Loading field map
How do we know that science gives us genuine knowledge about the world? This question has driven the philosophy of science for over two millennia. At its heart lies a persistent tension: is scientific knowledge built by a single rational method that any careful thinker can follow, or is it shaped by historical context, social forces, and pragmatic choices that no universal method can capture? The frameworks that follow represent different attempts to resolve this tension, each offering a distinctive picture of what makes science special—or whether it is special at all.
Aristotelian Demonstrative Science (c. 350 BCE–1600 CE) set the original template. For Aristotle, genuine scientific knowledge consisted of necessary truths deduced from self-evident first principles. The Posterior Analytics laid out a system in which demonstration from axioms guaranteed certainty. This framework dominated for nearly two millennia, but it faced a deep problem: where do the first principles come from? Aristotle himself allowed that they are grasped by intuition, but later thinkers found this answer unsatisfying.
Inductivism (1620–1900) offered a radical alternative. Francis Bacon argued that science should begin not with axioms but with careful observation and experiment, gradually ascending to general laws through inductive generalization. Where Aristotelian science moved from principles downward, inductivism moved from facts upward. This shift transformed the image of the scientist from a deductive reasoner to a patient collector of data. Yet inductivism faced its own crisis: David Hume showed that induction cannot be logically justified—no number of observed instances guarantees that the next case will resemble them. The problem of induction haunted every subsequent framework.
Hypothetico-Deductivism (1830–Present) responded to inductivism's difficulties by reversing the direction of inference. Instead of deriving laws from data, scientists first propose a hypothesis and then deduce observable consequences from it. If those consequences match observation, the hypothesis is confirmed; if not, it is refuted. William Whewell gave this approach its classic formulation, arguing that hypotheses are creative leaps guided by a 'consilience of inductions.' Unlike inductivism, hypothetico-deductivism acknowledged that theory precedes observation, but it struggled to specify how much confirmation a successful prediction provides.
Pragmatist Philosophy of Science (1877–Present) took a different route. Charles Sanders Peirce argued that the meaning of a scientific concept lies in its practical consequences—what we would expect to observe if the concept were true. For pragmatists, scientific inquiry is a self-correcting process guided by the goal of settling belief and enabling successful action. This framework did not try to eliminate uncertainty; it treated science as a fallible but effective tool for coping with the world. Pragmatism coexisted with hypothetico-deductivism, offering a more flexible account of scientific progress.
Conventionalism (1902–Present), associated with Henri Poincaré, added another layer. Some scientific principles, Poincaré argued, are neither empirical truths nor logical necessities but conventions—freely chosen frameworks that organize experience. Geometry, for instance, can be chosen for convenience rather than truth. Conventionalism narrowed the scope of what science must discover: many of its most abstract claims are not discoveries at all but decisions about how to describe the world.
Logical Empiricism (1920–1960) was the dominant framework of mid-twentieth-century philosophy of science. The Vienna Circle sought to unify science under a single logical method: all meaningful statements are either analytic (true by definition) or verifiable by sense experience. Metaphysical claims, including traditional philosophy, were dismissed as meaningless. Logical empiricism absorbed the hypothetico-deductive model and added a strict verificationist criterion of meaning. But the verification principle itself could not be verified, and the program collapsed under internal pressure: no purely observational language could capture the content of theoretical claims.
Critical Rationalism (1934–Present), Karl Popper's alternative, reacted directly against logical empiricism. Popper rejected verification entirely. Science progresses, he argued, not by confirming theories but by subjecting them to severe tests that could falsify them. A theory is scientific only if it is falsifiable—if it rules out some possible observations. This shifted the criterion of demarcation from meaning to testability. Critical rationalism preserved the idea of a single scientific method but replaced confirmation with falsification. Critics pointed out that no observation is ever certain enough to decisively falsify a theory, and scientists rarely abandon a theory after a single failed test.
The Deductive-Nomological Model (1948–1970), developed by Carl Hempel and Paul Oppenheim, aimed to capture the logical structure of scientific explanation. An explanation, on this view, is a deductive argument whose premises include at least one general law and whose conclusion describes the phenomenon to be explained. The D-N model was a direct extension of logical empiricist ideals: explanation and prediction had the same logical form. But it faced counterexamples—explanations that seemed good but did not fit the deductive pattern, and arguments that fit the pattern but did not explain. By the 1970s, the model was widely seen as too narrow.
Bayesian Confirmation Theory (1950–Present) offered a probabilistic alternative to both verification and falsification. On this view, scientific hypotheses are confirmed incrementally: evidence raises the probability of a hypothesis according to Bayes' theorem. Bayesianism absorbed the hypothetico-deductive insight that prediction matters but gave it a precise quantitative form. Unlike hypothetico-deductivism, it could handle degrees of confirmation and explain why surprising predictions carry more weight. Bayesian frameworks remain active today, especially in formal epistemology and philosophy of statistics, where they provide a unified account of learning from evidence.
Scientific Realism (1960–Present) emerged as a response to the anti-realist tendencies of logical empiricism and instrumentalism. Realists argue that successful scientific theories are approximately true and that the unobservable entities they posit—electrons, genes, black holes—really exist. The central argument for realism is the 'no miracles' argument: if theories were not at least approximately true, their predictive success would be a miracle. Scientific realism competes directly with Constructive Empiricism (1980–Present), Bas van Fraassen's alternative. Constructive empiricists agree that science aims to produce empirically adequate theories—theories that save the phenomena—but deny that acceptance requires belief in unobservable entities. The realism debate remains one of the most active in the field, with each side refining its arguments over decades.
Inference to the Best Explanation (1965–Present), developed by Gilbert Harman and later defended by realists like Richard Boyd and Peter Lipton, holds that we are justified in inferring the truth of a hypothesis when it provides the best explanation of the available evidence. This framework is often used to support scientific realism: the best explanation of experimental results is that the unobservable entities posited by the theory actually exist. Critics, including constructive empiricists, argue that explanatory power is not a reliable guide to truth.
The Semantic View of Theories (1960–Present) superseded the logical empiricist 'syntactic view,' which treated theories as sets of sentences in a formal language. Instead, the semantic view identifies a theory with a class of models—mathematical structures that satisfy the theory's laws. This shift mattered because it freed philosophy of science from the assumption that theories must be axiomatized in first-order logic. The semantic view coexists with the syntactic approach in some areas but has become the dominant framework for understanding theory structure.
Kuhnian Paradigm Theory (1962–Present) transformed the field by arguing that scientific change is not a gradual accumulation of knowledge but a series of revolutions. Normal science operates within a paradigm—a shared set of assumptions, methods, and exemplars. Anomalies accumulate until a crisis triggers a paradigm shift, after which the new paradigm is incommensurable with the old. Kuhn's framework reacted against logical empiricism by emphasizing the social and historical dimensions of science. It remains influential, though critics argue that incommensurability is overstated and that Kuhn's account of rationality is too pessimistic.
Naturalized Epistemology (1969–Present), proposed by W.V. Quine, argued that epistemology should become a branch of empirical psychology. Instead of trying to justify scientific knowledge from a priori principles, philosophers should study how humans actually form beliefs. This framework absorbed the pragmatist emphasis on fallibility and replaced the search for foundations with descriptive inquiry. Naturalized epistemology remains active, especially in cognitive science and philosophy of psychology, though critics worry that it abandons the normative task of distinguishing good from bad reasoning.
The Methodology of Scientific Research Programmes (1970–Present), Imre Lakatos's response to both Popper and Kuhn, tried to reconcile rational reconstruction with historical complexity. A research programme consists of a hard core of irrefutable assumptions and a protective belt of auxiliary hypotheses that can be modified to accommodate anomalies. Programmes are progressive if they predict novel facts and degenerating if they merely patch up anomalies. Lakatos preserved Popper's emphasis on testability while acknowledging that theories are not abandoned after a single falsification. His framework competes with Kuhnian paradigm theory as a model of theory change, offering a more rationalist picture.
Methodological Anarchism (1975–Present), Paul Feyerabend's provocative alternative, rejected the very idea of a universal scientific method. In Against Method, Feyerabend argued that science progresses precisely because scientists break every methodological rule. His slogan 'anything goes' was not a prescription but a description of actual scientific practice. Methodological anarchism coexists with Lakatos's programme as a radical critique of rationalist philosophy of science. It remains a minority position but continues to challenge assumptions about scientific rationality.
The Strong Programme (1976–2000), developed by David Bloor and Barry Barnes at Edinburgh, applied the sociology of knowledge to science itself. It demanded that explanations of scientific belief be symmetrical—the same types of social causes should explain both true and false beliefs. The Strong Programme absorbed insights from Kuhn but pushed further, treating scientific knowledge as entirely socially constructed. It provoked intense debate: realists accused it of relativism, while supporters argued that it revealed the social mechanisms behind theory choice. By 2000, the Strong Programme had largely faded as a distinct movement, but its influence persists in social epistemology and science studies.
Feminist Philosophy of Science (1980–Present) emerged from the recognition that gender biases have shaped scientific practice and theory. Feminist philosophers like Helen Longino and Sandra Harding argued that traditional ideals of objectivity conceal male-centered assumptions. Longino's critical contextual empiricism proposed that objectivity is achieved through social criticism within a community, not by eliminating values. Feminist philosophy of science does not reject science but demands that it become more inclusive and self-aware. It relates to the Strong Programme by emphasizing social context but retains a normative commitment to improving science.
Scientific Pluralism (1980–Present) holds that many scientific questions admit multiple, equally legitimate answers. Pluralists argue that the world is too complex to be captured by a single theory or method. This framework emerged partly as a response to the failures of reductionism and the success of diverse research programs in biology, psychology, and economics. Scientific pluralism coexists with realism and anti-realism: one can be a pluralist about models while remaining a realist about entities. It remains an active research program, especially in philosophy of biology and social science.
Social Epistemology of Science (1980–Present) studies how social structures—peer review, funding, collaboration, dissent—affect the production of knowledge. Unlike the Strong Programme, social epistemology does not assume that social causes undermine objectivity; it asks how social arrangements can promote or hinder reliable belief. Philip Kitcher's 'well-ordered science' and Helen Longino's 'critical contextual empiricism' are prominent examples. Social epistemology of science absorbs insights from feminist philosophy and scientific pluralism while maintaining a focus on epistemic norms.
Causal-Mechanical Explanation (1984–Present), developed by Wesley Salmon and later refined by James Woodward and others, offers an alternative to the D-N model. Instead of subsuming phenomena under laws, causal-mechanical explanation works by describing the causal mechanisms that produce the phenomenon. This framework revived interest in causation after decades of neglect and has become influential in philosophy of biology, neuroscience, and social science. It coexists with the semantic view of theories: models often represent causal mechanisms.
Structural Realism (1989–Present), defended by John Worrall and later by James Ladyman and others, emerged as a response to the stalemate between scientific realism and constructive empiricism. Structural realists argue that we should be realist about the mathematical structure of theories but agnostic about the nature of unobservable entities. This position preserves the realist intuition that successful theories capture something real while accommodating theory change: structure is retained across theory change even when entities are replaced. Structural realism remains an active research program, with ongoing debates about how to define 'structure' and whether it escapes the problems of standard realism.
Today, no single framework dominates the philosophy of science. The leading frameworks—scientific realism, constructive empiricism, Bayesian confirmation theory, scientific pluralism, social epistemology, and structural realism—agree that science is a fallible, historically situated enterprise that nonetheless produces remarkably successful predictions and explanations. They disagree sharply about what this success tells us about the world. Realists and constructive empiricists continue to debate whether we should believe in unobservable entities. Bayesians and hypothetico-deductivists offer competing accounts of confirmation. Pluralists and social epistemologists challenge the idea that any single method or perspective can capture scientific rationality. The field has broadened from its early focus on logical method to embrace the social, historical, and pragmatic dimensions of scientific practice, and this pluralism itself is now a subject of ongoing inquiry.