Accession Number : ADA255471


Title :   Probabilistic Inference


Descriptive Note : Technical rept.


Corporate Author : ROCHESTER UNIV NY DEPT OF PHILOSOPHY


Personal Author(s) : Kyburg Jr, Henry E


Full Text : https://apps.dtic.mil/dtic/tr/fulltext/u2/a255471.pdf


Report Date : Jan 1992


Pagination or Media Count : 11


Abstract : It is important to distinguish probabilistic reasoning from probabilistic inference. Probabilistic reasoning may concern the manipulation of knowledge of probabilities in the context of decision theory, or it may involve the updating of probabilities in the light of new evidence via Bayes' theorem or some other procedure. Both of these operations are essentially deductive in character. Contrasted with these procedures of manipulating or computing with probabilities, is the use of probabilistic rules of inference: rules that lead from one sentence (or a set of sentences) to another sentence, but do so in a way that need not be truth preserving. One could attempt to get along without probabilistic inference in AI, but it would be very difficult and unnatural. Instances of such rules are several classes of inference rules associated with statistics, and some rules discussed by philosophers. In artificial intelligence the rules that fall into this category are (mainly) default rules; these are not generally construed probabilistic, but obviously default rules that more often lead you astray than to the truth would be poor ones.


Descriptors :   *DECISION THEORY , *ARTIFICIAL INTELLIGENCE , REASONING , STATISTICS , THEOREMS , OPERATION , LIGHT


Subject Categories : Operations Research
      Cybernetics


Distribution Statement : APPROVED FOR PUBLIC RELEASE