QuestionsCan science explain everything?An experiment never condemns a single hypothesis: Duhem's holism
Can science explain everything?A human construction

An experiment never condemns a single hypothesis: Duhem's holism

The Duhem thesis (or Duhem-Quine thesis): confirmation holism

Pierre Duhem 1861 · 1916A physics experiment never tests a hypothesis alone, but a whole set of theories.
See profile
What it claims

A physics experiment can never condemn an isolated hypothesis, only a whole set; there is no crucial experiment; and the choice of the hypothesis to give up is a matter of the physicist's good sense.

Why

A prediction relies on a whole set of hypotheses: when it fails, logic does not say which one is wrong.

The same idea through history
Pierre Duhem1906Duhem, a physicist and historian of physics, shows from optics and mechanics that the physicist never tests an isolated hypothesis. Foucault's experiment (1850), often presented as a verdict against Newton's emission theory of light, in fact condemns Newton's whole system, not only the hypothesis that light is made of corpuscles.
Willard Van Orman Quine1951In "Two Dogmas of Empiricism", Quine generalises: our statements face experience as a body, and any of them can be held true if the rest is adjusted enough. Quine refers to Duhem for this idea; it has since been called the "Duhem-Quine thesis", but Quine's version goes further than Duhem's, who spoke of physics.
In short

"An experiment never condemns a single hypothesis: Duhem's holism" is an argument by Pierre Duhem (The Aim and Structure of Physical Theory (La Théorie physique), 1906, part II, ch. VI, 1906) on the question "Can science explain everything?" (answer: a human construction). Its claim: A physics experiment can never condemn an isolated hypothesis, only a whole set; there is no crucial experiment; and the choice of the hypothesis to give up is a matter of the physicist's good sense. According to AskPhilosophy's logic check (Z3 solver), the conclusion does follow from its 5 premises. AskPhilosophy presents 2 objections, 1 of them with a documented reply, and the original text behind each premise, checked word for word.

The Aim and Structure of Physical Theory (La Théorie physique), 1906, part II, ch. VI
Valid

We often imagine that an experiment "tests" an idea as one tests a part: if it breaks, the idea is false. Pierre Duhem, a French physicist, shows at the beginning of the twentieth century that it is not so simple. To predict what an experiment should give, the physicist uses not only the hypothesis he wants to test: he also uses the theory of his instruments, optics, calculations, etc. When the result differs from the prediction, something in the whole is therefore false, but the experiment does not say what. Duhem draws three consequences: no experiment condemns a hypothesis on its own; there is no "crucial experiment" that would decide between two rival hypotheses; and the choice of what to change depends on the physicist's "good sense", which is not a rule of logic. We list this argument under "A human construction" because it shows that human choices enter the decision, even in physics. Duhem thinks that a day comes when good sense clearly sorts out the theories (§ X). He speaks of physics, not of all science.

P1To predict the result of an experiment, the physicist always uses, besides the hypothesis he wants to test, a whole set of other hypotheses (the theory of his instruments, the laws of optics, etc.).
P2When the result contradicts the prediction, the experiment tells us that at least one of the hypotheses in the set is false, but it does not say which: logic alone does not say what to change.
P3 · implicit premiseAn experiment that does not point to the faulty hypothesis cannot condemn an isolated hypothesis.
P4A crucial experiment (one that decides between two hypotheses by refuting one) would require having listed all possible hypotheses for a group of phenomena.
P5The physicist is never sure of having exhausted all the imaginable hypotheses.
P6Since logic does not say what to change, no absolute principle guides this choice: different physicists can choose differently without being illogical. What guides them is what is called "good sense".
So a physics experiment can never condemn an isolated hypothesis, only a whole set; there is no crucial experiment; and the choice of the hypothesis to give up is a matter of the physicist's good sense.

click a premise to reject it

If you accept every premise, logic forces you to accept the conclusion. So the debate is about the premises.
Checked by a logic solver: the premises cannot all be true while the conclusion is false.
Without the implicit premise: Without P3, one could accept P1 and P2 and say that a well-designed experiment still isolates the tested hypothesis, because the other hypotheses are so solid that they can be taken as certain. Duhem recognises that this is what the physicist does, but he says this confidence is not "imposed by logical necessity".
The argument is valid. P1 and P2 say that the experiment does not point to the faulty hypothesis. With P3 (implicit), no isolated hypothesis is condemned. P4 and P5 show there is no crucial experiment. P6 says that, when logic does not name the culprit, good sense decides. The debate is about P2 and P3: in practice, can one find which hypothesis is at fault, and does the fact that one can always save a hypothesis suffice to say it can never be refuted?
Analogy. A car will not start. The mechanic cannot say "it is the battery" from the mere fact that it will not start: it may also be the starter, the fuel, a fuse. The fact points to nothing by itself; it is his experience, not a rule, that suggests where to begin.

Objections

2

Each objection is a conversation: click to see who replies and where the debate stands.

Karl Popper 1963 · attacks premise 2

Popper answers that the "holistic" argument goes much too far. It is possible, in quite a few cases, to find which hypothesis is responsible for the refutation.

a defender replies
see the argument
I
Imre Lakatos defender of the argument · 1970
Lakatos draws Duhem's lesson for method: no experiment, no observation statement can alone lead to refutation, and there is no refutation before a better theory appears. A theory is thus not rejected because an experiment contradicts it, but because another does better.
The debate goes on (our summary) Popper himself admits, in the same chapter, that often only a large chunk of a theoretical system can be tested, and sometimes the whole system; naming the responsible element is then sheer guesswork (he refers to Duhem: see his quote in Popper's argument). The disagreement is about how often one can find which hypothesis is at fault, and about the weight method should give to "good sense".
Where the debate stands (our summary)Our summary: everyone now admits that tests depend on auxiliary hypotheses. The core disagreement remains about what follows: for Popper, science still progresses by refutation; for Duhem and Lakatos, criteria beyond logic must be added (good sense, comparison between programmes).
A
Adolf Grünbaum 1960 · attacks premise 3

Grünbaum denies that logic proves what Duhem draws from it. That one can always save a hypothesis by changing the others is not proven: one still has to find alternative hypotheses that agree with all the results already established, and nothing guarantees that they exist. It is a question of fact, to be examined case by case, not a consequence of logic.

no reply
Duhem never answered this objection.
The debate goes on (our summary) Duhem shows on examples that a hypothesis can be saved by changing others (Neumann's case), without asserting it for all cases (our reading). The holists (including Quine) go further than Duhem in saying that any hypothesis can be kept if the rest is adjusted enough; it is this strong version that Grünbaum mainly targets.
Where the debate stands (our summary)Our summary: a weak version (tests depend on auxiliary hypotheses, admitted by all) is now distinguished from a strong version (any hypothesis can always be saved, much debated). Duhem rather defends the first.