On representation of source reliability through weights on rules

Rule‐based expert systems often require their users assess the state of an event and then “report” to the system that state. the system then uses that report of the event as the basis of inference, that is, the hypothesis. the problem is that the report is just that, a report. the report is not necessarily the same as the underlying event that occurred. the relationship between the report and the underlying event is referred to as the reliability of the report or the credibility of the report. This article integrates the difference between the report and the underlying event into the weights of two different uncertainty representation schemes used in expert systems. In each case the impact of that reliability is explored. For example, it is found that the model without reliability is equivalent to the model with reliability, in the case of complete reliability. Further, in some cases the weights are monotonic functions of reliability. Accounting for reliability is found to have a substantial impact on the value of the weights. Further, one approach is seen as influenced less by a lack of reliability than another approach.