Falsifiability (Popper)

The criterion stated

Popper's claim has two parts that are often confused. First, the demarcation thesis: a statement counts as scientific if and only if it is, in principle, falsifiable — there exists some conceivable empirical observation that would force its rejection. Second, the methodological prescription: scientists should propose bold, content-rich conjectures and then attempt seriously to refute them, treating survival under sincere attack as the only respectable kind of corroboration.

Both parts are responses to a problem Popper inherited from David Hume. Hume had shown that no finite collection of observations can deductively establish a universal law: the sun has risen on every observed morning, but no purely logical argument carries us to "the sun will rise tomorrow". The logical positivists of the Vienna Circle — Carnap, Schlick, Neurath — tried to rescue science by softening "verify" to "confirm" and proposing a verifiability criterion of meaning. Popper rejected this whole strategy. Look at the logical structure, he said: a universal claim like all swans are white is consistent with any number of confirming instances, but a single non-white swan refutes it. The asymmetry is the key. Science cannot prove its theories true, but it can prove them false — and that is enough to make progress.

Where the idea came from

Popper, born in Vienna in 1902, watched the post-1918 collapse of European certainties up close. He was briefly a Marxist as a teenager, knew Adler personally, and lived in a city saturated with Freudian thought. What struck him was how each of these systems seemed to confirm itself everywhere. A worker who supported the bourgeoisie was suffering false consciousness; a worker who didn't was developing class awareness; either way, Marxism was vindicated. A mother saving her child was sublimating; a mother harming her child was repressing; either way, psychoanalysis was confirmed. These theories were rich, sophisticated, and — Popper began to suspect — empirically empty.

The contrast came in 1919. Einstein's general relativity predicted that starlight passing close to the Sun would deflect by 1.75 arcseconds — twice the Newtonian prediction. Arthur Eddington's eclipse expedition to Príncipe and Sobral measured the deflection. Had it come back at zero, or at the Newtonian half-value, general relativity would have been dead. It risked something. That risk-taking, Popper thought, was the mark of genuine science.

Worked example: predicting an eclipse

Consider three claims about gravity:

1. Astrology — your fate is shaped by the configuration of planets at your birth.
2. Newtonian gravity — masses attract with force F = GMm/r².
3. General relativity — masses curve spacetime; geodesics determine motion and light bends in gravitational fields.

Astrology, when probed, retreats into vagueness: charts are interpretive, predictions hedged, failures explained by missing data. Nothing observable would lead an astrologer to declare astrology refuted. Newtonian gravity, by contrast, predicts a specific stellar deflection of 0.87 arcseconds during a solar eclipse — measurable, finite, refutable. General relativity predicts 1.75 arcseconds. Eddington's photographs in 1919 measured roughly 1.6 arcseconds — close enough to GR, fatally far from Newton's value at the precision available, and infinitely far from astrology, which made no quantitative prediction at all. Popper's point: theories 2 and 3 are scientific because they could have lost. Theory 1 is not, because it could not.

Falsifiability vs other demarcation proposals

Criterion	Champion	Mark of science	Trouble
Verificationism	Carnap, Schlick (1920s–30s)	Empirically verifiable in principle	Universal laws can never be fully verified
Falsifiability	Popper (1934)	Empirically refutable in principle	Duhem-Quine; theories never tested in isolation
Paradigm normal science	Kuhn (1962)	Puzzle-solving within accepted framework	Looks descriptive, not normative; what about pseudoscience?
Research programmes	Lakatos (1970)	Progressive problem-shifts; novel predictions	"Progressive" assessable only in hindsight
Anything goes	Feyerabend (1975)	No rules; methodological pluralism	Loses the ability to call astrology pseudoscience
Bayesian confirmation	Howson, Urbach (1989)	Updates priors in light of evidence	Subjective priors; no sharp pseudoscience demarcation
Cluster account	Hansson (2013)	Multiple imperfect markers, no single criterion	Vague; concedes Popper's project

The Duhem-Quine challenge

Pierre Duhem in The Aim and Structure of Physical Theory (1906) and W. V. O. Quine in "Two Dogmas of Empiricism" (1951) made what is now the standard objection. No hypothesis is ever tested alone. To test "light bends near the Sun" you need a theory of optics, of telescopes, of photographic emulsion, of atmospheric refraction at high altitude, of stellar parallax, and a great deal more. If the prediction fails, logic alone cannot tell you which of these many assumptions is the guilty party. You can always save the target hypothesis by sacrificing an auxiliary one.

Popper recognized the problem and called the move that rescues a theory by ad-hoc auxiliary modification a "conventionalist twist". He insisted that scientists ought to refuse such moves — should agree in advance which assumptions are protected and which are at risk. But "ought" is not "must", and critics argued that real scientific practice is shot through with such rescues. The Michelson-Morley null result was first explained away by Lorentz contraction; the perihelion of Mercury was patched onto Newton for half a century before being solved by general relativity. The history of science contains plenty of fertile evasions and plenty of barren ones, and falsificationism alone cannot tell them apart in the moment.

Lakatos's repair: research programmes

Imre Lakatos, a student of Popper's, tried to combine Popper's insights with Kuhn's. A "scientific research programme" has a hard core of central commitments treated as irrefutable by methodological decision, and a protective belt of auxiliary hypotheses that absorb refutations. Newtonian mechanics treated F=ma and the inverse-square law as the hard core; the protective belt included claims about how many planets exist, what the solar system's geometry is, and so on. The programme is progressive when modifications to the belt yield novel predictions that turn out true (Newton's belt produced Neptune); it is degenerative when modifications are mere face-saving (Newton's belt also produced unsuccessful Vulcan). Lakatos's criterion shifts the demarcation from individual theories to whole programmes assessed over time.

Variants of falsifiability

• Naïve falsificationism — one disconfirming observation kills a theory. Popper himself never held this; it is the cartoon version.
• Sophisticated falsificationism — a theory is rejected only when a better-corroborated rival replaces it. Closer to Popper's actual position.
• Methodological falsificationism (Lakatos) — researchers decide by convention which statements count as observational and which auxiliaries are protected; refutation is then sharp.
• Bayesian falsification — replace strict refutation with probability shifts: a failed prediction lowers posterior, a passed one raises it. Loses the asymmetry but tracks scientific practice better.
• Degrees of falsifiability — Popper himself proposed measuring a theory's content by how many possible observations it forbids. More improbable theories have more content if they survive.

Counterargument: Kuhn's normal science

Thomas Kuhn's Structure of Scientific Revolutions (1962) accepted that revolutions look something like falsification but denied that this is what scientists do most of the time. The bulk of scientific work, Kuhn argued, is "normal science" — puzzle-solving within an accepted paradigm, where anomalies are absorbed, postponed, or attributed to experimenter error rather than treated as refutations. The Mercury anomaly stayed in the literature for decades while astronomers searched for Vulcan rather than abandoning Newton. This was not bad science but rational behaviour: paradigms have track records, and sane researchers do not throw away a working framework on the first anomaly. Falsification, on Kuhn's view, is the exception (during revolutions), not the rule.

Popper's reply was that Kuhn was describing what scientists do, not what they should do, and that good science is precisely the kind that resists this conservatism. The disagreement is partly empirical (how do scientists actually behave?) and partly normative (what counts as good science?), and it remains live.

Common confusions

• Falsifiable does not mean false. A falsifiable theory might be true — it just must be the kind of claim that could in principle be shown false.
• Falsifiable does not mean easily falsifiable. Some scientific claims (the existence of the Higgs boson) require billion-dollar instruments to test. Costly is fine; impossible-in-principle is not.
• Unfalsifiable does not mean meaningless. Popper rejected the positivists' move from unfalsifiable to nonsense. "There is one unicorn somewhere in the universe" is unfalsifiable but perfectly meaningful.
• The criterion targets statements, not people. A psychologist using Freud may well make falsifiable predictions in their own lab work; the question is whether the theory itself, in its general form, forbids anything.
• Probabilistic theories are still falsifiable. "Half the births in this hospital next year will be female" is refutable by an observed extreme deviation; the bar is statistical, not deterministic.