Goal-directed diagnosis—a diagnostic reasoning framework for exploratory-corrective domains
Ron Rymon · Artificial Intelligence · 1996
In many diagnosis and repair domains, diagnostic reasoning cannot be abstracted from repair actions, nor from actions necessary to obtain information of diagnostic value. TraumAID 2.0, a consultation system for multiple trauma management, implements a reasoning architecture for exploratory-corrective domains which integrates diagnostic reasoning with planning and action. Taking the view that a diagnosis is only worthwhile to the extent that it can affect repair decisions, its goal-directed diagnosis (GDD) framework views the diagnosis object as secondary to the formation of appropriate goals for a complementary planner. We start this article with a specific case of multiple trauma injury. Demonstrating some of the key features of exploratory-corrective domains, we use this case to exemplify the suitability of our approach. Before focusing on the GDD reasoner itself, we overview TraumAID 2.0's exploratory-corrective management architecture in which it is embedded with a companion planner, and explain its iterative interaction with the physician. The paper's main body presents GDD's formalism, and knowledge representation scheme, and proposes a design methodology for GDD-based systems (the companion planner, and the implemented system are described elsewhere). Before concluding, we present a set of strategies characteristic of exploratory-corrective domains, and argue the suitability of our framework for their implementation.