It has been recognized that one of the limitations of the """"""""first generation"""""""" of medical artificial intelligence systems was the phenomenological nature of the knowledge to which they had access. To perform effectively in difficult diagnostic and therapeutic situations, such programs will need knowledge bases which are sufficiently rich to enable them to reason about pathophysiological mechanisms. Thus, the representation of the domain knowledge of physiology is becoming an important part of the development of such systems. In this research, several problems inherent in representing and reasoning about complex physiological knowledge will be investigated in the context of developing a medical knowledge base and building a module which uses physiological reasoning to aid in evaluating diagnostic hypotheses and therapeutic plans. The module will have access to a knowledge base which contains the domain knowledge of a subset of renal basic physiology, as well as clinical knowledge about fluid and electrolyte disturbances. Its performance will be evaluated by integration into a system with components which generate diagnostic hypotheses, therapy plans, and explanations of the system's reasoning and output. The following issues will be explored: a. qualitative vs. quantitative representation of parameter values and qualitative reasoning or simulation vs. numerical calculation; b. temporal representation and reasoning; c. classification of causal mechanisms, with emphasis on concepts specific to the domain of physiology. The above issues are relevant to the field of knowledge representation at large. Thus, the proposed investigation may lead to the development of knowledge representation and reasoning techniques which are particularly adapted to the domain of renal physiology but may have wider applicability as well.
