| Home | download | source | AI Book | Ruby |

Mycin Expert System

A Ruby Implementation

What is Mycin

Mycin is an expert system that can be programmed with knowledge database for different domains (e.g. medical diagnosis). It asks questions, reasons, gives explanations on why it asked the question, and deals with uncertainties. See the example below.

This is a Ruby rewrite of the Lisp code given in the book:

   Peter Norvig:  "Paradigms of Artificial Intelligence Programming. 
                   Case Studies in Common Lisp"
   Morgan Kaufmann Publishers, San Francisco. 1992
   ISBN 1-55860-191-0
   Original LISP source code is available HERE

Excerpts from the book:

MYCIN lead to the development of the EMYCIN expert-system shell... EMYCIN is a backward-chaining rule interpreter that has much in common with Prolog. However, there are four important differences. First, and most importantly, EMYCIN deals with uncertainty. Instead of insisting that all predications be true or false, EMYCIN associates *certainty factor* with each predication. Second, EMYCIN caches the results of its computation so that they not be duplicated. Third, EMYCIN provides an easy way for the system to ask the user for information. Fourth, it provides explanations of its behavior. This can be summed up in the equation:
EMYCIN = Prolog + uncertainty + caching + questions + explanations
The MYCIN expert system was one of the earliest and remain one of the best known. It was written by Dr. Edward Shortliffe in 1974 as an experiment in medical diagnosis.

A typical rule:

            param  context   operation value
            -----  -------   --------- -----

   (defrule 52                                         <--- rule number
       if  (site   culture         is  blood )
           (gram   organism        is  neg   )         <--- premise/condition
           (morphl organism        is  rod   )
           (burn   patient         is  serious)
       then 0.4                                        <--- confidence factor (cf)
           (identity   organism    is  pseudomonas )   <--- conclusion


There is nothing to install really. Install Ruby first, then


This is a result of a test run, user's entries are in bold. The example is from the book, although some questions are out of order (probably due to random order when reading a Hash).


------- patient-1 -------

Patient's name:  HELP
 Type one of the following:
    ?       - to see possible answers for this parameter
    rule    - to show current rule
    why     - to see why this question is asked
    help    - to see this list
    xxx     - (for some specific xxx) if there is a definite answer
    xxx .5 yyy .4 - if there are several answers with
                different certainty factors

Patient's name:  SYLVIA_FISHER


Age:  27
------- culture-1 -------

From what site was specimen CULTURE-1 taken?  BLOOD

How many days ago was this culture (CULTURE-1) obtained?  3
------- organism-1 -------

Enter the identity (genus) of ORGANISM-1?  UNKNOWN

The gram strain of ORGANISM-1?  ?
Must be one of: acid-fast, pos, neg

The gram strain of ORGANISM-1?  NEG

Is ORGANISM-1 a rod or coccus?  ROD

Is PATIENT-1 a burn patient? If so, mild or serious? WHY
[Why is the value of burn being asked for?]
It is known that:
    The site of the culture is blood
    The gram of the organism is neg
    The morphology of the organism is rod
Rule 52:
    If  The burn of the patient is serious
    Then with certainty of 0.4
        The identity of the organism is pseudomonas

Is PATIENT-1 a burn patient? If so, mild or serious? SERIOUS

What is the aerobicity of ORGANISM-1? AEROBIC

Is PATIENT-1 a compromised host?  YES
Findings for ORGANISM-1
    for these goals: identity

Is there another organism?YES
------- organism-2 -------

Enter the identity (genus) of ORGANISM-2?  UNKNOWN

The gram strain of ORGANISM-2?  NEG 0.8  POS 0.2

Is ORGANISM-2 a rod or coccus?  ROD

What is the aerobicity of ORGANISM-2? ANAEROBIC
Findings for ORGANISM-2
    for these goals: identity

Is there another organism?NO
Is there another culture?NO
Is there another patient?NO

by Zoran Lazarevic, laza@cs.columbia.edu