Models And Representation

Scientific knowledge rarely arrives as bare theory.

11 frameworks, from 1950 to today · Philosophy Of Science · Philosophy

still a working view ended · each leads to its entry below

Overview

Scientific knowledge rarely arrives as bare theory. It arrives in models: the double helix, the billiard-ball gas, the Lotka–Volterra predator–prey equations, a climate simulation, a trained neural network1. Models and representation is the part of philosophy of science that asks what these objects are, how they connect to the world, and what it means for one thing to stand for another2. The stakes are large. If models are the working units of science, then an account of science built only on theories and evidence will miss most of what scientists actually do. And if representation is not a simple copying relation, then the authority of a model—why we trust a simulation, or refuse to trust a black box—needs explaining rather than assuming.

Models as useful fictions in logical empiricism

The story begins inside logical empiricism, the mid-twentieth-century movement that sought to reconstruct science in formal logic. Its central target was the theory: a set of axioms in a formal language, connected to observation by correspondence rules3. Models had a modest place in this picture. In the 1950s and 1960s, philosophers such as Rudolf Carnap and Ernest Nagel treated a model as an interpretation of a formal calculus—an assignment of meanings that makes the abstract axioms concrete. A model of the axioms of geometry might be points and lines on a plane; a model of a physical theory might be a mechanical system. This "model heuristic" was pedagogically useful but philosophically secondary: the theory was the real bearer of content, and the model was a ladder to be discarded.

The limitation was practical. Working scientists did not seem to reason from axioms down. They built simplified systems—frictionless planes, ideal gases, isolated populations—and manipulated them. The logical-empiricist picture had no room for this middle layer between abstract theory and messy data.

The semantic view: theories as families of models

The first major break came with the semantic view of theories, developed from the 1960s onward by Patrick Suppes, Bas van Fraassen, Frederick Suppe, and others. On this account, a theory is not a linguistic object at all but a collection of models—typically mathematical structures—together with a specification of which real systems they are intended to represent. Van Fraassen's The Scientific Image (1980) made the view prominent: to present a theory is to specify a class of structures, and to accept a theory is to believe that the world is (or can be described as) one of them.

The change mattered because it put models first. A classical mechanics problem—say, a pendulum—is handled by constructing a model: a point mass on a massless rod, with equations of motion. The theory supplies a repertoire of such structures; the scientist selects and adjusts one. This fit actual practice far better than axiomatic reconstruction, and it made the question of how a model hooks onto a real system unavoidable. The semantic view answered it thinly, usually by appeal to an "intended application" or an isomorphism between model and system. That thinness set up the next generation of problems.

Isomorphism and its discontents: the similarity account

If representation is isomorphism—a structure-preserving mapping between model and world—then representation looks precise. But real models are rarely isomorphic to anything. The billiard-ball model of a gas does not preserve the structure of molecules; it deliberately distorts it. In the 1990s and 2000s, Ronald Giere proposed that representation is instead a matter of similarity, understood as sharing certain specified respects and degrees. A model represents a system in virtue of being similar to it in the ways the modeler intends and the user specifies. Giere's account, developed in Explaining Science (1988) and later work, made representation a three-place relation: model, real system, and the respects in which they are alike.

Similarity solved the rigidity of isomorphism but raised a worry: similarity is cheap. Any two things are similar in some respect. The account needed the "specified respects" to do real work, and critics pressed on whether those specifications are themselves part of the model or supplied by users. The debate pushed philosophers toward accounts that foreground use, purpose, and practice.

Models as mediators and as instruments

A parallel line of work, rooted in close study of scientific practice, treated models as tools rather than as pictures. Mary Morgan and Margaret Morrison's edited volume Models as Mediators (1999) argued that models are autonomous intermediaries: they are constructed from a mix of theory, data, and mathematical formalism, and they mediate between theory and the world rather than merely instantiating either. A model can be investigated in its own right—its behavior explored, its assumptions varied—before any claim about reality is made. This "models as mediators" framing drew on detailed case studies from economics, biology, and physics, and it shifted attention to what modelers do with models.

A related strand treated models as instruments for intervention and prediction. Robert Sugden's account of models as "credible worlds" in economics, and earlier work by Nancy Cartwright, emphasized that models earn trust by being manipulated and by surviving use. Cartwright's How the Laws of Physics Lie (1983) and later The Dappled World (1999) went further: she argued that fundamental laws rarely hold of real systems, and that what does the explanatory work are capacities—stable dispositions of things—realized in nomological machines: arrangements of components that, when assembled, produce regular behavior. A laser is her recurring example: the regularity of laser light is produced by a carefully engineered setup, not by a law that holds everywhere. On this view, models describe the machines, and representation is a matter of capturing the capacities that the machine realizes.

Fictions, idealization, and the epistemology of modeling

If models distort, what is the epistemic status of a distortion? A prominent answer came from the "models as fictions" literature. Nancy Cartwright and others had already noted that models involve deliberate falsehood. In the 2000s, philosophers such as Roman Frigg and, in a different register, Mauricio Suárez developed fictionalist accounts: models are like literary fictions, and our engagement with them involves imaginative participation governed by rules. Frigg's work on fiction and scientific modeling argued that model systems are stipulated entities, and that claims about them are true within the fiction. The payoff is an explanation of how false models can yield true predictions and genuine understanding: the fiction is a controlled construction, and its relation to the world is managed by the modeler.

A complementary line concerned idealization and its costs. Work by Ernan McMullin, and later by Michael Weisberg, analyzed how models simplify—through omission, abstraction, and idealization—and asked when such simplification is legitimate. Weisberg's Simulation and Similarity (2013) offered a taxonomy of modeling strategies and a similarity-based account of representation that tried to absorb the lessons of the earlier debates.

Surrogative reasoning: what models let us do

A distinct question is not what models are but what they enable. The surrogative reasoning account, associated with Chris Swoyer and developed by Mauricio Suárez and others, holds that a model represents a target when it can be used as a surrogate for it: we reason about the model and transfer conclusions to the target. Representation is thus tied to inference. If a model supports reliable inferences about a system—predictions, explanations, interventions—it represents that system for those purposes. This account connects the metaphysics of representation to the epistemology of model use, and it accommodates the fact that the same model can represent different systems for different purposes.

Computer simulation and the opacity problem

The rise of computer simulation from the mid-twentieth century onward—weather models, molecular dynamics, Monte Carlo methods—forced new questions. A simulation is a model run on a machine, and its outputs are often not analytically tractable. Philosophers such as Paul Humphreys, in Extending Ourselves (2004), and later Eric Winsberg, in Science in the Age of Computer Simulation (2010), examined how simulation changes the epistemology of science: results are produced by computational processes that no one fully traces, and trust must be distributed across code, hardware, and model assumptions. The question of whether simulation is a distinct kind of scientific method, or an extension of modeling, remains live.

The most recent turn concerns machine learning. A trained neural network is a model, but its internal structure is not interpretable in the way a differential equation is4. The opacity of such models—the difficulty of saying why they produce the outputs they do—raises sharp questions for representation4. If representation requires a grasp of the model's structure and its relation to the target, opaque models seem to represent without being understood. Philosophers have responded by distinguishing kinds of opacity (architectural, epistemic, explanatory) and by asking whether prediction alone can ground representation. This work is young, and its shape is still being contested.

The present

Today the field is plural. The semantic view remains a reference point, though few hold its original form; similarity and surrogative-reasoning accounts are widely used; fictionalism and the models-as-mediators tradition inform practice-oriented studies; and the simulation and machine-learning literatures are expanding fastest. The open questions are recognizable descendants of the old ones: what makes a distortion legitimate, whether representation can be separated from use, and how to hold on to epistemic responsibility when the model's workings outrun our understanding.

References

  1. Scientific modelling (Types), Wikipedia
  2. Philosophy of science, Wikipedia
  3. Received view of theories, Wikipedia
  4. Black box, Wikipedia

The frameworks of Models And Representation

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1950-1970

    Received view of theories

    What is a scientific theory made of?

    Fly to its tower in Nousfarer

  2. 1960-Present

    Semantic View of Theories

    Are theories sets of models or sets of sentences?

    Fly to its tower in Nousfarer

  3. 1968-Present

    Idealization

    When is simplifying a model legitimate?

    Fly to its tower in Nousfarer

  4. 1983-Present
  5. 1983-Present
  6. 1988-Present
  7. 1991-Present

    Surrogative Reasoning

    How do we reason with models about the world?

    Fly to its tower in Nousfarer

  8. 1999-Present

    Models as Mediators

    What work do models do between theory and data?

    Fly to its tower in Nousfarer

  9. 2002-Present

    Fictionalism

    Can false models yield true knowledge?

    Fly to its tower in Nousfarer

  10. 2004-Present

    Epistemology of Computer Simulation

    How do we trust results no one can check by hand?

    Fly to its tower in Nousfarer

  11. 2015-Present

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Received view of theoriesSemantic View of TheoriesIdealizationCapacities and Nomological MachinesModels as Instruments and Credible Wo…Similarity Account of RepresentationSurrogative ReasoningModels as MediatorsFictionalismEpistemology of Computer SimulationOpacity and Machine-Learning Models

Beyond the field

Where Models And Representation's ideas came from in other fields, and what they led to there: the strongest 12 of 16 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Models And Representation is a planet in Nousfarer

Its city has a tower for each of these 11 frameworks. Land, walk in, and climb one concept at a time.

Land on Models And Representation

Related fields

Fields its ideas are linked to

More in Philosophy Of Science

See where Models And Representation sits on the Map of Human Knowledge