From: ted.dunning@gmail.com   
      
   Yes, all of these systems have rule and learning components.  The rules   
   are useful for the essentially trivial parts of the system and the   
   learning components are useful for the content aspects.   
      
   Having worked on systems with thousands of rules in an academic setting   
   I would run, not walk from any project that tried to field a large   
   rule-based system in any practical setting.   
      
   Take as an example a large financial transaction processing company   
   that I had a hand in working with once.  They had monthly updates to   
   their procedures manuals.  They also had a team of 7 "knowledge"   
   engineers who were charged with reducing this material to a logical   
   form.  The knowledge team never came close to being up to date.   
   Generally, logical updates were so late that by the time they became   
   available, they were old news and had often been superceded by   
   additional changes.   
      
   This was the most successful instance of a traditional AI project that   
   I have ever encountered personally.  It didn't exactly blow apart.  It   
   just never had any positive net value and wound up being completely   
   replaced.  As it turned out, a text retrieval system could give people   
   access to the information they needed with a minimally complex system.   
      
   Contrast this with a system that I built over a decade ago.  The task   
   was to look at radar returns and determine when birds were about to   
   land on a cooling pond next to a power plant.  Since the cooling plant   
   had high levels of dissolved salts and such, the birds have to warned   
   off of landing on the pond.  The system used standard image processing   
   techniques, some minor learning systems and about 5 rules for   
   increasing variety in response.  The same system has been running   
   successfully without change for the last 11 years.   
      
   Anyway, my guiding principle has always been to avoid asking for   
   trouble.  I avoid building large rule systems if I possibly can.   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)