[Sigia-l] Findability

[Donna M. Fritzsche]

At 11:57 AM -0400 7/21/03, Peter Morville wrote:
>Thanks Karl. 
>
>I agree that Google's tremendous success came as a big shock to many of
>us with roots in the LIS community.
>
>However, I'm not (yet) convinced by your attempt to link Google to
>genetic (evolutionary) algorithms.
>
>My understanding is that Google's algorithms have been created by very
>smart humans.  Also, these algorithms are leveraging manually-created
>structural and descriptive metadata (e.g., structure of each web site,
>domain names, title tags, links between sites) in addition to full-text
>indexing.
>
>Do you have evidence Google is using genetic algorithms to enable
>fast-forward software evolution and emergence?  If so, please share!

I agree with a great deal of what is being said here.

  I wanted to step in here representing the point-of-view from the 
AI/supercomputing world. Goggle's success did not come as a shock to 
those of us who knew what the state-of-art was in the late 80's/early 
90s.  The technology that is being used in current search and 
clustering engines has been around that long,it just probably wasn't 
well known amongst the LIS community, just as LIS knowledge wasn't 
well known to the AI community.   I used to work with Brewster Kahle 
and friends at Thinking Machines where we developed and applied 
clustering and indexing tools that over time have migrated into the 
likes of google, vivisimo, and various data mining applications.   (I 
am not saying that the exact same algorithms are still being used, 
but the general bag of tricks probably hasn't changed that much.) The 
tools worked on extremely large datasets inculding 5 years of the 
Wall Street Journal, etc. To some, (especially Brewster), the vision 
was already there in terms of what was possible.

I recently had a conversation with Lou R. about the lack of 
communication between the LIS/IR community and the AI/algorithm 
community, and is seems that historically there had been a lack of 
communication and knowledge sharing in both directions.  Although, I 
do know that there was some attempt from folks at Thinking Machines 
to learn what was happening in the LIS/IR world and visa/versa.  But, 
unfortunately, probably not enough.  It is great that  the gap is now 
closing, because in my opinion that the best solutions in the future 
will be hybrid solutions.

There are different types of AI, its not limited to genetic 
algorithms.  Like Peter, I doubt that google uses genetic algorithms 
and I would agree with his assessment of how google works - which in 
the past might have been considered AI.  By far the best application 
of genetic algorithms that I have seen is the computer graphics work 
done by Karl Sims.  (Search for Karl Sims on vivisimo.com.)  It 
generally considered to be some of the most innovative work of this 
genre.

Got to run, interesting conversation though!

Donna