Inference Procedure with Uncertainty for Problem Reduction Method

摘要

In the application of artificial intelligence to complex decisionmaking problems, a decisionmaking tree is employed. Rational inference procedures for problem reduction are described which are based on either Bayes' theory or Dempster and Shafer's theory. At each branch of the decision tree, a probability is assigned to the evidence at that branch. In the Bayesian system, if there is a 75 percent probability that the evidence is true, for example, then there must be a 25 percent probability that the evidence is false. The Dempster and Shafer system allows for ignorance of evidence. A modification of the Bayesian, and Dempster and Shafer systems is presented which includes fuzzy reasoning, an approximate reasoning process which is compatible with human intuitions, and can yield a plausible answer even for problems in which all conditions for mathematical approaches are not satisfied. Fuzzy reasoning is presented as an alternative to statistical approaches in problem reduction.