Some implications of the emergence and diffusion of medical expert systems

This paper has two sections. The first introduces the area of medical decision making and the use of artificial intelligence techniques for developing medical expert systems. It also briefly discusses how a physician and an expert system interact, using the INTERNIST-I system as an example. An elaboration of several technical problems which expert systems must overcome for them to be accepted and used by physicians ends this discussion. The second section considers the implications medical expert systems for physicians and patient care. It suggests that medical expert systems regulate and routinize physicians' work. Moreover, the encoding of expert knowledge in a computer program will likely demystify the practice of medicine, nullifying much of the art of medicine. Alternatively, medical expert systems may improve the physician's decision making, extend the practitioner's capabilities, and hence increase the physician's prestige. Moreover, computer aided decision making could speed diagnosis, especially for difficult cases, thus providing the doctor with time for other pursuits. However, medical expert systems will be a mixed blessing for patient care. On the one hand, they could provide expert advice in medically underserved areas, decrease the number of referrals patients and physicians must contend with, while offering expert advice to a greater number of patients. On the other hand, expert systems could further alienate patients from practitioners and from the health system in general. Even more than other medical technologies, computers in the doctor's office could increase the separation between the doctor and patient, while lowering the physician's confidence in her or his own decision making. The conclusion discusses the importance of control over work in determining the use of technology and the influence of the changing medical power structure on the use of expert systems.I thank J. Zvi Namenwirth for his careful review and thoughtful comments on this and previous drafts of this paper. I also appreciate the useful suggestions offered by anonymous reviewers of an earlier draft.