ABSTRACT
Although this description of experimental inference is plausible in some respects, it is very seriously over-simplified and will not do as it stands. For one thing, it cannot elude the Russell-Goodman problem, which applies to it as well as to inductive generalization. Instead of thinking of the rival hypotheses about emeralds as being the results of a generalization, we can think of them as hypotheses to be tested. Since any prediction about the color of an emerald that we can verify will be made prior to time t, we have no way of confirming one of these hypotheses at the expense of the other. An analogous point can be made about the sort of hypotheses Russell considered - for example, the hypothesis that all cows are observed or found within ten miles of Konigsberg. Suitable observers will find the predictions they make on the basis of this hypothesis invariably borne out by their observations; and if they attempt to compare their predictions with evidence of other kinds (such as the testimony of others), the Goodman problem can be redirected to this other evidence. For example, the auxiliary hypothesis that the testimony of other observers is reliable can be shown to be no more likely, if we are restricted to the hypothetico-deductive method, than the alternative auxiliary hypothesis that other observers are reliable only on matters not pertaining to cows that I do not observe. Evidence pertaining to what I do not observe is always susceptible to more than one interpretation.
