The Falsifiability Criterion and Its Problems

Core Idea

While falsifiability offers an attractive demarcation criterion, it faces significant challenges. Scientific theories are often protected by auxiliary hypotheses that absorb anomalies without refuting the core theory. How does a single refutation falsify a theory if we can blame measurement error instead? Additionally, some scientific theories (evolutionary theory, quantum mechanics) seem flexible enough to avoid falsification while remaining central to science. These problems motivated Lakatos's account of research programs and Kuhn's rejection of falsificationism.

Explainer

From your study of Popper's falsificationism, you know the elegant core argument: a theory earns its scientific status by sticking its neck out. It must make predictions that could turn out to be wrong. If it is unfalsifiable — if it can accommodate any possible observation — it tells us nothing about the world. This gives us the demarcation criterion: science consists of genuinely falsifiable claims; metaphysics and pseudoscience do not. But the criterion runs into deep trouble when we try to apply it to actual science, because real scientific theories are never tested in isolation.

The central problem is the Duhem-Quine thesis. When you test a scientific prediction, you are not testing a single hypothesis — you are testing a whole bundle of claims: the core theory, auxiliary hypotheses about measuring instruments, background assumptions about the experimental setup, and much more. When an experiment goes wrong, logic alone cannot tell you which element of this bundle is false. Le Verrier and Adams predicted the existence of Neptune by invoking Newton's laws plus the hypothesis that they accounted for all massive bodies — when Uranus's orbit was anomalous, they added Neptune rather than abandoning Newton. They were right. But the same move can always be made: any anomaly can be absorbed by adjusting an auxiliary hypothesis rather than abandoning the core theory. Falsification is therefore never logically compelled by data alone.

This creates what philosophers call the problem of ad hoc modifications — changes made not because they reveal new structure but purely to protect the core theory from refutation. The worry is that sufficiently creative scientists can insulate any theory from falsification. Popper was aware of this and tried to distinguish legitimate theoretical development from dishonest ad hoc patching, but the line is notoriously difficult to draw in practice. When Freudian analysts explain both a patient's submission and a patient's aggression as expressions of the same repressed complex, they are pattern-matching post hoc in a way that generates no new testable predictions. But when physicists explain anomalous particle behavior by positing dark matter, they too are invoking an unobserved entity to save a theory — and they may well be right.

Some of science's best-confirmed theories also behave strangely with respect to falsifiability. Evolutionary theory explains why organisms are as they are, but it is notoriously difficult to specify in advance what would decisively refute it — almost any fossil discovery or behavioral adaptation can be incorporated. Quantum mechanics predicts probabilistic distributions, so any single outcome is compatible with the theory; only systematic deviations across many trials could falsify it. These are not weaknesses of these theories — they are among the best-supported claims in all of science. Yet they fit Popper's criterion awkwardly. This tension motivated Lakatos to replace the simple theory-as-unit picture with the richer notion of research programs with hard cores and protective belts, and Kuhn to argue that actual science operates by paradigm commitments that are not abandoned in response to anomalies but only replaced in revolutionary episodes. Both represent post-Popperian attempts to describe what makes science genuinely scientific without requiring that every theory be a hostage to a single possible observation.