View article

As long as knowledge-based systems have been built, facilities for handling uncertainty have been an integral part. In the early days of rule-based programming, the predominant methods used variants on probability calculus to combine certainty factors associated with applicable rules. Although it was recognized that certainty factors did not conform to the well-established theory of probability, these methods were nevertheless favored because the probabilistic techniques available at the time seemed to require either specifying an intractable number of parameters, or assuming an unrealistic set of independence relationships.

Today, methods based rmly in probability theory have once again begun to gain acceptance in the computer-science and uncertain-reasoning communities. The\breakthrough" was a graphical modeling language for representing uncertain relationships. Although Bayesian networks have gained widespread use in knowledge-based systems relatively recently, they are based on modeling