Loading field map
How can a model—a simplified, often fictional system—represent a real-world target? This question lies at the heart of the philosophy of science's subfield of models and representation. For almost a century, philosophers have debated whether representation is a matter of formal structure, practical use, or something else entirely. The history of these debates can be organized around six major frameworks, each emerging as a response to the limitations of its predecessors.
The earliest systematic framework, the Syntactic View (1920-1960), grew out of logical empiricism. It treated scientific theories as sets of sentences in a formal language, with models serving merely as interpretations of those sentences. On this view, a theory is a deductively closed set of axioms or laws, and a model is a structure that satisfies those axioms. Representation was thus a matter of syntactic derivation: a model represented a system if the system's behavior could be logically deduced from the theory's statements. This approach had the virtue of precision, but it faced a serious problem: actual scientific practice rarely fits such a rigid, linguistic mold. Scientists often use incomplete, idealized, or non-linguistic models—diagrams, analogies, scale models—that do not derive from axioms.
The Semantic View (1960-1990) emerged as a direct response to these limitations. Instead of starting with language, it started with models themselves. A theory, on this view, is a family of models (often defined set-theoretically), and representation is a relationship of isomorphism (or similarity) between the model and the target system. For example, the Bohr model of the atom represents the hydrogen atom because there is a structural mapping between the model's orbits and the atom's energy levels. The Semantic View preserved the formalist ambition of the syntactic approach—it still sought a precise, structural account of representation—but it broadened the notion of a theory to include non-linguistic entities. However, it inherited new difficulties: many models involve idealizations that break isomorphism (e.g., frictionless planes), and the assumption that representation requires a single, fixed structural mapping seemed too restrictive for models that are constantly adjusted.
By the 1980s, a new framework began to take shape, driven by close attention to how scientists actually construct and work with models. Models as Mediators (1980-Present) rejected the idea that models are simply parts of theories. Instead, it portrayed models as autonomous, often partial, and highly idealized representations that act as intermediaries between theory and the world. A model does not typically derive from a single theory; it may combine elements from multiple theories, empirical data, and pragmatic assumptions. For example, the Lotka-Volterra model in ecology does not flow directly from any fundamental theory—it is a deliberately simplified system that allows ecologists to explore predator-prey dynamics.
Representation, on this view, is not a two-term relation (model–world) but a three-term relation: a model stands for a target system in virtue of a user's intentions and the model's capacity for specific cognitive or practical tasks. The philosopher Ronald Giere emphasized that scientists actively use models to explore possibilities, make predictions, and explain phenomena. Models are not merely pictures; they are tools for reasoning. This framework thus shifted the focus from static structural correspondence to dynamic practice, including processes of denotation (picking out target entities), demonstration (deriving consequences from the model), and interpretation (applying results to the real world).
While Models as Mediators transformed the debate by emphasizing practice, philosophers also deepened the discussion about the nature of the representational relationship itself. Three overlapping frameworks arose after 1980, each offering a distinctive answer to the question: What, if anything, does a successful model tell us about the world?
Structural Realism (1980-Present) re-entered the realism/anti-realism debate by arguing that what survives theory change is the mathematical structure of models, not their ontology. Unlike the Semantic View's demand for full isomorphism, Structural Realism accepts that models are often false in their details but nevertheless preserve structural relations. For instance, the transition from Fresnel's ether theory to Maxwell's electromagnetism retained the structure of the equations even as the underlying ontology changed. Structural Realism thus provides a selective realist account: we can be realists about the structure represented by models, even if the model's fictional entities (like the ether) are abandoned.
Fictionalism (2000-Present) takes an even more radical step. It explicitly compares the use of models in science to the use of fictions in literature. On this view, models are akin to fictions—they involve make-believe, pretense, and imagined entities. When a physicist speaks of a point mass or an ideal gas, she is not making a literal claim about the world; she is engaging in a controlled fiction that yields useful insights. Fictionalism does not reject representation; rather, it argues that representation in science is not a matter of truth or correspondence but of guiding inference and prediction. This framework emerged partly in reaction to the prevalence of idealizations that cannot be eliminated from scientific practice.
Inferentialism (2000-Present), by contrast, focuses on the function of models in reasoning. Instead of asking whether models are true or similar to targets, Inferentialism asks: What can we infer from the model? Representation is defined in terms of surrogative reasoning: a model represents a target if we can draw conclusions about the target by manipulating the model and then translating the results. This view aligns with the practice-oriented spirit of Models as Mediators but reduces the question of representation to the pragmatic criterion of inferential utility. Inferentialism shares Fictionalism's rejection of a strong correspondence requirement, yet it differs by grounding representation in the activity of inference rather than in imagination or pretense.
Today, the four active frameworks—Models as Mediators, Structural Realism, Fictionalism, and Inferentialism—coexist without any having achieved dominance. They agree on several points: all reject the formalist reduction of representation to static isomorphism, all acknowledge the centrality of idealization and model autonomy, and all agree that representation is mediated by human purposes. Their disagreements lie in which aspects they emphasize:
The field remains pluralistic. Many philosophers now adopt a pragmatic stance, treating each framework as suited to different aspects of modeling. The central question—what makes a model represent the world?—has no single answer, but the sustained debate has deepened our understanding of scientific practice and the nature of theoretical knowledge.