Attempts to use computers as diagnostic aids in medical decision making: a thirty-year experience.

For more than 30 years our group of physicians, statisticians, and computer scientists has worked toward developing a computer program with the capability of a trained physician to make diagnostic decisions in the relatively broad medical subspecialty of hematology. We devised and tested many programs, none of which have been sufficiently useful to warrant carrying beyond the pilot-study stage. We analyzed the reasons for this failure. Our experience confirms the great difficulty, and even the impossibility, of incorporating the complexity of human thought into a system that can be handled by a computer. We concluded that we should stop trying to make a computer act like a diagnostician and concentrate instead on ways of making computer-generated relevant information available to physicians as they make decisions.