Perhaps the most powerful argument for methodological naturalism based on the goals of
science proceeds as follows. One of the central goals of all scientists is to justify their
claims in such a way that most reasonable persons with sufficient expertise will accept
them. But it is impossible to justify beliefs about supernatural entities in this way: no
public evidence can establish their (probable) truth or falsity. Thus, even though
supernaturalistic explanations might be true (and might even be justified for particular
individuals), scientists should not give them, for there is nothing scientists could do to
prove to other members of the scientific community that one such explanation should be
accepted and another rejected.
Clearly, the key premise of this argument is that the intersubjective justification of a
belief about supernatural entities is impossible. At least two reasons might be offered in
support of this premise. The first is that such a belief cannot be tested by evidence; that
is, it can neither be confirmed nor be disconfirmed either by new information or by what
we already know (cf. Pennock 1998, 206; Sober 2000, 46–57). Let us call this the
“testability problem.” Notice that a very broad sense of the verb “to test” is intended here.
It includes, of course, testing
end p.292
by experiment, but it also includes the sort of testing a historian might do: carefully
comparing the ability of some hypothesis to explain various known facts to the ability of
serious alternative hypotheses to explain those facts. A second reason for doubting that
supernaturalistic hypotheses can be intersubjectively justified is that their probability
prior to testing cannot be assessed. In other words, it is impossible to determine their
(initial) degree of plausibility or implausibility, and so impossible to make a rational
decision about which of them to test and impossible to determine the significance of any
testing that is done. Let us call this the “plausibility problem.” To refute this new
argument for methodological naturalism, both the testability problem and the plausibility
problem must be solved.
Of course, some scientists deny that plausibility judgments play any role in science. But
philosophers have shown that scientists presuppose such judgments all the time. Indeed,
even the claim that a fact is strong evidence for a hypothesis in the sense that it
significantly raises the ratio of the probability of its truth to the probability of its
falsehood presupposes a number of plausibility judgments. For a fact can significantly
raise this ratio only if it is antecedently more probable given that theory than it is given
its denial, and any precise assessment of a fact's antecedent probability given the denial
of a theory is impossible unless one can assess the relative plausibility of various
alternatives to that theory.^6 Furthermore, methods like statistical significance testing,
which are actually employed by scientists and which ignore prior probabilities (and thus
allegedly make science more “objective”), have been shown to be flawed for that very
reason (e.g., Edwards, Lindman, and Savage 1963).
One response to the plausibility problem grants that plausibility judgments about
supernaturalistic hypotheses are subjective, but denies that plausibility judgments about
naturalistic hypotheses are any less subjective. The claim here is that, in science and in
every other discipline, we just find ourselves taking certain hypotheses seriously and