In 1687 Isaac Newton set out his system of the world, and in its third book he stated, as a hypothesis, "That the centre of the system of the world is immovable." Something curious follows from his own laws. If the whole solar system were drifting through absolute space at a steady speed, nothing anyone could observe would change. Newton's physics with the system at rest and Newton's physics with it moving at any constant velocity make exactly the same predictions. Bas van Fraassen later used this to show that a single theory comes in endless rival versions that no evidence can separate.
What underdetermination means
A theory is underdetermined when the available evidence does not settle whether it or a rival is true. The best known form says that a failed test never shows which of the many assumptions behind it was at fault. Philosophers now distinguish a further form. Contrastive underdetermination arises when the evidence fails to single out one theory among rivals, because each fits the data. Newton's moving and resting systems are an extreme case: they predict exactly the same observations, so they are empirically equivalent, and no possible experiment favors one.
Most cases are milder. Transient underdetermination is the kind that later evidence can resolve: the rivals fit what we know now, but they part ways somewhere we have not yet looked. The deeper question is whether permanent underdetermination is common, or whether Newton's case is a rare curiosity.
Laudan and Leplin's reply
In 1991 Larry Laudan and Jarrett Leplin argued that the threat was overstated. A theory's observable consequences, they pointed out, depend on the auxiliary assumptions and instruments available, and these change. Two theories that are empirically equivalent today may come apart once a new instrument or a new auxiliary theory exists. So empirical equivalence can change as background knowledge grows. They added that evidence can support a theory indirectly, through its links to other well supported theories, even when the evidence is not among its direct consequences. Underdetermination, on this view, is a snapshot, not a permanent condition.
Inference to the best explanation
Scientists choose between rivals all the time, and one account of how they do it is inference to the best explanation: accepting the hypothesis that would best explain the evidence. The American philosopher Gilbert Harman named the pattern in 1965. Explanations are judged by virtues such as simplicity, scope, unification of separate facts, and fit with what else we know.
The worry is obvious. Why should the explanation we find most satisfying be the one most likely to be true? A universe that rewarded elegant theories would be convenient for us, but nothing guarantees we live in one. Critics of inference to the best explanation stress that explanatory appeal may not track truth, and defenders reply that simplicity and unification have, in fact, a long record of leading to theories that later passed new tests.
The problem of unconceived alternatives
The American philosopher Kyle Stanford gave underdetermination a historical twist in 2006. The trouble, he argued, is not rival theories we know about but rivals nobody has thought of yet. He studied the history of theories of inheritance and found a repeated pattern: scientists have repeatedly failed to imagine the theories that later replaced their own. Darwin, for instance, proposed a theory of inherited particles gathered from all over the body, unable to conceive of the genetics that came later. If the best theory available is chosen only from among those someone has happened to think of, its success may say less than we hope about its truth.
What underdetermination does not show
None of this means that any theory is as good as any other. Underdetermination does not mean anything goes, because evidence usually rules out most rivals, leaving a small number still in play. Underdetermination opens a narrower space, and part of what fills it, some philosophers argue, are judgments about simplicity, fruitfulness and sometimes social values. Others hold that the space is quickly closed by further evidence.
Worked example: three systems of the world
Around 1600 an educated astronomer had three systems to choose from. Ptolemy's put the Earth at the center with everything circling it. Copernicus's put the Sun at the center. The Danish astronomer Tycho Brahe, in 1588, proposed a compromise: the planets circle the Sun, and the Sun, carrying them, circles a motionless Earth. In 1610 Galileo's telescope showed Venus going through a full set of phases, like the Moon. That ruled out Ptolemy's arrangement, but it fitted Tycho's as well as Copernicus's, because in both Venus circles the Sun. As far as the relative motions of the heavens were concerned, the two systems were geometrically equivalent. The tie was transient. Newton's physics, later in the century, made it very hard to see how a vast Sun could circle a small Earth. Then came direct evidence of the Earth's motion: the aberration of starlight, reported by James Bradley in 1729, and later the long sought stellar parallax. Tycho's system was not refuted by one observation; it was squeezed out as new instruments and a new physics changed what counted as evidence.
