A Theory Has to Forbid Something

Karl Popper spent the 1920s in Vienna surrounded by three theories that each explained everything.

Marxism explained every political event, including the ones that had been predicted to go the other way. Freudian psychoanalysis explained a man who pushed a child into a river and a man who dived in to save one, with equal facility. Adler's individual psychology did the same. Their supporters treated this universal applicability as the great strength of the theories, and Popper began to suspect it was the opposite.

Then in 1919 Arthur Eddington's expedition photographed a solar eclipse.

Einstein's general relativity predicted that starlight passing close to the Sun would be deflected by a specific amount, roughly twice what Newtonian gravity implied. The number was fixed in advance, with no scope for adjustment afterwards. If the plates had shown the Newtonian value, or no deflection, the theory was finished. Einstein had staked everything on a photograph.

That, Popper decided, was the difference. Not that the astronomers had more evidence, but that they had exposed themselves to defeat. A theory compatible with every possible observation makes no claim about the world. It cannot be surprised by anything, and so nothing it survives counts in its favour. A theory that forbids something is telling you what you will not see, and every occasion on which you fail to see it is a test it could have failed.

So the criterion is falsifiability: a scientific claim rules out some observable state of affairs. The asymmetry is logical. No number of white swans establishes that all swans are white; one black swan settles it. Confirmation accumulates without ever reaching certainty; refutation is decisive.

Which raises the obvious objection, and Popper knew it. Theories are never refuted alone. A telescope result depends on the telescope, the atmosphere, the calculations, and a dozen auxiliary assumptions, and any of these can be blamed instead. When Uranus failed to follow its Newtonian orbit, astronomers did not abandon Newton; they proposed an unseen planet, and Neptune was duly found more or less where the calculation said. When Mercury's orbit failed in the same way, they proposed another planet, called it Vulcan, and it was not there; Mercury needed general relativity.

Same move, two outcomes. What distinguishes them is not whether the theory was rescued but whether the rescue risked anything: Neptune was a new, checkable prediction, and Vulcan was too, and one of them failed. An ad hoc hypothesis in the pejorative sense is one that saves the theory while forbidding nothing new: an addition that absorbs the anomaly and makes no further claim.

Thomas Kuhn then made the historical case that this is not how science actually proceeds day to day, and he was largely right. Practising scientists do not abandon a framework the moment something contradicts it, and they should not: anomalies are everywhere, most turn out to be problems with the apparatus or the application, and a theory that was dropped at the first discrepancy would never survive long enough to be developed. Normal science, in Kuhn's description, is puzzle-solving within a paradigm that supplies the questions, the methods and the standards for what counts as an answer.

A paradigm gives way not when it is refuted but when a rival exists that handles the accumulated anomalies better. That is a comparative judgement made by a community, and it can take a generation.

The two accounts are often presented as opponents, and it is more useful to treat them as answering different questions. Popper describes what makes a claim testable: a question about logic. Kuhn describes what communities of scientists do: a question about history. A field can be thoroughly Kuhnian in practice while falsifiability remains the thing that separates its claims from astrology's.

And the practical residue is worth carrying. When you meet a claim, ask what would have to be observed for it to be wrong. If the answer is nothing (if every outcome is accommodated, if failures are explained by the presence of sceptics or the wrong conditions or insufficient belief) then whatever the claim is doing, it is not making a statement about what you will observe. That is not the same as being false. It is worse, in a way: there is nothing there to be false.