Most of a decade ago I had to do an AI project for a class.  I was a
moderately skilled Unix (tm) system administrator, and had access to
the accumulated error logs of 30 Sun systems for a year.
 
I did "strings" on the Sun OS to get the fixed patterns of all Sun
kernel messages.  We did some form of cat|sort|uniq to categorize the
most common error messages (which, fortunately, were mostly 1-liners).
 
Then we set about parsing the error messages into failure catagories
using the AI program, and writing interpretations for them.
 
The result was quick & dirty, moderately usefull, and taught my 2
cohorts the diagnostic skills of running a Sun system.
 
What made it possible was the ability to bound the problem--in this case
to a fixed number of common error message patterns.
 
It would seem to be a useful project to write an AI program to interpret
LISTSERV error messages--which would perhaps interest some students.
The key requirement to beginning such a project is including the
diagnostics necessary to "snapshot" error conditions, and to begin
accumulating some error logs.
 
Stephen A. McFadden  [log in to unmask]