certainty factors

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

Certainty factors are way of performing uncertain reasoning. While Bayesian reasinong demands that belief be precisely coded as a single probability-like number, confidence factors allow a range with belief for and against a particlar conclusion to build as new evidence is found.

Initially , each initial fact/evidence is initially given a certainty factor where positive in the range +1 to -1, where positive values denote evidence for and negative values denote evidence against, with zero meaning no evidence and the extremes +1/-1 meaning incontrovertable evidence for/against). Derived facts are given two values, a measure of belief and a measure of diselief (each [0,1]) based on evidence, and these are accumulated as further positive or negative evidence arises. This works well in medical domains where there can be evidence both for and against a diagnosis. Certainty factors were used in the successful early expert system {{MYCIN} and still used in expert systems including neural network variants. However, they are less popular now than Bayesian methods and fuzzy reasoning.

Used in Chap. 3: pages 27, 29, 30, 34; Chap. 18: pages 275, 291

Also known as certainty factors, reasoning with certainty factors