Loading field map
Can we trust that the unobservable entities and processes described by our best scientific theories—electrons, genes, black holes—actually exist? Or is science merely a tool for predicting observable phenomena, its theoretical claims a useful fiction? This tension between believing in the reality of the unobservable and treating science as an instrument of prediction has defined the modern debate over scientific realism. The debate crystallized in the 1960s, when philosophers began to ask how the stunning empirical success of science could be explained if its theories were not at least approximately true.
The contemporary realist position took shape in the 1960s and 1970s, largely as a reaction against the Logical Positivist view that talk of unobservables was either meaningless or reducible to observation statements. Hilary Putnam and Richard Boyd articulated a powerful argument that became the realist's central weapon: the "no miracles argument." If our best theories were not approximately true, they argued, their ability to make novel predictions and generate successful technologies would be a miracle. The best explanation for science's success is that its theories correctly describe a mind-independent reality. This argument, a form of Inference to the Best Explanation, gave scientific realism a clear, positive case: we should believe in the truth of our best theories, including their claims about unobservables.
This confident realism faced its most influential challenge in 1980, when Bas van Fraassen published The Scientific Image and introduced Constructive Empiricism. Van Fraassen did not deny that unobservable entities might exist; he argued instead that the aim of science is not truth about the unobservable but empirical adequacy—saving the observable phenomena. A theory is acceptable, on this view, if it correctly predicts what we can observe, regardless of whether its unobservable posits are real. Constructive Empiricism thus offered a sophisticated alternative to realism: it accepted the full content of scientific theories (including their unobservable claims) but insisted that belief should stop at the observable level. The realist's no-miracles argument, van Fraassen argued, was itself an inference to the best explanation, and such inferences are not rationally compelling when the explanation appeals to unobservable truth rather than empirical success.
Constructive Empiricism was not the only threat to scientific realism. A more historically grounded challenge emerged from the work of Larry Laudan and others: the Pessimistic Induction. If past successful theories (phlogiston, caloric fluid, the ether) turned out to be false, the argument runs, then by induction our current successful theories are also likely false. The historical record seems to show that empirical success does not guarantee truth. This argument struck at the heart of the no-miracles argument: if success can be explained without truth, the realist's central inference collapses.
The Pessimistic Induction forced realists to become selective. Instead of defending the truth of entire theories, they began to argue that only certain parts of past theories were retained across theory change. Two selective realist strategies emerged in the 1980s, each offering a different answer to the question: what should we be realist about?
Entity Realism, defended by Ian Hacking and Nancy Cartwright, argued that we should believe in the existence of unobservable entities (electrons, genes) that we can manipulate and use to cause observable effects, even if the theories describing them are false. Hacking's slogan—"if you can spray them, they are real"—captured the idea that experimental manipulation provides a more secure basis for belief than theoretical truth. Entity Realism thus narrowed the realist commitment: it abandoned the claim that our best theories are true, while preserving belief in the entities that we can causally interact with in the laboratory.
Structural Realism, introduced by John Worrall, took a different selective path. Worrall argued that what survives theory change is not the entities but the mathematical structure of theories. Fresnel's theory of light posited an ether, which turned out not to exist, but its mathematical equations for the behavior of light were retained in Maxwell's electromagnetic theory. Structural Realism claims that we should be realists about the structure of our theories—the relations expressed in their equations—while remaining agnostic about the nature of the underlying entities. This position later split into two variants: epistemic structural realism, which holds that we can only know the structure of the unobservable world, and ontic structural realism, which goes further to claim that structure is all there is—that there are no underlying entities at all, only relations. The contrast with Entity Realism is sharp: Entity Realism puts its faith in manipulable objects, while Structural Realism puts its faith in mathematical form.
By the 1990s, the debate had become more nuanced. Realists and anti-realists had staked out clear positions, but each side faced internal pressures to refine its commitments.
Explanatory Realism, developed by Richard Boyd and others, attempted to strengthen the original no-miracles argument by focusing on the role of explanation in scientific practice. Instead of defending the truth of entire theories, Explanatory Realism argues that the success of scientific explanations—their ability to unify phenomena, generate novel predictions, and guide successful intervention—provides evidence for the approximate truth of the explanatory principles involved. This position aimed to answer Constructive Empiricism by showing that empirical adequacy alone cannot account for the explanatory power that scientists demand. If a theory's success depends on its explanatory depth, the realist can argue that truth is the best explanation of that depth.
Scientific Pluralism, emerging in the 1990s from the work of Helen Longino, Nancy Cartwright, and others, challenged a key assumption shared by many earlier realists: that science converges on a single, complete description of the world. Pluralists argued that different scientific models and theories can be successful in different domains without being compatible or reducible to a single unified account. This view undercut the realist's reliance on the success of science as a whole, since success might be local and fragmented. Pluralism also drew on insights from the sociology of science and feminist philosophy of science, which emphasized the role of values, interests, and social context in shaping scientific knowledge. The pluralist challenge was not anti-realist in the style of Constructive Empiricism; rather, it questioned whether realism needed to be committed to a single, unified picture of reality.
Perspectival Realism, developed by Ronald Giere and others in the 2000s, offered a way to combine realism with pluralism. Perspectival Realism holds that scientific knowledge is always from a particular perspective—shaped by the instruments, models, and theories we use—but that this does not prevent it from being genuinely about the world. Different perspectives can reveal different aspects of reality without being reducible to a single, perspective-independent description. This position draws on the history of science (where different theories have offered complementary insights) and on feminist philosophy of science (which has argued that knowledge is always situated). Perspectival Realism thus preserves the realist commitment to a mind-independent world while accommodating the pluralist insight that our access to that world is always partial and shaped by our cognitive and practical interests.
Today, the debate over scientific realism is more empirically informed than ever. Philosophers have turned to detailed case studies from the history of science, experimental practice, and the actual reasoning of scientists to test the claims of different frameworks. The leading positions—Scientific Realism, Constructive Empiricism, Entity Realism, Structural Realism, Explanatory Realism, Scientific Pluralism, and Perspectival Realism—all remain active, but they now occupy different roles.
What do these frameworks agree on? Most accept that the Pessimistic Induction poses a genuine challenge, and that selective realism of some kind is necessary. There is broad agreement that the no-miracles argument, while powerful, cannot be applied wholesale to entire theories; some form of selective commitment is required. There is also growing recognition that the debate cannot be settled by a priori argument alone—empirical work on how science actually succeeds and fails is essential.
Where do they disagree? The deepest disagreement concerns the proper target of realist commitment. Entity Realists and Structural Realists offer competing accounts of what survives theory change: manipulable objects versus mathematical structure. Explanatory Realists argue that the focus should be on explanatory success rather than manipulation or structure. Scientific Pluralists and Perspectival Realists challenge the assumption that there is a single, convergent story to tell, while Constructive Empiricists continue to deny that belief in unobservables is rationally required at all. The debate has thus shifted from a simple opposition between realism and anti-realism to a more complex landscape of selective realisms, each with its own criteria for what deserves our belief. The question is no longer simply "are you a realist?" but "what kind of realist are you, and why?"