What makes science different from astrology? From psychoanalysis? From Marxist historical theory? From religious cosmology? They all make claims about reality. They all have practitioners who believe in them. Astrology has made predictions. Freudian psychoanalysis has been applied clinically for decades. So what separates them from genuine science?
This is the demarcation problem, and Karl Popper (1902–1994) proposed the most influential answer of the 20th century. His solution, developed in The Logic of Scientific Discovery (1934, English translation 1959), is elegant and initially surprising: science is not defined by what it can prove, but by what it can refute.
Popper's starting point is a logical observation that sounds almost too obvious. You cannot prove a universal generalization from any finite number of observations. You have seen a million white swans, but that doesn't prove that all swans are white, because the next swan might be black. (This actually happened when European naturalists reached Australia and found black swans, demonstrating the point dramatically.) But you can definitively refute a universal generalization from a single counter-example. One black swan conclusively shows that "all swans are white" is false.
This asymmetry, you can falsify but not verify universal laws, led Popper to his criterion of falsifiability: a theory is scientific if and only if it is possible, in principle, for some observation or experiment to show it is false. The theory must stick its neck out, it must make predictions that could be wrong. If a theory is set up so that any possible observation can be accommodated, if it can explain everything by adjusting its auxiliary hypotheses, then it is not scientific. It is unfalsifiable, and therefore empty.
Popper's target examples are revealing. Freudian psychoanalysis and Adlerian psychology both claimed to explain human behavior. But Popper noticed that they explained it in every possible direction, whatever a person did, the theory could accommodate it as either expressing a repressed wish (Freud) or overcoming an inferiority complex (Adler). No possible human behavior counted as evidence against either theory, which meant both theories were saying something very vague, compatible with everything, ruling out nothing.
In contrast, Einstein's general relativity made a specific, falsifiable prediction: light from distant stars should be bent by the sun's gravitational field by a specific, calculable amount. In 1919, Arthur Eddington's eclipse expedition measured the bending of starlight during a solar eclipse. If the measurement had come out differently, general relativity would have been falsified. The theory was risky. It could have been wrong in a specific, testable way. That is what makes it scientific.
The criterion also explains the appeal and the danger of certain political ideologies. Marxist historical theory in its more dogmatic forms was not falsifiable, capitalism's survival was evidence of its internal contradictions (confirming the theory), and its weakening was evidence of its inevitable collapse (also confirming the theory). Whatever happened, the theory had an explanation. This is not a scientific virtue; it is a defect.
Quick reflection
Think of a belief you hold about human nature, history, or social life — something like 'people are fundamentally self-interested' or 'inequality always increases tension.' What would it take to falsify your belief? Is there any possible observation that would make you give it up? If not, is your belief scientific in Popper's sense — and does that matter for whether it's worth holding?