Why scientists deliberately build wrong versions of the world
John von Neumann claimed the sciences barely try to explain anything; mostly they make models, mathematical constructs justified only because they work. Every model is a simplified stand-in for reality, useful precisely because it leaves things out. The real skill lies in knowing what to discard and where the approximation stops holding.
Scientists build models to make some slice of the world easier to grasp, measure, picture or simulate. They come in many flavours: conceptual sketches for understanding, mathematical ones for quantities, graphical ones for visualising and computational ones for simulating. Models also differ by medium. Software versions run in silico, laboratory rats serve as in vivo models, and tissue cultures grown in glassware are in vitro.
Modelling steps in when direct experiment is impossible or impractical, though a controlled measurement will always beat an estimate. Each model starts with a question. Details irrelevant to that question are simplified away, and important but secondary information is bundled together through abstraction. Even our perception of reality is already a kind of model, limited by instruments and by current theory. Turning a concept into a running simulation adds further choices, like numerical shortcuts. Despite those compromises, simulation now stands beside theory and experiment as a third pillar of research.
Assumptions define where a model applies. Special relativity assumes an inertial frame of reference; general relativity later handled accelerating frames too, a common pattern in which a newer theory succeeds an older one precisely where its assumptions break. Any model that clashes with reproducible observations must be changed or dropped, sometimes simply by shrinking the territory where it is trusted. Fitting data is not enough on its own, though. Cost of use and refutability, which lets researchers gauge confidence, also count.
Two models of the same thing can differ profoundly, reflecting the needs of users or the tastes of builders: statistical versus deterministic, discrete time versus continuous. Engineers even pair a fast, crude model with a slow, accurate one, a technique called space mapping, to avoid expensive computation. Visual models have an ancient pedigree, running from cave paintings and hieroglyphs to Leonardo da Vinci's technical drawings. Today roughly 40 magazines are devoted to the craft.
Source: Scientific modelling
