The Consistency of the Medical Expert System CADIAG-2: A Probabilistic Approach

CADIAG-2 is a well known rule-based medical expert system aimed at providing support in medical diagnose in the field of internal medicine. Its knowledge base consists of a large collection of IF-THEN rules that represent uncertain relationships between distinct medical entities. Given this uncertainty and the size of the system, it has been challenging to validate its consistency. Recent attempts to partially formalize CADIAG-2's knowledge base into decidable Gdel logics have shown that, on formalization, the system is inconsistent. In this paper, the authors use an alternative, more expressive formalization of CADIAG-2's knowledge base as a set of probabilistic conditional statements and apply their probabilistic logic solver (Pronto) to confirm its inconsistency and compute its conflicting sets of rules under a slightly relaxed interpretation. Once this is achieved, the authors define a measure to evaluate inconsistency and discuss suitable repair strategies for CADIAG-2 and similar systems.