Watson video:
http://www.youtube.com/watch?v=I-gZkqCXOgs
 
Some interesting comments.  It was based on a system that took results from a 
number of probabilistic analytical methods that included natural language 
systems.  Two things.  Most nlp systems use highly structured relations between 
described parts of language as references for the statistical methods that they 
use.  So Watson does not demonstrate how language may be learned.  Secondly, as 
Watson solved the problem it had to handle a combinatorial explosion (of a 
manageable size).  One of the problems that we have is that even if we were 
able to begin writing programs of that kind of complexity we still would not be 
able to develop them on our computers because the run time would be too slow.  
(However, I am struggling with relatively simple programming problems so I am 
not even sure that is a relevant issue for me.)
 
I don't think that statistical methods are the way to go at this point.  The 
only evidence from Watson that I have to offer is that Watson did not truly 
learn a language, much of the structure of a language was incorporated into the 
model that they used and the statistical methods were derived from corpuses 
derived from statistical studies. However, even though I would use statistical 
models in a more constrained way than the contemporary prevailing paradigms 
such as Watson,  I would still use multiple paths to discovering possible 
'solutions' and multiple paths to evaluate the possible 'solutions'.
 
Jim Bromer
                                          


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to