MYCIN: A Rule-Based Expert System for Infectious Disease Diagnosis, Developed at Stanford University
Beginning around 1972, Edward Shortliffe at Stanford University developed MYCIN, a rule-based expert system written in Lisp that used approximately 600 if-then rules to diagnose bacterial blood infections and recommend antibiotic treatments adjusted for patient body weight, establishing a widely studied model for clinical decision support.
Background
By the early 1970s, computers had been used in medicine for about a decade, mostly for record-keeping and statistical work. Getting a machine to reason through a clinical problem was a different matter entirely. The programs that could do something like that were few, brittle, and mostly confined to laboratory demonstrations.
The underlying difficulty was not processing power. It was knowledge. A physician diagnosing a bacterial infection draws on years of training and a web of conditional judgements: which organisms are likely given this patient’s history, which antibiotics are effective against those organisms, which doses are safe given the patient’s weight and kidney function. Nobody had found a clean way to put that kind of knowledge into a computer and have it reason with it reliably.
There were also practical pressures. Bacteraemia, a bacterial infection of the blood, required fast treatment. Waiting days for laboratory cultures before starting antibiotics could cost a patient’s life. Clinicians were often making educated guesses about which organism was responsible and which drug to use, before the full evidence was in. A system that could reason through the same problem systematically, and explain its reasoning, would have genuine clinical value.
What happened
Around 1972, Edward Shortliffe, then a doctoral student at Stanford University working with the Stanford Heuristic Programming Project, began building a program to address exactly that problem. His collaborators included Bruce Buchanan, a computer scientist, and Stanley Cohen from the Stanford Medical School, who provided the clinical knowledge the system needed.
The result was MYCIN. It was written in Lisp and built around two separate components: a knowledge base of roughly 600 if-then rules encoding what the clinical experts knew, and an inference engine that worked through those rules against the facts of a particular case. Keeping the two parts separate meant that updating the medical knowledge did not require rewriting the reasoning machinery. That design choice would later prove influential well beyond this one project.
MYCIN would ask a physician a series of questions about the patient: symptoms, laboratory results, recent procedures. From the answers, it worked backwards through its rule base, a method called backward chaining, to identify which organisms were the most probable cause of the infection. It then recommended specific antibiotics along with dosages adjusted for the patient’s body weight. Shortliffe described the system in detail in his 1974 doctoral thesis and in a 1976 monograph published by Elsevier. A preliminary account appeared even earlier, in a 1973 paper in Computers and Biomedical Research, co-authored with colleagues including Buchanan and Cohen.
The system was evaluated against the judgements of infectious disease specialists, and those evaluations, published in the late 1970s, found that MYCIN’s recommendations were broadly comparable to those of human experts in its narrow domain. It was a research system and was never deployed in routine clinical practice. But what it showed about encoding specialist knowledge as explicit, inspectable rules gave the field a concrete model to build from.
Why it mattered
MYCIN demonstrated that a computer system encoding domain expert knowledge as explicit rules could match or exceed the diagnostic accuracy of specialists in a narrow clinical domain, a result formalised in controlled evaluations published in the late 1970s. Its architecture, separating a knowledge base of rules from a general inference engine, became the canonical template for expert systems research through the 1980s. The project also raised foundational questions about how to evaluate AI systems in high-stakes settings, influencing later work in medical informatics and AI safety.
People
Edward H. Shortliffe, Bruce Buchanan, Stanley Cohen, Avram Bar Hillel
Organisations
Stanford University, Stanford University School of Medicine, Stanford Heuristic Programming Project
Sources
- Computer-Based Medical Consultations: MYCIN.Elsevier / North-Holland (Springer reprint via DOI).Primary source
- Edward H. Shortliffe Papers, MYCIN project materials.Stanford University Libraries, Special Collections.Secondary
- An Artificial Intelligence Program to Advise Physicians Regarding Antimicrobial Therapy (Shortliffe et al., 1973, Computers and Biomedical Research).Academic Press via ACM Digital Library.Primary source
Cite this page
AI Achievements. (1972). MYCIN: A Rule-Based Expert System for Infectious Disease Diagnosis, Developed at Stanford University. Retrieved 2026-08-22, from https://achievements.ai/milestone/mycin-expert-system-stanford-university
@misc{achievements_mycin_expert_system_stanford_university,
title = {MYCIN: A Rule-Based Expert System for Infectious Disease Diagnosis, Developed at Stanford University},
author = {{AI Achievements}},
year = {1972},
url = {https://achievements.ai/milestone/mycin-expert-system-stanford-university}
}