# Falsifiability

> A claim is falsifiable when some possible observation could count against it, though real tests examine a bundle of assumptions.

- HTML version: https://robbiepalmer.me/ideas/falsifiability
- Source: https://plato.stanford.edu/entries/scientific-method/#PopFal

Falsifiability asks whether a claim takes a genuine risk. A claim is falsifiable
when we can describe a possible observation that would conflict with it. Karl
Popper used this condition to distinguish scientific claims from claims that
can accommodate every outcome.

The core argument rests on an asymmetry. Many successful predictions cannot
deductively prove a universal claim, but one genuine counterexample can conflict
with it. Seeing a thousand white swans does not entail that every swan is white.
Finding one black swan is enough to refute that unrestricted statement.

Falsifiability is not the claim that a failed prediction automatically tells us
which belief to discard. That is where a clean logical principle meets the
messier practice of inquiry.

## Make the risk visible

A useful test states the claim, the expected observation, and the outcome that
would count against it before the result is known. "This treatment always
prevents the reaction under these conditions" takes a clear risk. "The
treatment works unless an unknown influence blocks it" can absorb any failure
and so offers a weaker test.

Popper called a theory's possible counterexamples its potential falsifiers. A
claim with more potential falsifiers says more and is easier to test. Surviving
a demanding test corroborates the claim. It does not establish that the claim
is true or assign it a probability of truth.

This distinction guards against two common mistakes. Confirming evidence is not
proof, and an explanation that fits every result may explain very little. Good
inquiry looks for observations that could expose an error rather than collecting
only friendly cases.

## A test examines a bundle

Pierre Duhem argued that a physical hypothesis is not tested alone. A prediction
also relies on auxiliary hypotheses about instruments, initial conditions,
background laws, and the way the experiment was performed. If the prediction
fails, the conjunction has failed. Logic does not identify the faulty member.

For example, a key that does not open a lock may show that the key is wrong. It
may instead show that the lock was replaced, the key was cut badly, or the
mechanism is jammed. The trial tests a bundle of claims about the key, the lock,
and the procedure. Repeating the test or inspecting the mechanism can narrow the
fault, but the first failure does not name it.

W. V. O. Quine extended the problem beyond physics. Duhem argued that a failed
experiment reveals an error somewhere among the hypothesis, background theory,
instruments, and procedure. Quine treated statements about the external world
as a connected body. On his account, a statement can survive contrary evidence
if other commitments change. The name "Duhem-Quine thesis" hides that
extension.

This limits simple falsification. A contrary result conflicts with the tested
system. Further experiments and judgement locate the error and determine what
to revise.

## Revision without evasion

Revising an auxiliary assumption can produce a better test. When the observed
orbit of Uranus conflicted with existing predictions, adding an unknown outer
planet predicted where astronomers should look. They found Neptune there.

An amendment earns its place when it explains the failure, predicts another
observable result, and applies beyond the case that prompted it. A patch that
shields a favoured claim from every contrary result reduces its falsifiability.

[Theory-ladenness](/ideas/theory-ladenness) deepens the problem because evidence
itself depends on assumptions. [Postpositivism](/ideas/postpositivism) responds
with fallibility, open criticism, and attempts to detect error. The aim is not a
mechanical verdict. It is a disciplined way to learn which part of an account no
longer deserves confidence.

## Sources

* Hepburn, Brian, and Hanne Andersen. ["Scientific
  Method"](https://plato.stanford.edu/archives/spr2025/entries/scientific-method/#PopFal).
  *The Stanford Encyclopedia of Philosophy*, 2025.
* Popper, Karl. *The Logic of Scientific Discovery*. Hutchinson, 1959.
* Ariew, Roger. ["Pierre
  Duhem"](https://plato.stanford.edu/archives/spr2025/entries/duhem/#AgaNewMetDuhThe).
  *The Stanford Encyclopedia of Philosophy*, 2025.
* Quine, W. V. O. ["Two Dogmas of
  Empiricism"](https://doi.org/10.2307/2181906). *The Philosophical Review* 60,
  no. 1, 1951.

## Related ideas

- [Epistemology](https://robbiepalmer.me/ideas/epistemology.md): How we know what we claim to know, what justifies a belief, and when new evidence should change it.
- [Postpositivism](https://robbiepalmer.me/ideas/postpositivism.md): Empirical inquiry can pursue objectivity while accepting fallible evidence, theory-laden observation, and the influence of values.
- [Theory-ladenness](https://robbiepalmer.me/ideas/theory-ladenness.md): Observations depend on prior ideas about what to notice, how to measure it, and what the result means.

## Where it appears

- Blog post: [The Philosophy of Data Science](https://robbiepalmer.me/blog/2022-03-02-the-philosophy-of-data-science.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt