Models Are Not Reality

Why it attracts me

Models are among my favorite tools. A good one lets me try a change in minutes that would take weeks, or be unsafe, to try for real. That power is also the risk. After enough time with a model, it is easy to start defending the model instead of checking the world.

The idea

Every model is selective. It keeps the features that matter for one question and drops the rest. So the useful question about a model is not "is it true?" but "what can it answer, and where does it stop?" A good model puts its assumptions where people can see them.

An example

The Queueing Simulation Lab puts two models side by side. One is a formula that predicts how long customers wait at a service counter. The other plays out every customer one at a time. The lab is open about where each falls short. When arrivals stop, it pauses the comparison because "the formula assumes customers keep arriving." It also warns that in a busy line, even 2,000 simulated hours can miss the true average by 10 percent or more. Both models are useful. Neither is reality.

Where it connects

This is the warning label for Digital Copies for Testing What-Ifs: a detailed virtual copy of a factory is still a model, not the factory. It is a close cousin of The Map Is Not the Territory, about words and diagrams standing in for what they describe, and it shapes how I read results from Simulating a System Event by Event.

Questions I'm still exploring

• How do I notice when a model's simplifications have started to matter?
• Can a model predict well and still explain the wrong thing?
• Which of my personal beliefs work like models, useful but incomplete?

Sources and further reading

• Roman Frigg and Stephan Hartmann, "Models in Science", Stanford Encyclopedia of Philosophy