INTERNIST-I Developed by Jack D. Myers and Harry E. Pople Jr.
Jack D. Myers and Harry E. Pople Jr. at the University of Pittsburgh developed INTERNIST-I, an expert system for internal medicine diagnosis, with its principal public evaluation published in the New England Journal of Medicine in 1982. The system encoded knowledge of roughly 500 diseases and 3,500 symptoms, demonstrating that algorithmic clinical reasoning could approach specialist performance on complex cases.
Background
Doctors diagnosing complex internal medicine cases draw on an enormous body of knowledge: hundreds of diseases, thousands of possible symptoms, and the awareness that several conditions can be present in the same patient at the same time. By the 1970s, that scale of knowledge was exactly what made the diagnostic problem hard to automate. Researchers had built programmes that handled narrow, well-defined tasks reasonably well, but internal medicine was not narrow. A realistic case might involve overlapping diseases with overlapping symptoms, and no straightforward rule-based shortcut would get you through it.
The approach most systems used was pattern recognition: match what you observe against stored templates. That worked well enough when cases were clean and typical. It struggled when they were not, and real clinical cases frequently are not. What was missing was something closer to the way an experienced physician actually reasons: weighing evidence, considering and discarding alternatives, and following diagnostic logic rather than pattern similarity.
What happened
Jack D. Myers, a senior internist at the University of Pittsburgh, and Harry E. Pople Jr., a computer scientist working alongside him there, spent roughly a decade building INTERNIST-I. Myers supplied the clinical knowledge; Pople built the computational framework to make it usable. The system encoded about 500 diseases and around 3,500 findings, where findings means symptoms, signs, and laboratory results. Each disease was linked to the findings associated with it, along with numerical weights expressing how strongly each finding supported or argued against that diagnosis.
When a clinician entered a patient’s findings, INTERNIST-I generated competing diagnostic hypotheses and scored them against the evidence. It was not matching a pattern. It was working through a kind of differential diagnosis, the process of systematically considering and ruling out possibilities, in a way that reflected how a specialist might reason through a complicated case.
The principal public evaluation appeared in the New England Journal of Medicine in 1982. The results were examined alongside assessments by clinicians, and the system showed it could reason at a level that invited serious comparison with specialist performance on difficult cases. The evaluation also documented where it fell short: when a patient had more than one illness simultaneously, the system’s architecture did not handle the interaction between competing diagnoses well. That limitation was clearly identified rather than glossed over, and it became the starting point for follow-on work.
That follow-on work, a revised system called CADUCEUS, tried to address exactly the multiple-illness problem INTERNIST-I had exposed. The 1982 paper gave the research a public, peer-reviewed record in one of medicine’s leading journals, which brought the work into direct contact with clinical readers who had no background in computing.
Why it mattered
INTERNIST-I was one of the most knowledge-intensive expert systems built to that point, encoding years of a senior clinician's diagnostic expertise in a structured, computable form. Its published evaluation in a leading medical journal brought expert-system research into direct dialogue with clinical medicine, influencing subsequent work on decision-support tools. The system's documented weakness with multiple concurrent illnesses in a single patient also defined a clear research problem, later addressed in CADUCEUS, that shaped the next generation of medical AI.
People
Jack D Myers, Harry E Pople Jr
Organisations
Sources
- The INTERNIST-I / CADUCEUS Project.New England Journal of Medicine.Primary source
- The INTERNIST-I / CADUCEUS projects: retrospective and current directions.ACM.Secondary
Cite this page
AI Achievements. (1982). INTERNIST-I Developed by Jack D. Myers and Harry E. Pople Jr.. Retrieved 2026-08-22, from https://achievements.ai/milestone/internist-by-jack-myers-and-harry-pople
@misc{achievements_internist_by_jack_myers_and_harry_pople,
title = {INTERNIST-I Developed by Jack D. Myers and Harry E. Pople Jr.},
author = {{AI Achievements}},
year = {1982},
url = {https://achievements.ai/milestone/internist-by-jack-myers-and-harry-pople}
}