TL;DR: Falsifiability is a standard proposed by Karl Popper in 1934 in Logik der Forschung (the expanded English edition of 1959 is The Logic of Scientific Discovery): a theory is considered scientific only if its logical structure allows pointing to an observation that could contradict it. Popper used astrology and psychoanalysis as examples of systems that explain everything, therefore cannot be wrong, therefore do not provide useful predictive information. This standard has been highly influential, was once cited in Daubert v. Merrell Dow Pharmaceuticals (1993), but has also been criticized for ignoring background assumptions in practical testing.
What is falsifiability?
Falsifiability is the standard for evaluating scientific statements, including theories and hypotheses. A statement is falsifiable if it belongs to a language or logical structure capable of describing an empirical observation that contradicts it. For a theory, falsifiability requires that, given an initial condition, the theory must theoretically exclude certain observations, meaning it must make formal predictions.
The concept was introduced by the philosopher of science Karl Popper in Logik der Forschung (1934), with an expanded English edition in 1959 titled The Logic of Scientific Discovery. Popper stressed that the contradiction lies within the logical structure itself, without needing to appeal to methodological considerations outside that structure. He proposed falsifiability as the cornerstone solution to both the problem of induction and the demarcation problem, that is, the boundary between science and non-science.
Why did Popper take astrology as an example?
Popper recounted his encounter with psychoanalysis in the 1910s, especially with Alfred Adler, in chapter 1, 'Science: Conjectures and Refutations', of Conjectures and Refutations (London: Routledge & Kegan Paul, 1963), a lecture given at Peterhouse, Cambridge in 1953. No matter what observation one brought forward, psychoanalysis could explain it. The reason it could explain everything was that it excluded nothing. Popper held that this was a failure, because it meant the standard could not lead to any prediction.
In the same chapter, Popper contrasted psychoanalysis with Einstein's theory of relativity, and the benchmark he named explicitly was the year 1919. With Einstein, a risky prediction was made in advance and could fail; with systems such as psychoanalysis or astrology, every outcome was maneuvered into place.
Astrology belongs to this same group: it can be adjusted to fit any outcome. A failed prediction can be rescued by appealing to a planet not yet considered, an inaccurate birth time, or a different interpretation. Precisely because no observation is excluded in advance, the system takes no risk, and therefore carries no information.
From a logical standpoint, observations that do not contradict a law do not mean the law is true. A confirmation has no value in itself. But if a hypothesis makes risky predictions and those predictions are borne out, Popper said, that is a reason to prefer this hypothesis over others that make less risky predictions or none at all.
The logical mechanism: modus tollens and asymmetry
The problem of induction is often called the problem of Hume. David Hume studied how human beings acquire new knowledge that goes beyond known laws and observations. He understood that deductive logic could not explain this learning process and argued in favor of a psychological learning process that does not require deductive logic. Popper accepted Hume's argument and therefore viewed progress in science as the result of something close to induction, that is, induction without a rule of inference, which he also called "the way of science".
Popper's idea for solving this problem was: while it cannot be verified that all swans are white, finding one black swan shows that not all swans are white. Such falsification uses the valid inference of modus tollens: if Q can be deduced from law L, but ¬Q is observed, then L is false.
More concretely, the deductive statement yielding Q can be split into an initial condition and a prediction, in the form C ⇒ P, where C is "the object here is a swan" and P is "the object here is a white swan". If C is observed to be true while P is false, that is C ∧ ¬P, then the law is false by modus tollens.
The asymmetry here is clear: verifying a universal statement such as "all swans are white" requires observing every swan, which is practically impossible; whereas a single black swan is enough to refute it. Popper contrasted falsifiability with the concept of verifiability then prevalent in logical positivism.
Demarcation and meaning: where do science and non-science differ?
Popper said that some philosophers of the Vienna Circle conflated two problems: the problem of meaning and the problem of demarcation. They proposed in the verification theory a single solution to both: a statement that cannot be verified is considered meaningless. Popper held instead that there exist theories that are meaningful but non-scientific, and therefore the criterion of meaning does not coincide with the criterion of demarcation. He set out this argument in chapter 1, 'Science: Conjectures and Refutations', of Conjectures and Refutations (London: Routledge & Kegan Paul, 1963).
Methodologically, Popper proposed choosing the hypothesis that has been most severely tested: "the hypothesis which, in the light of our critical discussion, appears to be the best so far". In his words, since only the negative approach is supported by logic, Popper applied a negative methodology to prevent "the policy of immunizing our theories against refutation".
Table comparing the key conceptual pairs
| Conceptual pair | Content | Consequence |
|---|---|---|
| Falsifiability and verifiability | Falsification: a single contradictory observation is enough to refute. Verification: every case must be observed to affirm | Falsification is logically feasible; universal verification is practically impossible |
| Logical criterion and methodology | Falsifiability is a logical criterion within an empirical language. Falsificationism is a methodological approach | Popper said falsifiability should not be equated with falsificationism |
| Falsification and empirical refutation | Falsification points to the logical structure that permits contradiction. Empirical refutation involves background assumptions | The Duhem–Quine thesis holds that decisive empirical falsification is impossible |
| Science and non-science | Science makes risky predictions that can fail. Non-science can explain everything | Astrology and psychoanalysis belong to the latter group according to Popper |
How does the Duhem–Quine thesis push back?
The Duhem–Quine thesis holds that decisive empirical falsifications are impossible, and that no scientific hypothesis by itself can make predictions, because an empirical test of a hypothesis requires background assumptions. This is the presentation in the entry Underdetermination of Scientific Theory in the Stanford Encyclopedia of Philosophy (Kyle Stanford). The acceptance of those background assumptions involves methodological decisions within Lakatos's falsificationism.
Popper's response was that falsifiability is a logical criterion. Empirical research has the Duhem problem and other problems, such as the problem of induction, but according to Popper, logical induction is a fallacy, and statistical tests, which are only feasible when a theory is falsifiable, are nonetheless useful in critical discussion.
Popper's distinction between logic and methodology does not help falsifiability escape certain criticisms aimed at methodology. For example, Popper's rejection of Marxism as non-scientific because it resists negative evidence is a methodological stance, but the problems of that stance are still presented as a limitation of falsifiability.
Induction, probability and a different conception of induction
In practice, some observation-based steps can be justified under certain assumptions. For instance, Bayesian inductive logic is justified by theorems that state the assumptions. These theorems are obtained by deductive logic. They are sometimes presented as supporting inductive steps, because they appeal to the laws of probability, though they do not go beyond deductive logic. This is a different conception of induction, overlapping with deductive logic in the sense that it is supported by it. Hume's argument does not refute the possibility of a general hypothetical procedure for explaining the progress of science, but it says that the problem of choosing initial hypotheses and proving their correctness generates an infinite regress.
Philip N. Johnson-Laird agrees with Hume that there can be no general method of justification for induction, but induction does not need justification. Instead, these steps use inductive patterns, which are not expected to have a general justification: they may or may not apply depending on context. Here there is one detail I have not been able to verify: some secondary sources attribute to Johnson-Laird a sentence to the effect that induction is simply what animals, including humans, do to make life possible, but when I checked the paper A model theory of induction (International Studies in the Philosophy of Science 8(1), 1994) I could not find that sentence, so I will set it aside and not attribute it to him.
Popper accepted the possibility of a psychological explanation for the learning process, especially when psychology is regarded as an extension of biology, but held that biological explanations do not belong to epistemology. He proposed an evolutionary mechanism to explain the success of science, but did not regard it as part of his epistemology, and called the opposing view psychologism.
Practical applications and multi-criteria standards
Falsifiability, as a key concept in separating science from non-science and pseudoscience, has figured prominently in many debates and applications. It has been used as a legal precedent, specifically in Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), where the U.S. Supreme Court named falsifiability as a factor in considering the admissibility of expert testimony. In the Stanford Encyclopedia of Philosophy entry on science and pseudoscience, the problem of distinguishing science from pseudoscience is placed within the larger task of determining which beliefs are epistemically warranted.
Practical fields that need this standard include: climate policy, where the scientific consensus on human-caused climate change leaves no room for reasonable doubt; healthcare, where pseudoscience leads to ineffective and sometimes dangerous interventions; environmental policy; expert testimony in court; science education, especially in the face of efforts to introduce creationism into the curriculum; and journalism, where there is a need to distinguish legitimate scientific debate from efforts to promote pseudoscientific views as though they were science.
Because many criteria have been proposed, some authors take a multi-criteria approach, that is, using a list of criteria rather than a single one. A notable point is that there is more agreement on specific cases than on the general criteria on which that judgment should rest.
Where does Popper place falsifiability in the overall picture?
One distinction must be made clearly: falsifiability differs from falsificationism, that is, the methodological approach in which scientists actively seek evidence to refute a theory. It also differs from Lakatos's falsificationism. The purpose of falsifiability is to make a theory predictive, testable and useful in practice.
In other words, Popper did not require every scientist to constantly try to refute their own theory. He required that the theory be stated in such a way that, in principle, there is an observation that could contradict it. A guess that cannot be wrong excludes nothing, and therefore says nothing that can be tested.
Is falsifiability the same as falsificationism?
No. Falsifiability is a logical criterion about the structure of a statement, whereas falsificationism is a methodological stance about how a scientist ought to work. Popper said the two should not be equated.
A theory can be falsifiable without anyone being obliged to go looking for counterexamples to it right away. What the standard requires is only this: when stating the theory, one must specify which observation, if it occurred, would make the theory false.
Has falsifiability been completely refuted?
Not exactly. The strongest criticism comes from the Duhem–Quine thesis: an empirical test never tests a single hypothesis alone, but tests a whole cluster of hypotheses and background assumptions, so decisive falsification is impossible in practice.
Popper answered that falsifiability is a logical criterion, not a description of laboratory practice. The criterion therefore remains useful as a way of arranging theories by degree of risk and degree of testability, even though it is not enough on its own to demarcate science from non-science.
Why does a guess that cannot be wrong say nothing?
Because it excludes no outcome whatsoever. A statement carries information only when it says that the world will not turn out in certain ways.
If every observation is maneuvered by the system into place, then no matter how the world is, the statement remains true. In that case it does not help us predict, does not help us distinguish one situation from another, and poses no risk by which it could be tested. That is precisely the point Popper used to separate science from systems that can explain everything.
