Realism vs Anti-Realism

Three axes of disagreement

The single word "realism" hides at least three independent positions. Untangling them is the first step in any serious discussion.

Metaphysical realism. There is a world whose nature does not depend on what anyone thinks or says about it. Mountains, electrons, and the past obtain (or fail to obtain) regardless of human cognition. The opposing view — idealism, in its various forms — holds that what exists depends on minds: Berkeley's esse est percipi, Kant's transcendental idealism, or post-Kantian constructivisms.

Semantic realism. Statements about a domain have determinate truth values, fixed by how things stand in that domain, regardless of our capacity to verify them. "It rained in London on 6 January 1407" is either true or false even though no one alive can confirm which. Semantic anti-realism, developed most rigorously by Michael Dummett, denies this — meaning is constituted by use, and a statement we could not in principle settle has no determinate truth value.

Epistemic realism. Our best theories are at least approximately true, and their central theoretical terms refer. Scientific realism is the most-discussed version: when physics says there are quarks, there are quarks. Anti-realism here can take many shapes — instrumentalism (theories are calculation devices, not true or false), constructive empiricism (theories aim at empirical adequacy, not truth about unobservables), structural realism (only the mathematical structure of theories tracks reality).

One can mix and match. Van Fraassen is a metaphysical realist (the world exists) and a semantic realist (statements about electrons have truth values) but an epistemic anti-realist (we should not believe those statements about electrons even when we use them).

The realist's master argument

The most cited argument for scientific realism is Hilary Putnam's no-miracles argument, formulated in Mathematics, Matter and Method (1975) and refined by Richard Boyd in the late 1970s. The argument runs:

1. Mature science predicts new phenomena with stunning accuracy and produces working technology — semiconductors, antibiotics, GPS, mRNA vaccines.
2. The simplest explanation for this success is that the theories are at least approximately true, and that their unobservable posits exist.
3. If we deny this, we make the success a miracle — an unexplained run of luck spanning four centuries.
4. By inference to the best explanation, we should accept that the theories are approximately true.

Realists treat this as the closest thing in philosophy to an empirical argument. Critics like Arthur Fine and Larry Laudan attacked premise 2 — there are other explanations of success, including selection effects (we only call successful theories good) and survivor bias (we forget the equally-confident failures). Bas van Fraassen pressed a related point: the no-miracles argument is itself an inference to the best explanation, but inference to the best explanation is precisely the kind of move realists are trying to license. The argument is therefore circular — or at least begs the question against an opponent who rejects IBE for unobservables in the first place.

The anti-realist's master reply

Larry Laudan's "A Confutation of Convergent Realism" (1981) gave the canonical anti-realist comeback: the pessimistic meta-induction. Look at the history. The phlogiston theory of combustion, the caloric theory of heat, the crystalline-spheres model of the heavens, optical and electromagnetic ether, the elastic-solid theory of light, classical absolute space and time — all were predictively successful theories of their era, all are now considered to posit non-existent entities. By induction, our current best theories will probably go the same way. Their success is poor evidence for their truth.

Realists have replied in two main ways. Selective realism — championed by Philip Kitcher, Stathis Psillos, and John Worrall — agrees that we should not be globally realist about whole theories but argues we can be realist about the specific posits and structures that did the actual predictive work. The ether dropped out of physics, but Maxwell's equations did not; structural realism (Worrall's structural realism, 1989) holds that what survives theory change is mathematical structure, not specific furniture. Entity realism (Ian Hacking, 1983) takes a different tack: believe in entities you can manipulate experimentally — "if you can spray them, they're real" — even when the surrounding theoretical narrative is provisional.

Worked example: do electrons exist?

An electron is unobservable. No microscope sees it directly; we infer it from cathode-ray deflection, photoelectric effects, and a thousand experiments since J. J. Thomson's 1897 measurement of e/m. The realist says: electrons exist, with charge ≈ −1.6 × 10⁻¹⁹ coulombs, mass ≈ 9.1 × 10⁻³¹ kg, exhibiting wave-particle duality. The constructive empiricist (van Fraassen) says: electron talk is a wonderful organising device for predictions about cloud chambers and cathode tubes, and we should use it freely; but believing in electrons goes beyond the evidence. Empirical adequacy is enough.

The selective realist replies: structural features survive even radical theory change. Thomson's electron, Bohr's electron, and the QFT electron-as-excitation-of-a-field are not the same object metaphysically — but they share enough mathematical structure (charge, lepton number, fermionic statistics) that we can plausibly say "electrons exist" while remaining cautious about the underlying ontology. Hacking presses a different point: experimentalists routinely spray electrons in particle-beam apparatus to probe other targets. If you can use electrons as instruments, the case for their reality is no longer purely inferential.

Realism vs anti-realism by axis

Axis	Realist position	Anti-realist alternatives	Major figures
Metaphysical	Mind-independent world exists	Idealism; constructivism; Putnam's internal realism	Berkeley, Kant, Putnam (later)
Semantic	Bivalence: every statement true or false	Verificationism; intuitionism; truth-as-warranted-assertibility	Dummett, Brouwer, Wright
Scientific entity	Unobservable posits really exist	Instrumentalism; constructive empiricism	Van Fraassen, Mach, Duhem
Scientific structure	Theories approximately true throughout	Structural realism (only structure survives)	Worrall, Ladyman
Mathematical	Numbers, sets, functions are objects	Nominalism; fictionalism; if-thenism	Gödel, Maddy / Field, Balaguer
Moral	Moral facts hold mind-independently	Expressivism; error theory; constructivism	Moore / Mackie, Blackburn, Korsgaard
Modal	Possible worlds genuinely exist	Linguistic ersatzism; modal anti-realism	D. Lewis / Plantinga

Dummett's manifestation challenge

Michael Dummett, in essays beginning with "Truth" (1959) and developed across Truth and Other Enigmas (1978), reshaped the debate by tying it to philosophy of language. His manifestation argument goes like this: the meaning of a statement is what a competent speaker grasps when they understand it. Whatever they grasp must be manifestable in their use of the statement — in the conditions under which they assert, deny, or accept it. But realists treat truth as potentially evidence-transcendent: a statement could be true even if no one could ever verify it. How could a speaker manifest grasp of that? Dummett's conclusion was that the only intelligible notion of truth, for the bulk of our discourse, is one tied to assertibility conditions — making him a semantic anti-realist.

The manifestation challenge does not deny the world; it denies that our language can describe it in evidence-transcendent ways. Critics including Crispin Wright (a sympathetic critic) and McDowell pushed back on Dummett's verificationism while preserving its force. The dispute fed directly into intuitionistic logic in mathematics, where bivalence is dropped and proofs constructively understood.

Counterargument: Putnam's internal realism

By 1976, Putnam had reversed himself. In Reason, Truth and History (1981) he abandoned what he called metaphysical realism — the "external" view from nowhere — for internal realism: truth is idealised rational acceptability, not correspondence to a ready-made world. His model-theoretic argument purported to show that, under reasonable assumptions, even an ideal scientific theory has multiple incompatible models that satisfy all evidential constraints — so realism's notion of "the" correspondence between theory and world is incoherent. By the late 1990s Putnam moved again, to a "natural realism" closer to common sense. The trajectory itself is instructive: a great realist tested realism's foundations and found them harder to hold than he had supposed.

Putnam's later moves drew sharp replies. David Lewis defended a robust realism, arguing that Putnam's model-theoretic argument illicitly assumed all reference is fixed by us, ignoring the world's role in selecting the intended interpretation. Saul Kripke's causal theory of reference, by tying names and natural-kind terms to baptism-and-causal-chain rather than to descriptions, gave realists a story about how reference can be stable across theoretical revolution.

Map of positions

• Scientific realism (Boyd, Psillos, Sankey) — mature theories are approximately true; their posits exist.
• Structural realism (Worrall, Ladyman) — only mathematical structure tracks reality; entities themselves may not survive.
• Entity realism (Hacking, Cartwright) — believe in what you can manipulate; remain agnostic about high theory.
• Constructive empiricism (Van Fraassen) — accept theories as empirically adequate; do not believe their unobservable claims.
• Internal realism (Putnam, middle period) — truth as idealised rational acceptability within a conceptual scheme.
• Pragmatism (James, Dewey, Rorty) — truth is what works; the realism debate is a metaphysical pseudoproblem.
• Natural Ontological Attitude (Fine 1984) — the dispute is misframed; accept what science says, but neither realism nor anti-realism adds anything to that acceptance.

Common confusions

• Realism is not the same as physicalism. A realist about numbers (Platonism) thinks abstract objects exist; a physicalist thinks only physical objects do. The two can come apart.
• Anti-realism is not relativism. Van Fraassen accepts that there is one world; he just thinks we should not commit to detailed descriptions of its unobservable parts.
• "Realism" varies by domain. One can be realist about middle-sized objects, anti-realist about colours, realist about quarks, anti-realist about possible worlds. The arguments are mostly local.
• Putnam changed his mind multiple times. The "Putnam" who proposed the no-miracles argument is not the "Putnam" who defended internal realism, who is not the late Putnam who returned to natural realism. Cite carefully.
• Anti-realism does not deny that science works. The whole point of constructive empiricism is to do justice to scientific success without committing to literal truth — empirical adequacy already entails success.