The Argument for Realism from the Success of Science
and Laudan's Criticism of It

Modern day realists recognize that the fact of empirical underdetermination of theory choice by evidence blocks any attempt to infer from the fact that a theory is succesful at predicting observed phenomena to the conclusion that the theory is true in a correspondence sense.  They recognize that such empirical confirmation is the basis for the acceptance of theories, and thus that all one can directly conclude from the fact that a theory is accepted is the fact that is it indeed succesful at "getting the numbers right," or, on in van Frassen's terminology, it is "empirically adequate."

Realists want to distinguish between these two questions:

                                      "Why do we accept a theory?" and
                                       "Why is that theory so succesful at getting the numbers right?"
For the anti-realist the answer is effectively the same for both questions: We accept theories because they get the numbers right; if we apply this criterion of success long enough in accepting and rejecting theories, always selecting for that theory which gets more numbers, more right, we eventually end up with theories which are indeed remarkably succesful at getting the numbers right.  This is the logical consequence of our criterion for acceptance.  (If we view theoreis as tools for predicting observed outcomes, as do some anti-realists, we keep improving our "tools" until we get ones that are remarkably succesful at doing whatever they are used to do.)  Accepted theories are remarkably succesful simply because the criterion for acceptance is that they be succesful.

"No," say the realists. "You anti-realists are right, of course, in saying that the basis for acceptance is empirical success, but this success itself needs an explanation. It is no 'expanation' of why these particular theories are the succesful ones to say that they have been selected for acceptance because they are successful.  We realists can explain this success: these theories are as succesful at getting the numbers right as in they are because they are approximately true and the unobservable or 'theoretical' entities, states and processes postulated by these theoreis are real entities, states, and process in the world; they are not just constructions of our theories."

"The best explanation for the success of the currently accepted theories of  science," say these theory realists, "is that these theories are approximately true, so by the rule known as 'inference to the best explanation' which we use all the time in science, we realist philosophers can legitimately infer that we are rational to accept our realist interpretation of theories because this realist 'hypothesis' of the approximate truth of theories is the best explanation of these theory's empirical successes (aka 'empirical adequacy').

Larry Laudan is an anti-realist who accepts both the naturalistic view that scientific methodology is itself a subject for scientific study and explanation, and the pragmatic view that acceptance of theories is rationally warranted by such "pragmatic utilities" as simplicity, fertility, precision, scope, and unification.  Thus he hopes to refute the realists' argument from the "success of science" (the so-called "no miracles" or "cosmic co-incidence argument").  To do so, he must first characterize his opponent; he calls this opponent's position "Convergent Epistemological Realism."

Laudan characterizes the realists' argument under his conception of "Convergent Epistemological Realism" thus:
First Argument
Premiss 1: If a theory is approximately true, it will be emprically succesful at getting the numbers right (it will be "empirically adequate").
Premiss 2: If the non-observational terms of a theory refer to real entities, states, and processes, that theory will be empirically succesful at getting the numbers right (it will be "empirically adequate")..
Premiss 3: The theories of most mature sciences are in fact empirically succesful at getting the numbers right (they are "empirically adequate").
Conclusion: It is rational to believe that currently accepted theories in mature sciences are (probably) approximately true and their non-observational terms really do refer to real entities, states, and processes.
Second Argument:
Premiss 1: If a theory is approximately true and its non-observational terms genuinely refer to real entities,states, and processes, then a successor theory to that theory (one which replaces it in the acceptance of the scientific community in time) will preserve those parts of the predecessor theory in which it was emnpirically succesful (i.e. the earlier trheory will be shown to be a "limiting case" of the later theory).
Premiss 2: In the mature sciences, scientists do in fact endeavor to preserve earlier theories as limiting cases of later theories, and in fact generally succeed in doing so.

Conclusion: It is rational to believe that (probably) accepted theories in the mature sciences are approximately true and their non-observational terms really do refer to real entities, states, and processes.

Arguments of this form are what C.S. Peirce called "abductive inferences" or "inferences to the best explanation."  Thus "approximate truth" and "reference to real entities, states, and processes" are intended to be the "best explanation" for the emprical success of scientific theories.  Such inferences to the best explanation are made all the time in science, so they should be justified here in philosophy of science as well (this is the naturalistic view).  But of course no such inductive arguments are valid, so the most they could ever establish would be the probability of the truth of their conclusions.

Laudan attacks both of these arguments in Chapter 5 of Science and Values.  He concedes that what was identified as Premiss 3 of the First Argument  may very well be true (although he complains that realists are very vague on what they mean by "succesful" in this claim).  However, he seeks to disprove Premisses 1 and 2 of both arguments in order to show that the realists' alleged defense of their central claim is unsound.

Laudan points out that both of the Premisses 1 and 2 of both of the above arguments are historical claims about what is alleged to be the case in fact.  Thus he seeks to disprove them by pointing to specific examples of historical theories which he claims show these premisses are false.

Premiss 1 of the first argument is the weakened version of what the realist would like to be able to say, but can't:

If a theory is true, then it will be empirically succesful.
This would indeed be a strong premiss, because, Laudan concedes, "it is self-evident."  But in moving from "truth" simply to "approximately true" the plausibility of the claim, he argues, is greatly diiminished.  This is because to admit that the theory is only approximately true is equivalent to admitting that some of its empirical predictions will not be confirmed.  Every theory implies an infinite number of empirical consequences, thus those empirical consequences of a theory which have in fact been observationally verified always form only a small part of all logically possible empirical consequences.  Now it may very well happen that those consequences which have been empirically verified will be among just those phenomena where the only approximately true theory goes astray.  Thus there can be no guarantee that approximate truth will lead to success in empirical predictions among those which have in fact been observationally verified.

Moreover, Laudan argues, we cannot go the other way either, from empirical success to approximate truth.  To support this claim he argues that if the central non-observational terms of a theory are admitted not to refer to real entities, then that theory cannot be even approximately true in what it claims about such non-existent entities, states, or processes.  He then proceeds to list (p.121) a variety of theories which were in fact historically quite empirically succesful, but which today, by the realists own admission, we regard as not in fact referring to real entities.  Such theories referring to "phlogiston," "caloric," "electrical fluid," "bodily humors," and various "aethers" all have been quite succesful in predicting a wide variety of empirical phenomena.  However today we do not regard such theories as only approximately true; we regard them as wholly false because we regard their central explanatory constructs as not referring to any real entities at all.  Thus Premiss 2 of the First Argument is also "refuted" by historical counterexamples because here we see that non-referential theories (i.e. theories which we today regard as not having referred to anything real at all) such as the phlogiston theory of combustion  and the caloric theory of heat have in historical fact been 'succesful.'

Furthermore there have been theories which realists today would want to insist are theories which really do refer, but which have been notably unsuccesful.  Examples of such unsuccesful theories would be mid nineteenth century atomism or plate tectonics prior to World War II; while realists today would want to insist that there really are chemical atoms as Dalton hypothesized and tectonic plates as hypothesized by Wegner, given the phenomena then known (in those times) the theories of atoms and plate tectonics were not very succesful at all in predicting the then known phenomena.

Finally Laudan calls attention to repeated historical episodes where successor theoreis have dispensed altogether with the central terms of their predecessor theory.  Even if we allow that occasionally scientists do preserve earlier theoreis as limiting cases of later theories, this is a small number of theory changes and not sufficient to support the realists' broad claim in Premiss 2 of the Second Argument.