The Inconsistent Use of Measures of Certainty in Artificial Intelligence Research
Eric Horvitz, David Heckerman
There has been a great deal of interest among artificial intelligence researchers in methodologies for using subjective measures of belief for reasoning about uncertain events. We emphasize the fundamental difference between the use of measures of absolute belief and measures of change in belief. We show that expert system investigators have elicited measures of absolute belief and have used them as though they were measures of change in belief. We argue that such use is counterintuitive and formally inconsistent. In our discussion of the inconsistency, we introduce a belief-updating paradigm used frequently in plausible reasoning. Finally, we attempt to characterize the potential impact of the inconsistency.
Keywords: Measures of Uncertainty
PDF Link: /papers/85/p137-horvitz.pdf
AUTHOR = "Eric Horvitz
and David Heckerman",
TITLE = "The Inconsistent Use of Measures of Certainty in Artificial Intelligence Research",
BOOKTITLE = "Uncertainty in Artificial Intelligence Annual Conference on Uncertainty in Artificial Intelligence (UAI-85)",
PUBLISHER = "Elsevier Science",
ADDRESS = "Amsterdam, NL",
YEAR = "1985",
PAGES = "137--151"