Instead of trying to write AI programs directly perhaps we   
   should be trying to write formal AI specifications.  Used   
   as a programming language the formal specification might   
   then be transformed by automatic means into Java or   
   other executable code.   
      
   One might start from Werbos' definition of intelligence:   
   "a system to handle all of the calculations from crude   
   inputs through to overt actions in an adaptive way so   
   as to maximize some measure of performance over   
   time" (IEEE Trans. Systems, Man, and Cvbernetics,   
   1987, pg 7) and informal specifications like:   
      
   Task intelligence   
      
        Goal:   
              what action/output do I do now   
      
        Roles:   
              Input:   
                   what do I sense around me now   
      
              Output:   
                   category:   
                           what case best describes the world   
                   evidence:   
                           what list of inputs (and outputs) have   
                           I seen to date   
      
              Spec:   
                   find a set of cases that describe the inputs,   
                   outputs, and utility seen   
      
   Task method   
      
            ..................etc...................................   
            ....................................................   
      
   Perhaps (genetic or other) learning algorithms could be   
   applied to such formal specifications in order to evolve   
   and (hopefully) improve them.   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)