What’s striking about the Duhem–Quine thesis is: Instead of discarding your hypothesis when it fails to fit the data, you can always make broad revisions elsewhere in your web-of-beliefs, so that your collection of beliefs as a whole does fit the data even though it keeps your hypothesis. I think most people don’t understand this until they hear many examples of how scientists have done this and can do this in their work, quite irrespectively of whether their hypothesis turned out to be right (by our lights). To the end of making the thesis vivid, please give a series of examples which either actually happened historically or at least very plausibly could have happened, where a scientist opted (or could have opted) to make broad web-of-belief revisions (e.g. in philosophical or methodological assumptions) rather than discard his empirical hypothesis when the data failed to match it.
Love this question. The Duhem–Quine point really pops when you watch scientists protect a favorite hypothesis by moving other pieces of the web. Here’s a mix of real and strongly plausible cases. For each: hypothesis under pressure → the mismatch → what got revised instead (auxiliaries/methods/ontology) → how it eventually shook out.
-
Geocentrism saved by epicycles (actual)
H: Earth at rest, planets on circles. → Data: retrograde motion, varying brightness. → Revise: add deferents/equants/epicycles and tweak parameters rather than abandon geocentrism. → Worked “well enough” for centuries; replaced by heliocentrism. -
Heliocentrism saved by epicycles (actual)
H: Sun-centered cosmos. → Data: Copernicus’s circular orbits don’t fit positions precisely. → Revise: keep circles + add epicycles to rescue fits (don’t give up heliocentrism). → Kepler’s ellipses eventually remove the epicycle scaffolding. -
Tychonic compromise (actual)
H: Earth does not move. → Data: planetary phenomena push toward heliocentrism. → Revise: Sun orbits Earth, other planets orbit Sun (geo-heliocentric kinematics), preserving immobile Earth while matching observations (e.g., phases of Venus). → A serious rival until Newtonian dynamics. -
Neptune for Uranus’s odd motion (actual)
H: Newtonian gravity is right. → Data: Uranus deviates from predicted path. → Revise: postulate unseen planet as an auxiliary cause (not the law). → Neptune found—classic “save the theory by changing boundary conditions.” -
Vulcan for Mercury’s perihelion (actual)
H: Newtonian gravity again. → Data: Mercury’s extra perihelion precession. → Revise: hypothesize intra-Mercurial planet (Vulcan) or tweak solar quadrupole moment; keep Newton. → No Vulcan; Einstein’s GR explains it. -
Ether saved by Lorentz–FitzGerald contraction (actual)
H: Luminiferous ether exists. → Data: Michelson–Morley null result. → Revise: lengths contract and time dilates for moving bodies; adopt “local time,” keep ether as background. → Einstein drops the ether; the kinematical revisions remain (as relativity). -
Phlogiston with negative weight (actual)
H: Combustion releases phlogiston. → Data: metals gain mass when calcined. → Revise: say phlogiston has negative mass or that air is absorbed while phlogiston departs—rearrange ontological and chemical auxiliaries to keep the core hypothesis. → Oxygen theory supplants phlogiston. -
Caloric theory patched after friction heating (actual)
H: Heat is a conserved fluid (caloric). → Data: seemingly limitless heat from boring cannons (Rumford). → Revise: treat processes as releasing pre-stored caloric from materials or invoke hidden reservoirs/measurement caveats instead of abandoning caloric. → Kinetic theory wins with energy conservation. -
Planck’s quantization as a “fit device” (actual)
H: Classical electrodynamics/stat mech. → Data: blackbody spectrum failure (UV catastrophe). → Revise: restrict energy exchange to ε = hν for oscillators (a surgical auxiliary move) without discarding Maxwell’s theory. → The patch births quantum theory and expands into a revolution. -
Parity conservation protected by “two kaons” (actual)
H: Parity is universally conserved. → Data: the θ–τ puzzle (same mass/lifetime, different decay modes implying different parities). → Revise: treat them as two distinct particles rather than give up parity. → Wu’s experiment shows weak interactions violate parity; the auxiliary split dissolves. -
Energy conservation saved by the neutrino (actual)
H: Energy is strictly conserved in all decays. → Data: continuous β spectra look like missing energy. → Revise: postulate a light, neutral particle (Pauli’s neutrino) to carry it off; protect conservation laws. → Neutrino detected; auxiliary becomes new ontology. -
Static universe saved by Λ (actual)
H: The cosmos is static. → Data: Einstein’s field equations predict dynamic universes. → Revise: add the cosmological constant to hold it static. → Expansion observed; Λ later returns, but as dark energy for acceleration, not to save stasis. -
Solar neutrino problem (actual)
H: Standard Solar Model or standard weak interaction picture (pick one). → Data: deficit of detected solar ν_e. → Revise (two paths): astrophysicists tweak solar opacities/mixing; particle physicists allow flavor oscillations (neutrino mass). → Oscillations confirmed; the “particle” auxiliary won. -
Galaxy rotation curves and the mass discrepancy (actual)
H: Newton/GR are right on galactic scales. → Data: flat rotation curves, cluster dynamics. → Revise: add nonluminous dark matter to the ontology (plus new structure-formation priors) rather than modify gravity. → DM remains standard; MOND/modified gravity is the competing way to shift the web. -
Accelerating universe (actual)
H: GR + matter-only cosmos. → Data: Type Ia supernovae too dim at redshift. → Revise: re-admit Λ or a dark-energy field; adjust cosmological priors and parameterization (not GR’s core structure). → ΛCDM becomes concordance. -
Land bridges vs. drifting continents (actual)
H: Continents are fixed. → Data: biogeography and fit of continental margins. → Revise: hypothesize ephemeral land bridges, complex subsidence histories, and methodological cautions about reconstruction fidelity—keep fixism. → Plate tectonics eventually reframes the whole web. -
Stokesian ether drag (actual)
H: Ether exists. → Data: stellar aberration (Bradley) seems hard for a static ether. → Revise: let matter partly drag the ether (with carefully tuned indices) to reconcile aberration and other optics. → Abandoned with relativity. -
“Instrument first” patches (very plausible, many real episodes)
H: A favored empirical claim (e.g., superluminal neutrinos, early CMB B-modes). → Data: surprising signal. → Revise: recalibrate clocks, fiber connections, dust foreground models, pipeline assumptions—methodological/instrumental auxiliaries—before giving up relativity or ΛCDM. → In both famous cases, auxiliary fixes solved the anomaly. -
Precession without GR (plausible counterfactual)
H: Newtonian gravity is exact. → Data: Mercury anomaly again. → Revise (if GR hadn’t arrived): adopt distance-dependent G, or a small r^−3 correction from Solar oblateness as a systematic, or a distribution of dust torque—keeping Newton while shifting background astrophysical assumptions. → Would have postponed a theory change. -
“Hidden variables” to save classical intuitions (plausible/historically explored)
H: Physics is local and deterministic. → Data: EPR/Bell-type violations. → Revise: widen the ontology (nonlocal hidden variables, superdeterminism), or alter methodological canons about what counts as an acceptable causal explanation; preserve the target hypothesis about determinism. → Mainstream accepts standard QM with nonlocal correlations; alternatives persist on the margins.
How these illustrate Duhem–Quine
- The empirical hit lands on a bundle: the core hypothesis plus approximations, ceteris-paribus clauses, boundary conditions, modeling choices, data-reduction pipelines, instrument models, even metaphysical commitments (e.g., determinism, symmetry principles).
- In every case, one can keep the target hypothesis by shifting weight elsewhere—adding entities (Neptune, neutrino, dark matter), adjusting kinematics (Lorentz contraction), revising cosmological priors (Λ), or altering methodology (foreground subtraction, calibration).
- Whether such “saves” are good science is judged retrospectively by fruitfulness, unification, independent tests, and simplicity—not by logic alone.
If you want, I can turn these into a one-page handout with columns for “core hypothesis / anomaly / auxiliary change / independent check” or add more cases in specific fields (biomed, econ, CS).
No comments:
Post a Comment