IV. Adjacent Territory Essay 22
The Map Is Not the Territory
Models that omit on purpose, and why you still need the ground
You are going somewhere you have never been. Somebody has given you a map. It is a good map. It has the roads in the right colours, the station marked, the river where the river is, and a scale in the corner so you can tell that the walk is longer than it looks on the page.
And none of the things that will actually happen to you are on it. The way the light comes off the car park at seven in the morning. The smell of the place you are looking for. The fact that the road bends and you cannot see the bend until you are in it. The hour it takes to get there because the bus is every twenty minutes and the timetable says twelve. The map has none of these and you will need all of them.
The map is not a failed version of the territory. It is a different thing, made for a different job, and the omissions are the price of the job. The question is not how to make a perfect map. There is no such map, and if there were it would be useless. The question is how to know which omissions you are being given, and that is a question about your purpose.
Consider the tube map. The lines are not in their real geographical positions. They are straightened, they are set at fixed angles, they are spaced evenly regardless of how far apart the stations actually are, and a station that is a five minute walk from another station in the real world may be drawn as if it were next door.
Every one of these is a deliberate lie, and the lies are in service of one particular thing: making the network legible. A geographically accurate map of a large rail network with dozens of lines crossing is unreadable, and an unreadable map is useless for the one job a map of a rail network has, which is helping a person get from one place to another when they do not know the place.
So the map gets used anyway. People who have never navigated the network in the real city can use the tube map correctly, because the map is a model of the network and not of the surface. The journey it describes is real, and the journey it draws is not the journey you walk. The distance on the page is not the distance on the ground, and the interchange that looks like a single point is often a long walk through four corridors.
Omission is the method
It is worth saying plainly that a model is not defined by what it includes. A model is defined by what it leaves out, because the leaving out is what makes it a model rather than a copy.
Every copy would be the territory. A copy of a city is the city, and it is the size of the city, and you cannot put it in your pocket. A map is small, and being small means it cannot contain everything, and what it does contain is chosen for a reason. The omissions are the edits, and the edits are where the intention is.
So a useful test of any model is to ask what it would have to include in order to be useless. For a map of a rail network, the useless version includes the colour of the walls, the names of the trees, and the weather. For a recipe, the useless version includes every possible measurement of every ingredient in every possible state.
Distortion on purpose
The other thing worth seeing is that a good model's distortions are systematic rather than random. The tube map does not distort the distances by accident, and it does not distort them evenly. It distorts them in a specific direction: it makes the network look more connected than it is, because what the user needs to know is what connects to what, not how far apart things are.
What if the map were perfectly accurate? Down to the metre, showing every path, every tree, every puddle, and the exact position of every person who is standing in the park at that moment. It would be a different kind of useless. It would be too large to use, it would change every second, and it would not answer the question you had, which was how to get from the station to the door. The accuracy would be in the territory and useless in the map, because the question was about the territory only by accident.
Now change the condition. What if the map were the opposite, a diagram with no geography at all: a network of nodes and lines showing only what connects to what? That is the tube map with the last layer of detail removed, and it is more useful for some questions and less for others. The question how do I get there is answered by the network diagram. The question what will it look like when I arrive is not answered by anything you can draw.
"The map and the territory are the same thing, roughly." This is how people end up managing their lives by a spreadsheet. The spreadsheet is a model, with a purpose, and the purpose is not the same as the purpose of the thing being modelled. A forecast is a model of a future, and a future cannot be read off a model without a step, and the step is where the mistakes happen. A model and its subject are related the way a description of a person is related to the person. A good description helps. A description taken as the person is a category, and you have read about categories earlier on this site.
The other misconception: "a model that leaves things out is therefore unreliable." It may be perfectly reliable, for its purpose. A recipe that leaves out the temperature of the room is not unreliable because of it. The mistake is not the omission. The mistake is using a model outside the purpose it was made for, which is using a map to find out what a place feels like.
So every once in a while a team discovers that the thing everybody was arguing about was never in the model. The spreadsheet did not have a row for it. Nobody omitted it on purpose, and everybody assumed somebody had, and the argument had been running for a month. This is what a model with an unmarked scope does. It does not lie. It simply has nothing to say, and silence in a conversation reads as agreement.
When to leave the map
Here is the practical part. A model is trusted inside the boundary of its purpose, and the boundary is not marked on the model, which is why people walk confidently off the edge of one.
The signs that you have left the map are worth learning. The model is being used to predict something it was built to describe. The model is being used to make a decision about a case that the model does not contain. The model's own categories have started to feel like facts about the world. And the most common one: the model is being used to decide what to do, and the answer it gives is comfortable.
And the moment to reach for the territory is when the model is the only thing in the room. Go and look at the actual thing. Stand in the place. Touch the material. Ask the person. The model is a preparation for that, and a preparation that has replaced the thing it was preparing for is a category.
Maps for the mind
It is worth saying that this is not only about paper. Every plan, every model, every description of a person, every theory, every curriculum, is a map. The same structure applies: each one omits most of the territory, each one distorts it in a direction that serves a purpose, and each one is more useful for being a map and more dangerous for being believed in. The one you are most likely to over-believe is the one about yourself, because a personality test or a diagnosis fits you the way the tube map fits the city: well enough to be useful, not at all completely.
Limits
This essay is not an argument against models, and it should not be read as one. Models are among the most powerful things human beings have made, and the ability to make a model that is useful for a purpose nobody had in mind is the core of science, engineering, and craft. The claim is much smaller than that. It is that a model is a tool with a scope, that the scope is invisible from inside the model, and that the most common error in using one is using it outside the scope it was built for.
Second limit: the essay is written from the position of somebody who trusts models. There is a stronger version of the claim, which is that all models are equally distorting and the differences are matters of taste, and that version is not being argued here. Some models are much better than others at their job, and finding out which is a real skill, and this essay does not give you a shortcut to it.
Third limit: knowing what a model leaves out does not tell you what to do instead. Sometimes the right move is to gather better data and rebuild the model. Sometimes it is to abandon the model. The territory does not automatically produce a better model. It produces a better basis for one.
Stuck
You are stuck with a model when the model has become the world. The plan is the project, the chart is the patient, the spreadsheet is the business, and the territory has stopped being consulted. The tell is that you can produce the model at any time and you cannot describe the actual situation without it.
The way out is to describe the situation without the model, in words, for ten minutes. If you cannot, you have lost the territory, and the map has become a self-referential thing that only refers to itself. Go and look at the actual thing and write down what you see, including the parts the model has no place for. The parts the model has no place for are the parts it is hiding, and they are the parts that will explain the failure.
Sources
- The phrase about the map not being the territory is associated with Alfred Korzybski, who wrote about the relationship between maps and territories in the early twentieth century. The broader claim that models simplify is standard throughout science and philosophy, and no specific text is quoted here.
- The description of the underground rail diagram is given in general terms. No date, designer, or specific feature is claimed, and readers wanting the history should look it up directly.
- All other examples are ordinary and hypothetical.