When physicists describe a frictionless plane or economists posit a perfectly rational agent, everyone involved knows these things don't exist. The assumptions are, strictly speaking, false. Yet entire disciplines build their most powerful explanations on precisely these kinds of deliberate distortions. How should we make sense of that?

This is one of the most persistent puzzles in philosophy of science: the epistemological status of scientific models. These highly idealized representations sit at the heart of how communities of researchers generate understanding, communicate findings, and make predictions. They are neither raw data nor finished theory—they occupy an unusual middle ground that resists easy categorization.

The question matters beyond academic philosophy. How we understand the role of models shapes public trust in science, determines how policymakers interpret expert advice, and influences which research programs receive institutional support. If models are merely useful fictions, their authority looks different than if they are genuine approximations to how the world works. The answer, as we'll see, is more interesting than either option alone.

Scientific theories are often too abstract to make direct contact with the world, and the world itself is too complex to be grasped without some structuring framework. Models serve as mediators—cognitive tools that stand between high-level theory and the sprawling messiness of empirical phenomena. The philosopher Mary Morgan has argued that models are partly independent of both theory and data, which is precisely what gives them their distinctive power.

Consider the Bohr model of the atom. It depicts electrons orbiting a nucleus in fixed paths, like planets around a sun. We've known for nearly a century that this picture is wrong in fundamental ways—electrons don't have definite orbits. Yet the Bohr model still shows up in chemistry textbooks because it mediates between quantum theory and observable spectral lines in a way that makes certain phenomena intelligible. It translates an abstract formalism into something a community of learners and practitioners can reason with.

This mediating function is deeply social. Models don't just help individual scientists think—they enable entire research communities to coordinate. A shared model provides a common language, a set of expectations about what counts as a good explanation, and a framework for identifying anomalies worth investigating. When climate scientists use general circulation models, they aren't claiming these models perfectly reproduce Earth's atmosphere. They're creating shared instruments of inquiry that allow hundreds of researchers across institutions to compare results and build on each other's work.

The key insight here is that a model's value isn't exhausted by its literal truth or falsity. Models earn their place in scientific practice by what they enable: explanation, prediction, communication, and the coordination of collective inquiry. They are tools shaped by and for communities, not solitary attempts to mirror nature.

Takeaway

Models are not simplified copies of reality but mediating instruments—their value lies in what they enable communities to explain, predict, and investigate together.

Here is the puzzle at its sharpest. Scientists routinely introduce assumptions they know to be false—perfectly spherical cows, infinite populations, zero transaction costs—and then use these distortions to generate claims they present as genuine knowledge. If the starting assumptions are wrong, how can the conclusions be right? This is what philosophers call the idealization paradox, and it strikes at the foundations of how scientific communities validate understanding.

One influential response comes from the tradition of de-idealization. On this view, idealized models are starting points that can be progressively corrected. A frictionless plane is not the endpoint—it's the first step. Researchers add friction, air resistance, and material imperfections layer by layer, moving closer to accurate descriptions of specific real systems. The idealization is justified because it reveals the dominant causal structure before complicating details are introduced. You learn what matters most by first stripping everything else away.

But not all idealizations work this way. Some distortions are essential to the model's explanatory power and cannot be removed without destroying the insight. The economist's perfectly rational agent isn't a rough draft of a real person—it's a deliberate fiction that reveals structural features of markets that would be invisible in a more realistic model. The philosopher Michael Weisberg calls these targetless idealizations: they don't aim at any particular real system but illuminate patterns that hold across many.

What the idealization paradox reveals is that scientific knowledge isn't simply a matter of accumulating true statements. Communities of inquiry sometimes understand more by using representations they know to be false, because those representations make certain structures, dependencies, and possibilities visible. The relationship between truth and understanding turns out to be far less straightforward than common sense would suggest.

Takeaway

Deliberately false assumptions can generate genuine understanding—not because falsehood is harmless, but because strategic distortion can reveal causal structures that literal description would obscure.

If models were straightforward approximations to a single truth, we would expect scientific progress to converge on one best model for each phenomenon. But this is not what we observe. In practice, scientists routinely maintain multiple incompatible models of the same system—and they do so not out of confusion but out of epistemic necessity. The London model and the Ginzburg-Landau model of superconductivity, for instance, contradict each other in their core assumptions, yet both remain indispensable for different questions.

This phenomenon—what we might call model pluralism—poses a serious challenge to the view that science aims at a single unified picture of reality. If two incompatible models both provide genuine understanding, then understanding cannot simply be a matter of getting closer to the one true description. Instead, it suggests that the world is complex enough to require multiple representational strategies, each illuminating different aspects, scales, or dimensions of the same phenomenon.

From a social epistemological perspective, model pluralism is not a failure of coordination—it's a feature of well-functioning epistemic communities. Helen Longino has argued that productive scientific inquiry requires a diversity of approaches and assumptions, because no single perspective can capture every relevant dimension of complex phenomena. Pluralism keeps inquiry open, prevents premature closure, and ensures that the blind spots of one model are compensated by the strengths of another.

This has practical consequences for how we structure knowledge-producing institutions. If pluralism is epistemically valuable, then funding agencies, peer review systems, and educational curricula should actively resist the temptation to prematurely standardize on a single model. The communities that produce the most robust understanding are often those that sustain productive disagreement—where multiple incompatible models are explored, tested, and refined in parallel rather than forced into artificial consensus.

Takeaway

Multiple incompatible models of the same phenomenon aren't a sign of scientific failure—they reflect the irreducible complexity of the world and the strength of communities that sustain productive disagreement.

Scientific models are neither mere fictions nor transparent windows onto reality. They are socially embedded instruments of understanding—shaped by communities, validated through collective practice, and valuable precisely because they do something more interesting than simply being true or false.

This should change how we think about scientific authority. When experts present model-based conclusions, they aren't claiming to have a perfect copy of reality. They're offering the best understanding their community has been able to construct using carefully chosen distortions, tested against evidence and refined through sustained disagreement.

The institutions that produce knowledge—laboratories, universities, funding bodies, journals—should be designed with this in mind. The goal isn't consensus for its own sake but the cultivation of productive pluralism: communities where multiple models can compete, complement, and challenge one another. That is how societies come to understand a world too complex for any single representation.