Some problems related with probabilistic interpretations for certainty factors

Abstract

Some problems identified by J.B. Adams (Mathemat. Biosciences, vol.32, p.177, 1976) in the MYCIN-model are discussed. It is found that these problems are related only with an inappropriate interpretation for certainty factors assigned to rules. This interpretation is purely formal and has no influence upon the actual elicitation and use of such factors. It is pointed out that one of D. Heckerman's (1986) arguments about the MYCIN-model is incorrect. Then, the interpretations for certainty factors given by Heckerman, namely, CF/sub 1/, CF/sub 2/, are critically reviewed. It is shown that similar problems to those in the original MYCIN-model exist also in CF/sub 1/ and CF/sub 2/. This supports further the point of view that certainty factors should not be interpreted as a measure of belief-updating.<>