Evidential Confirmation as Transformed Probability
This work addresses the challenge of aggregating confirmatory evidence in AI, offering a unification that clarifies and integrates existing methods for researchers in uncertain reasoning.
The paper tackles the problem of unifying alternative approaches to evidential confirmation in AI by showing that a revised MYCIN Certainty Factor method is equivalent to a special case of Dempster-Shafer theory, resolving issues like the 'taxe-them-or-leave-them' problem of priors and providing a well-understood axiomatic basis.
A considerable body of work in AI has been concerned with aggregating measures of confirmatory and disconfirmatory evidence for a common set of propositions. Claiming classical probability to be inadequate or inappropriate, several researchers have gone so far as to invent new formalisms and methods. We show how to represent two major such alternative approaches to evidential confirmation not only in terms of transformed (Bayesian) probability, but also in terms of each other. This unifies two of the leading approaches to confirmation theory, by showing that a revised MYCIN Certainty Factor method [12] is equivalent to a special case of Dempster-Shafer theory. It yields a well-understood axiomatic basis, i.e. conditional independence, to interpret previous work on quantitative confirmation theory. It substantially resolves the "taxe-them-or-leave-them" problem of priors: MYCIN had to leave them out, while PROSPECTOR had to have them in. It recasts some of confirmation theory's advantages in terms of the psychological accessibility of probabilistic information in different (transformed) formats. Finally, it helps to unify the representation of uncertain reasoning (see also [11]).