive mode each one operated in.

GPS is lookup navigation. Turn here. In 400 meters, turn left. Each instruction is correct, atomic, and requires no model of the surrounding space. You get where you're going without ever assembling the city into a shape you can hold in your head. Efficient. No residue.

"The Knowledge" forces spatial modeling. You're not following a sequence of instructions — you're navigating a structure you've internalized. When a passenger says "Paddington to the Barbican," you don't look it up. You think across a map you've built, notice the alternatives, pick the route.

These are two genuinely different ways of finding your way. One leaves a map behind. The other doesn't.

Your note-taking app is a GPS

The canonical use of any modern knowledge tool — Notion, Bear, Evernote, Obsidian — goes like this: you have a thought, you write it down, you move on. Later, when you need it, you search. You type some keywords. You find it, or you don't.

This is lookup. The structure you add — folders, tags, notebooks — is metadata designed to make search marginally more precise. It doesn't create a navigable space. You can't browse from a note about a decision you made to the people it affected or the project it was part of, unless you remembered to add explicit links. There's no map. There's a very large pile with a search engine on top.

That works fine until the pile gets large enough that you're missing things. You recreate a decision you already made. You forget a relevant conversation because you can't find it through the words you're currently reaching for. You stop browsing because browsing stopped working.

The GPS problem, but for your knowledge.

What a map requires

A spatial map has properties a pile doesn't: location, adjacency, and — crucially — visible empty space. When your memory palace has a room for "decisions about hiring," you can walk into it and see what's there. You can also see what's absent. The empty chair tells you something.

Most knowledge tools have no empty chairs. They can't. A blob of untyped text has no schema, so there's no basis for incompleteness. A note either exists or it doesn't.

Typed entities change this. A person record with fields for last contact, current role, and relationship context has a shape. When the last-contact field is empty, you see the gap — you can query for it. A decision record that asks for "alternatives considered" creates a slot for the thing you almost did instead. The schema defines structure; the structure creates something navigable.

This isn't a subtle point about retrieval efficiency. It's a different model of what a knowledge base is for.

When you have fifty person records, you can browse them the way you'd walk through a neighborhood — notice who's there, notice who's missing, let adjacency surface things you wouldn't have searched for. When you have a hundred decision records, the ones with empty outcome fields aren't just incomplete. They're open loops, visible as such, waiting.

Not about memorizing

None of this means you should be walking through your knowledge base manually, or that the point is richer browsing. You still want search. The AI reading your notes will use retrieval, not walking.

But the map matters for a different reason: it reflects the quality of what was captured in the first place. GPS users learn cities worse because the lookup mode removes the cognitive work of modeling. Knowledge bases built around search-and-dump accumulate notes that were never really processed — captured and filed, never integrated into anything with edges and adjacency and meaning.

Typed entities impose the minimum cognitive work. Choosing what kind of thing this is — a person, a decision, a preference, a project — is the moment you decide where in the map something belongs. That's a small act of structuring. But it's the act that makes the map.

The London cabbies with larger hippocampi didn't just find their way faster. They understood the city differently. That understanding persisted for years after individual routes had faded — not a lookup cache but a navigable model they could reason from.

A knowledge base with structure works the same way. Not because you'll wander through it. But because the structure you built to make it navigable is also the structure that makes it useful to anyone who reads it — including, eventually, you.

---

Asgeir Albretsen is the founder of Harbor.