Route Knowledge and Survey Knowledge

Lost in Las Vegas

I am in Las Vegas for a conference. It's not a city I like or know well, although the first day or so can be fascinating. I'll talk a walk up and down the strip just to get out of the hotel and conference center and there are always changes, big and small, to notice.

On this trip, when I arrived at the airport my ride to the hotel was quite disrupted because the Las Vegas Marathon had closed down the strip. The rideshare app showed my driver a route which knew about that, but then it turned out the police had unexpectedly closed another road. So my driver had to find her own way. (A few years ago, I would have said she was left to her own devices but today that would be confusing!)

My rideshare driver was clueless. She had driven round Vegas many many times, but always with a GPS and she really didn't know the geography. After 5 minutes of going in circles, I walked the rest of way; I could see the hotel and I was traveling light.

A london cab driver looking at a map. A color palette of grays, browns, and light blue neutral colors. A simple and minimalistic design, with flat colors and shading.

Finding your Way

John Huth, in his wonderful book The Lost Art of Finding Your Way makes distinction between route knowledge and survey knowledge.

If we have route knowledge, we know routes and landmarks along these routes. We know of a network where different routes join each other but not what lies between the routes themselves. The quip attributed to the crusty Maine lobster-man, "You can't get there from here," would seem to indicate route knowledge, where no combination of routes connect from "here" to "there."

Survey knowledge, on the other hand, is a complete familiarity with an environment. In your mind you see the region as if you are hovering over the landscape and seeing everything below in miniature.

So, route knowledge is the step-by-step memory of how to get from point A to B, tied to one’s own orientation (i.e., turn left at the red house, then go straight for two blocks).

Survey knowledge, on the other hand, is more map-like, allowing individuals to imagine relationships among locations beyond immediate perception (e.g., seeing how multiple streets connect or hypothesizing shortcuts). We use both forms: route knowledge ensures competence in specific contexts, while survey knowledge promotes flexibility and broader navigational insight.

Route knowledge is local, situated expertise: personal, tacit, but it is often resistant to abstract models. You can see this with friends and relatives (and ourselves!) who always take the same "short cut" and simply cannot imagine another route is better.

Doing The Knowledge

A recent study from the Champalimaud Center for the Unknown looked in detail at how London taxi drivers navigate. It's a fascinating study, because London cabbies are famous for The Knowledge, an arduous and complex exam of route planning around 26000+ streets. And, for 25 years, we have known they have an enlarged posterior hippocampus in the brain as a result of this intense cognition.

One thing I like about this study from Champalimaud is that it proved me totally wrong on a prior assumption. I imagined that doing the knowledge gave London cabbies something like survey knowledge of the city: an abstract, overall view of how it all fits together which they can mentally navigate in advance. But it's not at all like that.

I'll take this more or less directly from the paper: the drivers deploy a rational complexity–based decision precaching strategy where they prioritize high-complexity junctions (e.g., intersections with 5+ branches) and then fill in local street details after establishing critical nodes, with response times influenced by real-world metrics (e.g., longer streets.)

This means they don't try to figure out every single turn in advance. Instead, they first focus on the tricky intersections along a possible route where lots of roads come together: the complicated parts that could cause confusion, like the Old Street roundabout, Hyde Park Corner or the infamous Hammersmith Gyratory. They use something like route knowledege to navigate between these complex junctions and, having gone beyond them, use route knowledge again to reach the destination.

Taxi drivers use a hybrid of route and survey knowledge; which could be a model for designing more effective analytics and AI. Rather than treating experiential (route) and theoretical (survey) knowledge as separate, more powerful systems could integrate both. We could create hybrid cognitive maps that combine metric knowledge (statistical patterns) with procedural understanding (domain-specific processes).

Many a wrong turn leads you right

There’s a lot to think about here.

The taxi drivers also demonstrate "hierarchical detangling" by resolving critical junctions first: in analytics this minimizes computational overhead. Instead of linear processing of an entire route, they break down problems by complexity.

They do this by "non-sequential sampling," that is they access remote environment states during planning. This ability to jump between different points in the problem space enables more flexible, creative solutions. For AI design, imagine an architecture that can attend to distant but relevant information rather than being constrained by purely sequential processing.

I can also see how this approach would help to balance automated systems with active human learning to prevent degradation of institutional knowledge and human-AI interactions.

It's exciting to be wrong. So many more possibilities and landscapes open up, compared to being right, when your mind just trudges along the same old routes. Oh, see what I did there? For my next post ...