# MYCIN

MYCIN was an early 1970s expert system developed at Stanford University to identify bacteria causing severe infections and recommend antibiotics, using a rule-based inference engine and certainty factors. It is considered a pioneer in the field of expert systems.

MYCIN was an early [artificial intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) program developed in the early 1970s at [Stanford University](https://www.wikiprompt.org/wiki/stanford-ai-lab) to assist physicians in diagnosing and treating severe bacterial infections, such as bacteremia and meningitis. It used a backward chaining inference engine and a knowledge base of approximately 600 production rules to reason about patient data and recommend antibiotics, with dosages adjusted for body weight. The name derived from the suffix "-mycin" common to many antibiotics. MYCIN is widely regarded as a foundational expert system, with computer scientist Allen Newell calling it the "grandaddy of them all - the one that launched the field."

MYCIN was written in Lisp as the doctoral dissertation of Edward Shortliffe, under the direction of Bruce G. Buchanan, Stanley N. Cohen, and others. It emerged from the Stanford Heuristic Programming Project, which also contributed to other early AI systems. The program demonstrated that domain-specific knowledge, clearly separated from inference procedures, could produce high-performance reasoning in a medical context.

## Method

MYCIN operated by querying the physician through a series of simple yes/no or textual questions. At the end of the session, it provided a ranked list of possible culprit bacteria, each with a confidence level, the reasoning behind each diagnosis, and a recommended drug treatment. The system could also answer follow-up questions from physicians, such as why it asked a particular question, how it reached a conclusion, or why it did not consider certain factors.

The knowledge base consisted of rules encoded by experts, each representing an individual inferential fact. At the time, no other AI program contained as much domain-specific knowledge clearly separated from its inference engine. The inference engine used backward chaining, starting from a goal hypothesis and working backward through rules to find supporting evidence.

Developers conducted studies showing that MYCIN's performance was minimally affected by perturbations in the uncertainty metrics associated with individual rules. This suggested that the system's power came more from its knowledge representation and reasoning scheme than from the details of its numerical uncertainty model. Some observers argued that classical Bayesian statistics should have been used, but the developers countered that this would require either unrealistic assumptions of probabilistic independence or an unfeasibly large number of conditional probability estimates from experts.

## Certainty Factors

MYCIN used a certainty factor model to combine evidence from multiple rules. Each rule could conclude a hypothesis with a weight between -1 and +1, where -1 represented "false," +1 represented "true," and 0 represented "uncertain." When two or more rules drew conclusions about the same parameter, MYCIN combined their certainty factors using a specific formula.

The combination formula was designed to have several desirable properties. Combining an unknown value with anything left it unchanged. Combining true with anything except false gave true, and similarly for false. Combining true and false resulted in a division-by-zero error. Combining +x and -x gave unknown. Combining two positives gave a larger positive, and combining a positive and a negative gave a value in between.

Subsequent studies showed that the certainty factor model could be interpreted in a probabilistic sense, though they also highlighted problems with the implied assumptions. However, the modular structure of the system proved influential, leading to the development of graphical models such as Bayesian networks.

## Contexts

A context in MYCIN determined what types of objects could be reasoned about. Contexts were similar to variables in Prolog or environment variables in operating systems. They allowed the system to organize knowledge about different aspects of a medical case, such as the patient, the culture, or the organism being identified.

## Legacy

MYCIN led to the development of EMYCIN ("essential MYCIN"), an expert system shell that could acquire knowledge, reason with it, and explain results without the specific medical knowledge. EMYCIN was described as "EMYCIN = Prolog + uncertainty + caching + questions + explanations + contexts - variables." It allowed developers to build expert systems in other domains by replacing the medical knowledge base.

MYCIN's influence extended beyond its immediate application. It demonstrated the potential of rule-based systems for [machine learning](https://www.wikiprompt.org/wiki/machine-learning) and reasoning, and its approach to knowledge representation influenced later work in [AI](https://www.wikiprompt.org/wiki/artificial-intelligence) and expert systems. An introduction to MYCIN appears in Chapter 16 of the book *Paradigms of Artificial Intelligence Programming* (PAIP).

Despite its technical success, MYCIN was never widely deployed in clinical practice, partly due to legal and ethical concerns about computer-generated medical advice. Nevertheless, it remains a landmark in the history of AI, illustrating both the strengths and limitations of early expert systems.

## See Also

- [Xerox PARC](https://www.wikiprompt.org/wiki/xerox-parc)
- [MIT CSAIL](https://www.wikiprompt.org/wiki/mit-csail)
- [Carnegie Mellon University](https://www.wikiprompt.org/wiki/carnegie-mellon-university)
- [Berkeley AI Research](https://www.wikiprompt.org/wiki/berkeley-ai-research)

---
Source: https://www.wikiprompt.org/wiki/mycin
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-08T06:10:26.968574+00:00