From: mailbox@dmitry-kazakov.de   
      
   On Tue, 23 Sep 2003 22:42:59 GMT, larix_occidentalis@yahoo.com   
   (Western Larch) wrote:   
      
   >"Dmitry A. Kazakov"  wrote:   
   >   
   >[...]   
   >   
   >> Right, it is to death beaten issue.   
   >   
   > ... which is the first relevant remark in the whole debate.   
   >   
   >No wonder the AI field seems to be moving so slowly -- its proponents   
   >waste time in language wars from the eighties.   
      
   There is also a saying that people are interesting in artificial   
   intelligence, mostly because they are lacking in a natural one. (:-))   
      
   >But seriously, folks, whether you use one common programming   
   >language or another is mostly (not completely, but mostly) irrelevant.   
   >If only there were a language that came close to expressing the   
   >reasoning needed for natural language processing, vision, etc.   
   >Existing programming languages are not exactly "too low level" --   
   >the more serious problem is that no notion of the problem domains   
   >of interest are built into them. I don't mean that there are no   
   >built-in classes such as Image or AuditoryExperience, I mean that   
   >there are no mechanisms for effectively reasoning about those things,   
   >even if there were such objects.   
      
   Mmm, actually the trend in language design is to reduce the number of   
   built-in types, but to provide instead a better abstract data type   
   (ADT) support, so that the user might develop its own types. One   
   cannot foresee and built-in everything. Then the cross dependencies   
   between built-in types and issues of concurrency, persistency,   
   distributed systems, genericity, aggregation produce a geometric   
   explosion of variants. It is a dead end.   
      
   >Over the years I've become convinced that an appropriate programming   
   >language for AI, whatever that is, will be much more strongly   
   >declarative than common programming languages. That is, the programmer   
   >would state the relationships between things -- then how you get from   
   >"what I know" to "what I should do" is something that can be   
   >derived by turning the crank, in this case, applying rules of   
   >decision analysis. Let a compiler turn the crank for you.   
      
   Probably true.   
      
   >It would be much more efficient to spend time thinking about   
   >issues like, "What can be inferred about the present position of   
   >a bird from past images of it?"  At present, unfortunately, such   
   >reasoning is deeply mired in irrelevant issues; to solve any   
   >minor reasoning problem requires solving ten problems about an   
   >ordinary programming language.   
   >   
   >So, here's a question that's more interesting than type systems.   
      
   You can't slip past this issue. Types are fundamental to programming   
   languages. The things you are talking about can be built only if the   
   problem of ADT be firstly clearly understood and secondly resolved.   
   Presently it is neither.   
      
   >What would a general purpose language for "AI" (whatever that is)   
   >look like? Is there such a thing yet, even in prototype?   
   >There were "AI" languages back in the eighties. Why didn't they work?   
      
   Because they didn't solve that fundamental problem of ADT. They saw   
   water, and thinking that it is a puddle, they tried to jump over an   
   ocean ...   
      
   ---   
   Regards,   
   Dmitry Kazakov   
   www.dmitry-kazakov.de   
      
   [ 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)