On the measurability of knowledge acquisition and query processing Original Research Article

The adaptation of a priori knowledge to new relevant facts and the use of the so enriched knowledge base for inference and response is the subject of this paper. Here knowledge processing is realised in a conditional and probabilistic environment under maximum entropy (MaxEnt) and minimum relative entropy, respectively. It is measurable: the amount of knowledge acquired, the remaining (first order) uncertainty, the inferential strength when focussing things, and the (second order) uncertainty in given answers, are quantifiable as well as the relevance of new facts; they all measure in [bit]. This makes the MaxEnt-probability distribution accessible to clear and substantial interpretations: due to incomplete information about the domain parts of the knowledge are significant or reliable, others are not. And this non-reliability can be identified and evaluated. The expert system shell SPIRIT supports this sort of knowledge evaluation, which is shown by suitable examples.