From: saalweachter@purdue.edu   
      
   makc.the.great@gmail.com wrote:   
   > Dmitry A. Kazakov wrote:   
   >   
   >>2. Simulating human behavior is a questionable issue as well. It boils down   
   >>to Turing test. But how our *inability* to decide (machine vs. human) could   
   >>characterize anything as intelligent?   
   >   
   >   
   > Agree. What if AI will be more rational than humans? Then, they may   
   > fail Turing test very well because human judge himself would not be   
   > able to comprehend supreme rationality, and will consider machine   
   > behavior nothing better than random noise.   
      
   I think this view gives too little credit to both 'supremely rational   
   beings' and 'humans'.   
      
   When you ask the supremely rational AI, during the Turing test, "So, how   
   'bout them Bears?", do you expect it to spit out a terse "ERROR.   
   BASEBALL IS ILLOGICAL."?  Or would you prefer that the exsquisitly   
   rational being launch into a perfectly thought out, well reasoned   
   analysis of how they're doomed this season, having traded Larry Bird to   
   the Nicks for Roy Orbison, when what they REALLY need is a fullback?   
      
      
   There is certainly a computational tie-in with reasoning: proof   
   construction is an NP-complete problem.  Even if you assume that the   
   entity has 'sufficient knowledge', it's still going to take an   
   exponential amount of time to infer arbitrary conclusions from this   
   knowledge.   
      
   However, consider that, while proof-construction is NP-complete,   
   proof-checking is not.  This means that if we had some exquisitely   
   rational fellow sitting around, he might occassionally reach conclusions   
   we never could.  Why?  He's just a better reasoner; he constructs proofs   
   that are beyond our capacity to construct.  HOWEVER, it is perfectly   
   reasonable to expect that he could then /explain/ his reasoning to us,   
   and we could say, "Oh, that makes perfect sense."  Checking a proof is   
   computationally simple, and we should expect ourselves to be able to   
   check proofs we ourselves could never discover.   
      
      
   Thus, 'perfectly rational' or even just 'spectacularly rational' AI   
   shouldn't seem incomprehensible, irrational to humans.  Its reasons   
   should be utterly simple to comprehend -- and utterly impossible to   
   discover.   
      
      
   Jack Saalweachter   
      
   [ 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)