Abstract

Subjective Bayesianism is the current orthodoxy in confirmation theory. In broad outline, this view claims a) that confirmation is a relation of positive relevance, viz., that a piece of evidence confirms a theory if it increases its probability; b) that this relation of confirmation is captured by Bayes’s theorem; c) that, hence, the only factors relevant to confirmation of a theory are its prior probability, the likelihood of the evidence given the theory and the probability of the evidence; d) that the specification of the prior probability of a theory is a purely subjective matter; e) that the only constraint on an assignment of prior probabilities to several theories should be that they obey the axioms of the probability calculus; f) that, hence, the reasonableness of a belief does not depend on its content; nor, ultimately, on whether the belief is made reasonable by the evidence; and g) that degrees of belief are probabilities and that belief is always a matter of degree. This theory has had many successes. Old tangles, like the ravens paradox or the grue problem, are resolved. New tangles, like the problem of old evidence, have been, to some extent, resolved. But there is still a pervasive dissatisfaction with subjective Bayesianism. This dissatisfaction concerns all of the theses to above, but it is centred mostly around the point that subjective Bayesianism is too subjective to offer an adequate theory of confirmation and of rational belief. Yet, up until recently this dissatisfaction had not borne fruits in a fully developed alternative theory of confirmation. Perhaps, only Deborah Mayo’s error-statistical approach was a fully developed alternative, but this too faced considerable problems, one central among them being that she takes all probabilities to be relative frequencies. One thing is sure. Probability theory should place a central role in a theory of confirmation. But how are we to interpret probabilities if we want to avoid the subjective element of Bayesianism and the objective element of the error-statistical approach?