Models Are Maps: What A Model Can And Cannot Do is a guide for readers who want clarity instead of slogans. The purpose is simple: learn how to tell what a claim means, what would count as support, and what would count as a real correction.
The purpose of this page is to help you use models without being used by them. If you want the technical map of formal material on this site, the Research Library is the clean index. Here we stay in plain language and focus on a single habit: keep the boundary between a tool and reality visible.
People often say “the model says” as a shortcut, but the shortcut is costly. It can quietly replace reality with a diagram, and then the debate becomes a contest over diagrams rather than a search for what is true. When that happens, the conversation can grow louder while learning stops.
A model is a chosen simplification. That is not a flaw. It is the point. The question is whether the simplification matches the purpose and whether the limits are admitted. A model becomes dangerous when the limits are hidden and the simplification is treated as if it were the whole story.
Throughout the site you will see a recurring theme: constraints shape what is possible. In the most technical material, that theme is pursued with explicit assumptions and checkable steps. In the more interpretive material, it appears as a habit of honesty about what is being claimed. If you want that human-level reading, Being Human is a good companion.
Key definition: model, map, and target
A model is a representation built for a target. The target might be a physical system, a social pattern, or a mathematical structure. The model captures some features of the target and ignores others. The choice of what to capture is guided by purpose.
A map is the easiest example. A road map ignores the height of every hill, and a topographic map ignores which restaurants are open. Neither map is “wrong” because it leaves things out. Each is judged by whether it helps you do the task it was built for.
The target is what you are trying to understand. The map is your tool. When the map works, it is tempting to stop looking at the target. That temptation is precisely why the boundary must remain visible.
If you keep purpose in view, the question becomes sharper: what does this model help me predict, compare, or explain. If you lose purpose, the model becomes a slogan, and slogans have no natural stopping point.
- Name the purpose before you argue about accuracy.
- Separate “fits what we saw” from “must be the cause.”
- Look for what the model cannot represent by design.
What models are good for
A good model compresses complexity. It lets you see relationships that are hard to hold in your head. It can show you which variables matter more and which details are noise for the task at hand.
A good model also produces stable comparisons. Even if it is not perfect, it can still tell you that option A is likely to have higher cost than option B under a range of conditions. That kind of comparative clarity can be more valuable than fragile precision.
Models can also serve as scaffolding. They can help you generate questions you would not have asked otherwise. The model gives you a vocabulary for exploring the space, even if the vocabulary later needs correction.
Finally, models can guide measurement. If a model says two quantities should be related, you can go measure the relation. Without the model, you might not have known what to measure.
Keeping a model honest: tests, stress, and scope
If a model matters, it should face stress. A gentle test is one where the model is tuned to the data and then praised for matching the data. A stronger test is one where the model commits to a prediction it did not get to choose.
Stress can also be structural. You can vary the input conditions and see whether the conclusion is stable. If a tiny change in assumptions flips the conclusion, that does not mean the model is worthless. It means the model is sensitive, and the sensitivity should be part of the message, not hidden in a footnote.
Another honesty practice is to separate what the model guarantees from what it merely suggests. Some formal work on this site focuses on exactly that separation: stating clear conditions and deriving what follows. If you want an example of that style, Rigidity & Reconstruction is a good destination.
Finally, keep scope explicit. A model can be reliable inside a narrow corridor and misleading outside it. Scope is not embarrassment. Scope is what makes a model usable, because it tells you where the tool is safe to use.
- Ask: what would count as a real counterexample here.
- Ask: which assumptions matter, and which are cosmetic.
- Ask: does the claim survive variation in inputs, or only in one tuned setting.
- Ask: is the conclusion descriptive, explanatory, or normative.
What models are not good for
Models are not good for granting certainty when the situation is unstable. If the system can change regimes, a model fitted in one regime may mislead in another. This happens in weather, in markets, and in human communities.
Models are not good for replacing judgment. A model may recommend an action that is locally optimal under its assumptions, while the real situation includes values, constraints, or risks the model cannot represent.
Models are also not good for deciding what matters. They can help you reach goals, but they do not select the goals. When a model is treated as a value machine, it ends up smuggling values in through hidden choices: what is measured, what is ignored, and what is treated as success.
When a model is used beyond its purpose, the failure can look like reality misbehaving. In fact, it is the tool being used outside its design window.
A concrete example: fitting a curve and mistaking it for a cause
Suppose a simple curve fits a dataset extremely well. It is tempting to say, “That curve explains the process.” But a curve fit is often only a compressed description. It summarizes what was observed without revealing why it happened.
If you treat the curve as a cause, you may make confident predictions outside the observed range and then feel betrayed when the predictions fail. The problem is not that modeling is useless. The problem is that a descriptive tool was treated as a mechanism.
A better approach is to separate tasks. Use the curve where you are interpolating inside the observed range. Use a mechanism model when you need to extrapolate. And when you do extrapolate, state the assumption that makes it reasonable, so that readers know exactly what would break the conclusion.
This is one reason technical writing insists on stating assumptions. A stated assumption is not weakness. It is the boundary that keeps a useful tool from becoming a misleading idol.
A common misread and a clean correction
A common misread is to think, “If models are not reality, then nothing can be known.” That is too extreme. The correction is that models support knowledge when they make risky predictions and those predictions hold up across meaningful variation.
Another misread is to assume that the most detailed model is always the best. Detail can hide ignorance. A simpler model can be better if it captures the stable structure and stays honest about what it ignores.
A final misread is to treat disagreement about models as disagreement about motives. Often people are trying to protect the same value, but they believe different representations are reliable. When you name the boundary and the purpose, disagreement becomes less personal and more productive.
When models become identity
A model becomes identity when criticism of the model is experienced as criticism of the person. That shift can happen even among thoughtful people, especially when the model is tied to a community, a career, or a moral cause.
Once identity is on the line, evidence becomes secondary. People interpret every objection as hostility, and every limitation as sabotage. In that mood, the model stops functioning as a tool for learning and becomes a badge for belonging.
The correction is not cynicism. The correction is to return to purpose. Ask what the model helps you do. Ask what it cannot do. Ask what a reasonable opponent would need to see to take the model seriously. These questions lower the temperature and restore the possibility of shared learning.
When you can hold a model lightly, you can improve it. You can also replace it when the situation changes. That flexibility is a strength, not a betrayal.
Where to go next
- Philosophy of Meaning and Checkable Claims: How to Read Models Without Confusion
- What Does “Checkable” Mean: Claims, Tests, And Limits
- Falsifiability and Testing: What It Clarifies And What It Doesn’t
Helpful next step
Behavioral Science Under Constraints: Decisions, Learning, and Coordination