Knowledge-Based Systems in Medicine

Abstract

In early 1990, Miller and Maserie described their belief that early approaches to computer-based decision support in medicine had been flawed in their adoption of the “Greek oracle” model of consultation [1]—computer programs to which a physician would turn for advice, answering patient-related questions but otherwise deferring to the dialog style and recommendations of the machine. The authors correctly observed that this model, adopted by many knowledge-based systems during the first decade of medical artificial-intelligence (AI) research [2], failed to capitalize on the special observational skills of physicians and failed to allow the user adequately to control the interaction, selecting the ways in which the program’s knowledge could most effectively be brought to bear in the consideration of a particular case. Their redesign of Internist-1’s interactive model in the development of the QMR system reflected their belief that the “Greek oracle” model should be laid to rest and replaced by a much more interactive and user-controlled approach to decision support.