Other Means for Ideology to Affect Science

In addition to the mechanisms discussed in section 3.2, ideology may enter into academic study through attempts to produce a posterior probability distribution, in which the policy which will minimize the expectation of one's own loss function is the same policy which will minimize the expectation of the loss function of someone else with decision-making power, i.e. through research as a coalition-building tool.

At its most sinister, this seems to be the process that Proctor [57] describes as the attempt by the Nazis to `depoliticize' repressive policies against several minorities. Harding [29], by reference to various commentators [66,63,54], and in the context of developing standpoint epistemology, proposes as the mechanism by which de-politicization was achieved, `the creation of a ``normal'' or ``authoritative'' science.' A Bayesian might interpret the phrase `a ``normal'' or ``authoritative'' science' to mean that a single theory $T$, which, although it may have given Nazi policy $y_N$ a high $L(y_N;T)$, where $L$ is the loss function for a more moderate ideology than the Nazis' own, claimed that the available alternative policies $y_a$ would have loss functions $L(y_a;T)$ that were almost as high as $L(y_N;T)$, or even higher, was held up as certainly true, $P(T)=1$. If the probabilistic nature of truth is accepted, as Bayesians recommend, such a claim of certain truth for one theory will not seem very convincing.

The author does not, however, wish to claim that Bayesian thinking provides infallible protection against similar deceptions by extremists seeking to build coalitions: one obvious loophole is the possibility of popularizing a prior probability distribution, designed to lead to a high posterior probability for the de-politicizing theory; if an unscrupulous researcher or politician chooses a posterior probability distribution in advance, Bayes' theorem (equation 7) can readily be inverted, to generate a ``rigged'' prior probability distribution, which, along with the available evidence, will produce that posterior probability distribution. The attempt to make one theory appear certainly true is an extreme example of this: if a theory has prior probability $1$, it will have posterior probability $1$. Another such loophole is the possibility of selectively publicizing evidence, in such a way as to lead to a high posterior probability for the de-politicizing theory.

Even for those who support more moderate ideologies, and are familiar with the methods of Bayesian inference, it is easy to slip into the making of a theoretical claim with little evidence, as a way of convincing believers in other ideologies to support the policies that serve one's own ideology. A relatively trivial, but highly illustrative, example, is the author's reaction to sweeping condemnation of manufactured pop groups in general, and the recent Popstars television series, in the UK, which constructed such a group through a series of televised auditions, in particular, for allegedly producing music of poor quality. This reaction was: even if the criteria used to select members for a manufactured pop group are only partly based on musical ability, such a process provides a better approximation to fair equality of opportunity than waiting for a group to arise spontaneously. Therefore, in principle, one would expect manufactured pop groups to produce better music than pop groups which arise spontaneously. The hidden theoretical assumption, (subconsciously) intended to persuade those whose ideology places high worth on good music to agree, on policy, with the author, whose ideology places high worth on fair equality of opportunity for its own sake, is that fair equality of opportunity is associated with high productivity. This is not an unreasonable assumption, but it would have been dangerous to use it in this concealed and non-probabilistic way.