Deep and shallow models in medical expert systems

摘要

In the context of medical expert systems a deep system is often used synonymously with a system that models some kind of causal process or function. We argue that although causality might be necessary for a deep system it is not sufficient on its own. A deep system must manifest the expectations of its user regarding its flexibility as a problem solver and its human-computer interaction (dialogue structure and explanation structure). These manifestations are essential for the acceptability of medical expert systems by their users. We illustrate our argument by evaluating a representative sample of medical expert systems. The systems are evaluated from the perspective of how explicitly they incorporate their particular models of expertise and how understandably they progress towards solutions. The dialogue and explanation structures of these systems are also evaluated. The results of our analysis show that there is no strong correlation between causality and acceptability. On the basis of this we propose that a deep system is one that properly explicates its underlying model of human expertise.