Is Science Really Self-Correcting?
There's more than one variable.
There's the sociological variable. Some views are more popular than others in the scientific and its surrounding political community. It can reach the point where holding particular views is a condition of academic hiring and for achieving tenure. Holding particular views becomes a condition for publishing in the leading journals and for gaining standing in one's specialty. When influential voices start calling for criminal prosecution of people who disagree with the 'correct' views, then the chances of those views being successfully challenged if they are in fact mistaken grow exceedingly remote.
There's obviously going to be an access to information variable. If a current mistaken view can only be corrected in light of a particular kind of new information, and if human beings have no access to that new information, then it's hard to see how the mistaken view can be corrected. The needed information might not be accessible because of our current location in space and time, because of some inherent limitation in human cognitive abilities, or because it's only available behind event-horizons, or some other place that isn't observable.
Another more philosophical variable is associated with the distinction between instrumentalism and scientific realism. Scientific realism is the view that physical theory is
descriptive, that the entities named in scientific theory actually exist in physical reality and that the structure of the scientific theory actually captures how physical reality interacts and behaves. Instrumentalism is the view that scientific theory is merely
predictive and only serves to allow scientists to predict what the results of observations will be. With instrumentalism, any theory that allows observations to be successfully predicted with suitable accuracy is a good theory. There is no expectation that whatever is imagined as part of the theory in the course of generating its predictions correspond to anything in the physical reality that is being observed. (The physicist Ernst Mach questioned the literal existence of atoms on this basis.)
The paradigmatic example of that is geocentric cosmology and Copernicus' heleocentric cosmology. Copernicus' system was no more accurate than the late medieval variants of geocentric cosmology, at least when restricted to observations of heavenly bodies in the sky as seen from the surface of the earth. What's more, geocentric cosmology could be made as accurate as desired, simply by adding additional epicycles and equants. So the theories were basically equivalent in observational terms, despite their imagining very different heavenly movements. Another example might be contemporary quantum mechanics. As I understand it, there are at least two mathematical formalisms that produce accurate instrumental results, Schroedinger's wave mechanics and Heisenberg's matrix mechanics.
The relevance of all this to the question in the subject line is that if there are multiple theories capable of accounting for all observations to any desired level of accuracy, then there might not be any easy way to distinguish which theory is actually a true description of physical reality and which one isn't, assuming that either one is.