Loading field map
Scientific realism is a position in the philosophy of science about the relationship between scientific theories and the world they describe. At its core, it is the view that the aim of science is to produce true descriptions of the world, including its unobservable aspects, and that successful scientific theories are at least approximately true. The subfield is not a single doctrine but a family of positions defined by a set of questions about how to interpret scientific knowledge, what it can legitimately claim, and how it should respond to the history of scientific change.
The fundamental issue that organizes scientific realism is the problem of unobservables. Science routinely posits entities and processes that cannot be directly perceived: electrons, genes, black holes, gravitational waves, and the deep past of the universe. These posits are indispensable to the explanatory and predictive power of modern science, yet they are not given in experience. The central question is whether we are justified in believing that such entities actually exist and that our descriptions of them are accurate, or whether they are merely useful instruments for organizing observations and making predictions.
A second, closely related question concerns the relationship between scientific success and truth. Scientific theories change over time. Older theories are often discarded or radically revised. If past theories were wrong in fundamental ways, why should we trust current ones? This is known as the pessimistic meta-induction: if most past scientific theories have turned out to be false, then by induction, most current theories will also turn out to be false. A realist must explain why current science is different, or why the falsehood of past theories does not undermine the claim that they were approximately true in their successful domains.
A third question concerns the nature of scientific progress. Realists typically hold that science makes genuine progress toward truth, not merely toward better predictive power. This raises the question of what "approximate truth" means and how it can be measured or recognized. It also raises the question of whether different scientific theories that are both successful can be incompatible with each other, and if so, what that says about the relationship between theory and reality.
The stakes of these questions extend beyond academic philosophy. If scientific realism is correct, then science provides genuine knowledge about the deep structure of reality, and its claims carry ontological weight. If anti-realism is correct, then science is a powerful tool for prediction and control, but its theoretical claims are not to be taken as literal descriptions of an independent world. This distinction matters for how we understand the authority of science, the meaning of scientific disagreement, and the relationship between science and other forms of inquiry.
The roots of scientific realism lie in the early modern period, when the mechanical philosophy of the seventeenth century claimed that natural phenomena could be explained by the motions and interactions of invisible particles. Figures such as Robert Boyle and John Locke distinguished between the observable properties of bodies and the underlying corpuscular structure that was supposed to explain them. This raised the question of whether the unobservable structure could be known or only inferred. Locke himself was cautious, suggesting that our knowledge of the real essences of things was likely beyond reach.
The term "scientific realism" itself is of twentieth-century origin, and the modern debate took shape in the mid-twentieth century. The logical positivists of the Vienna Circle, who dominated philosophy of science in the first half of the century, held that statements about unobservables were either reducible to observation statements or were meaningless. Their view was a form of anti-realism, though it was motivated by a theory of meaning rather than by skepticism about science. The later work of philosophers such as W. V. O. Quine and Wilfrid Sellars challenged the positivist distinction between observation and theory, arguing that all observation is theory-laden and that scientific theories are continuous with common sense in their claims about the world.
The modern debate was crystallized in the 1960s and 1970s by a series of influential works. Thomas Kuhn's The Structure of Scientific Revolutions (1962) argued that scientific change is not cumulative but involves shifts in paradigms that are not fully comparable with one another. This posed a serious challenge to realism, because it suggested that the history of science is a sequence of incommensurable worldviews rather than a gradual approach to truth. Bas van Fraassen's The Scientific Image (1980) articulated a sophisticated alternative called constructive empiricism, which holds that the aim of science is empirical adequacy—getting the observable phenomena right—rather than truth about the unobservable. This book set the terms for much of the subsequent debate.
The field is organized around a set of rival positions, each of which addresses the central questions in a different way. These positions are not always mutually exclusive, and many philosophers hold hybrid views, but they provide the main landmarks of the conceptual map.
Entity realism is a selective form of realism that focuses on the existence of specific theoretical entities rather than on the truth of entire theories. Its central claim is that we are justified in believing in the existence of entities that we can manipulate and use experimentally, even if the theories that describe them are imperfect or likely to change. For example, we can use electrons to produce images, accelerate particles, and generate radiation, and this practical success gives us good reason to believe that electrons exist, regardless of whether our current theory of electrons is the final word.
The motivation for entity realism is to avoid the pessimistic meta-induction. If we do not commit to the truth of whole theories, but only to the existence of the entities that those theories successfully manipulate, then the replacement of one theory by another does not undermine our belief in the entities. The approach is associated with the work of Ian Hacking and Nancy Cartwright, who argued in the 1980s that experimental practice provides a more secure foundation for realism than theoretical explanation does.
The main limitation of entity realism is that it is difficult to specify exactly what it means to "manipulate" an entity, and whether this criterion applies to all unobservables. Many important theoretical entities, such as black holes or the early universe, cannot be manipulated in any straightforward sense. Entity realism also faces the challenge of explaining how we can know the properties of an entity without committing to the truth of some theory about it.
The most common form of scientific realism is explanationist realism, also called the no-miracles argument. This position holds that the success of science is best explained by the truth of its theories. The argument is simple: if scientific theories were not at least approximately true, then their remarkable predictive and explanatory success would be a miracle. The fact that theories can make precise predictions about phenomena they were not designed to explain, and that these predictions are confirmed, is powerful evidence that the theories capture something real about the world.
This approach is associated with philosophers such as Hilary Putnam and Richard Boyd, who developed it in the 1970s and 1980s. It treats the success of science as an empirical phenomenon that itself requires explanation, and it argues that realism provides the best explanation. The approach is "explanationist" because it uses inference to the best explanation—the same kind of reasoning that science itself uses—to justify realism.
The main challenge to explanationist realism is the pessimistic meta-induction. If past theories were also successful in their day, and they turned out to be false, then the success of current theories does not guarantee their truth. Realists have responded in several ways. Some argue that past theories were not as successful as current ones, or that their success was limited to domains where they were approximately correct. Others argue that the theoretical terms of past theories can be reinterpreted as referring to the same entities as current theories, so that there is continuity beneath the apparent discontinuity. This is known as the "pessimistic induction" debate, and it remains one of the most active areas of the field.
Constructive empiricism, developed by Bas van Fraassen, is the most influential anti-realist position in the contemporary debate. It holds that the aim of science is to produce theories that are empirically adequate, meaning that they correctly describe the observable phenomena, and that acceptance of a theory involves only a belief in its empirical adequacy, not in its truth about the unobservable. The "constructive" part of the name reflects the view that scientific theories are human constructions that we use to organize experience, not discoveries of an independent reality.
Van Fraassen's position is not a form of skepticism. It does not deny that unobservable entities might exist, nor does it claim that we cannot have knowledge of them. It claims only that science does not require belief in them, and that the proper epistemic attitude toward theoretical claims is agnosticism. The view is motivated by a commitment to empiricism: the idea that all our knowledge is ultimately grounded in experience, and that we have no experiential access to the unobservable realm.
The main strength of constructive empiricism is that it avoids the problems of the pessimistic meta-induction and the difficulty of defining approximate truth. Its main weakness is that it must explain why the distinction between the observable and the unobservable is epistemically significant. Van Fraassen argues that the distinction is drawn by human sensory capacities, but critics have pointed out that this makes the epistemic status of a claim depend on the biology of the observer, which seems arbitrary. The debate over whether the observable/unobservable distinction can bear the weight that constructive empiricism places on it is a central topic in the field.
Structural realism is a position that attempts to capture the best of both realism and anti-realism. It holds that we can be realists about the structure of the world—the mathematical and relational properties described by scientific theories—while being agnostic or anti-realist about the nature of the entities that instantiate that structure. The view is motivated by the observation that when scientific theories change, the mathematical structure often survives even when the ontology does not. For example, Fresnel's theory of light as a wave in a mechanical ether was replaced by Maxwell's electromagnetic theory, but the mathematical equations describing the propagation of light were preserved.
Structural realism comes in two main versions. Epistemic structural realism holds that we can only know the structure of reality, not the nature of the entities that have that structure. Ontic structural realism, associated with philosophers such as James Ladyman and Steven French, goes further and claims that structure is all there is: there are no underlying entities, only relations. This view is motivated by developments in modern physics, particularly quantum mechanics, where the notion of individual particles with definite identities breaks down.
The main challenge to structural realism is the problem of defining "structure" in a way that is both precise and captures what is preserved across theory change. Critics have also argued that structure without entities is unintelligible, since a structure must be a structure of something. The debate between epistemic and ontic versions of structural realism, and between structural realism and more traditional forms of realism, remains active.
Selective realism is not a single position but a family of approaches that argue for realism about some parts of science while remaining anti-realist about others. The motivation is to respond to the pessimistic meta-induction by identifying which components of scientific theories are likely to be preserved across theory change and which are likely to be discarded. Different versions of selective realism focus on different criteria: some emphasize the parts of a theory that are essential to its successful predictions, others emphasize the parts that are confirmed by multiple independent lines of evidence, and still others emphasize the parts that are involved in experimental manipulation.
The most influential version of selective realism is the "divide and conquer" strategy, which argues that the pessimistic meta-induction fails because it treats theories as monolithic wholes. In reality, theories have many components, and only some of them are responsible for the theory's success. By identifying the "working posits" of a theory—the parts that do the explanatory and predictive work—we can be realists about those while remaining agnostic about the "idle" components that are later discarded.
The main challenge to selective realism is to provide a principled criterion for distinguishing the parts of a theory that are likely to be preserved from those that are not. Without such a criterion, the position risks being ad hoc, simply retrofitting our realism to whatever parts of past theories happened to survive. The debate over how to formulate and defend such a criterion is a major area of current research.
These positions are not arranged in a simple linear sequence, and they do not exhaust the field. They overlap and interact in complex ways. Entity realism and selective realism are both responses to the pessimistic meta-induction, but they focus on different aspects of scientific practice. Explanationist realism provides the general argument for realism, while structural realism offers a specific account of what it is about theories that we should be realist about. Constructive empiricism is the main anti-realist alternative, but it shares with structural realism a skepticism about our ability to know the nature of unobservable entities.
There are also positions that do not fit neatly into the realism/anti-realism dichotomy. Some philosophers argue that the debate itself is misguided, because it presupposes a distinction between the observable and the unobservable, or between truth and empirical adequacy, that cannot be sustained. Others argue that the question of realism should be settled by looking at specific scientific practices rather than by general philosophical arguments. These "practice-based" approaches, which draw on the history and sociology of science, tend to dissolve the traditional debate rather than take a side in it.
The contemporary field is characterized by a high degree of specialization and a willingness to engage with the details of scientific practice. Philosophers of science working on realism increasingly draw on case studies from specific sciences—physics, biology, chemistry, the cognitive sciences—to test and refine their positions. The debate has also become more historically informed, with philosophers examining the actual record of theory change to assess the strength of the pessimistic meta-induction.
One notable development is the increased attention to the relationship between realism and scientific modeling. Many scientific theories are not intended to be literally true descriptions of reality but are idealizations or approximations that are useful for specific purposes. This raises the question of whether realism should apply to models, and if so, in what sense. Some philosophers argue that the success of idealized models supports a form of realism about the structures they capture, while others argue that it supports a more instrumentalist view.
Another development is the growing interest in the realism debate in relation to specific scientific domains, such as quantum mechanics, evolutionary biology, and the cognitive sciences. In each of these domains, the question of what the theory tells us about reality is complicated by the nature of the subject matter. Quantum mechanics, for example, raises questions about the reality of wave functions and the nature of measurement that go beyond the general realism debate. Evolutionary biology raises questions about the reality of species and the nature of selection. These domain-specific debates are not always continuous with the general debate, but they draw on its concepts and arguments.
The field remains divided, and no consensus has emerged. The realism debate is one of the oldest and most persistent in philosophy, and it shows no signs of being resolved. What has changed is the sophistication of the arguments and the extent to which they are grounded in the details of scientific practice. The central questions remain the same: whether science tells us the truth about the world, what kind of truth it tells, and how we can know that it does.