Does science progress in a cumulative way? In other words, do later scientific theories build upon earlier ones by keeping older established truths about the world, and adding us more truths? The standard answer, familiar from Whiggish history and brief historical asides in scientific textbooks, is that it does. Aristotelian-Ptolemaic astronomy predicted the motions of the planets; Copernican-Galilean astronomy did a better job of prediction and was simpler; Newtonian physics explained how force and mass accounted for the motions studied by Galileo; Relativity Theory adds an explanation of the motion of objects close to the speed of light.
The problem is that this account is either false or misleading at every stage. The Copernican astronomy was NOT more accurate than Ptolemaic, and required just as many "epicycles." Newtonian physics made planetary motion much more complex (and less circular) than Galileo had imagined. In the Newtonian universe mass is preserved and energy is preserved, and both are found in absolute space and absolute time. In Relativity Theory, mass-energy is preserved in a space-time manifold.
It's tempting to say that, despite the differences, the theories are still "talking about the same things." But what would those things be? Aristotle talked about the Four Elements -- earth, water, fire, air -- and quintessence (the last being the stuff the planets and stars are made of). Galileo had to drop this elemental theory in order to reconcile terrestrial and non-terrestrial motion, but he didn't have anything better to put in its place. Newton talked about objects with mass exerting the otherwise inexplicable force of "gravity" over potentially infinite reaches of space. Einstein talked about mass-energy causing deformations in space-time.
We might say that the observations are the same. But what we see is, at the least, heavily influenced by what we think we are seeing. A "pendulum" is a special thing, a central model for all kinds of regular motion, beginning with Galileo. Prior to Galileo, a "pendulum" is a silly toy, illustrating nothing but constrained fall. Mass is another good illustration of how nothing is simply observed. If "mass" could be observed, cave men would have known about it. Mass is inferred based on observations whose intent is to measure it: absent the intent, there is no mass to observe.
Are we left with the conclusion that science does not advance? Perhaps it advances only in the sense that later theories can explain why our ancestors said what they said, and saw what they saw, whereas earlier theories cannot do the same for the latter. Contemporary chemists can explain why Priestley obtained a gas that encouraged combustion, by saying that it is oxygen. However, Priestley's theory that he had isolated "dephlogisticated air" cannot explain many of the phenomena familiar to the modern chemist. We might say that scientific theories are "trap doors": once one has gone through them, one cannot go back.
But are we sure that the crossroads of science can only go one way? Was there a way to save Aristotle's theory of the Four (terrestrial) Elements if there had been a brilliant defender of it? I can imagine a science fiction novel about an alternative present in which we treat illnesses and travel in space, perhaps even better than we do now, using updated Aristotelian physics.