XPost: alt.folklore.computers   
   From: antispam@fricas.org   
      
   In alt.folklore.computers Lawrence D’Oliveiro  wrote:   
   > On Thu, 1 Jan 2026 10:30:54 -0000 (UTC), Waldek Hebisch wrote:   
   >   
   >> But are 'expert systems' really AI?   
   >   
   > What is really “AI”? At one point, the argument was over whether   
   > computers could “think”. Then you had to define “thinking”, and   
   > somebody tried to settle the question by saing: “thinking is what   
   > computers cannot do”.   
   >   
   > The only succinct definition of “AI” I ever saw was: “solving NP   
   > problems in polynomial time”.   
      
   Well, for me AI is process (and its results) of trying to solve   
   problems that we can not solve using known (at given time) methods   
   and which seem to require inteligence.  Important aspect is   
   that "solving" is supposed to produce a method that works not   
   for a single problem, but generalized to some class of problem.   
      
   One could split hairs asking if AI obtains some result and it   
   becomes well-known, is it still AI?  As an example, early AI   
   spent a lot of effort on search.  Regardless if now we consider   
   resulting alogrithms AI or not, important results were obtained   
   as part of AI research.   
      
   Coming back to 'expert systems', principle "you need domain   
   expert" seem to indicate the approach really was "study the   
   domain, once you collect enough knowledge you can write a   
   program".  Which is what a good programming book of that   
   period told you to do.  So more refined question may be: did   
   'expert systems' bring something new an valuable?  They   
   boasted about a handful of successful systems, but details   
   about them seem to be quite sparse.  And hard to see   
   anything systematic.   
      
   In "LESSONS LEARNED; Building Expert Systems" (AI EXPERT,   
   SEPTEMBER 1988) A. Bahil, P. Harris and E. Senn wrote:   
   "If it would take a human two days to solve the problem,   
   it is far too complicated for an expert system."  They   
   also wrote that most system they looked at did not use   
   uncertainty factors or explanation capability of some   
   expert system shells.  That looks for me as rather low   
   limitation on complexity of tasks solved by expert   
   systems and that solutions typically did not use fancy   
   features of expert system shells.   
      
   BTW: I asked a question.  I would love to see an example   
   showing that 'expert system' methodology or tools really   
   made a difference in solving a problem.  But I was not   
   able to find such an example.   
      
   --   
                                 Waldek Hebisch   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)