[Web4lib] AI answering, IBM Watson, whatever...

[David -]

Sure, they can teach a computer to answer reference questions, but can they 
teach it to kick someone off a public access computer for looking at porn?



________________________________
From: Ernest Perez <ernest.r.perez at gmail.com>
To: web4lib at webjunction.org
Sent: Tue, February 8, 2011 2:52:14 AM
Subject: [Web4lib] AI answering, IBM Watson, whatever...

Hi, Web4Lib folks,

Re the recent exchange by Sloan, Balliot, et al, about IBM/Watson AI
question-asking & -answering.

Might something like Watson be a good near-term approach for a small public
or college library? Yeah, if you happen to have a spare 2,500 or so parallel
processor cores lying around. Along with an Industrial-strength IT staff.
(per the IBM system)

Part of the challenge here is that an "answer" is often not a short & sweet
sentence or two. The information and knowledge content will probably be
expressed somewhere on a variable continuum of data and text. The Question
AND the Answer are always dependent on the context of both the question and
the user. So the answer has to be continuously spoon-feedable, until the
user is satisfied.

Twenty years in newspaper libraries, leaning how to use relevance-ranking on
our editorial text databases, and a mid-career Ph.D. convinced me of the
power of using text analysis to produce "answers."

A more productive approach may be using text-mining to examine the "communal
wisdom" in our existing global network. This is perhaps more productive than
AI in the short- and middle-run. Best of all, this approach doesn't have to
bother with the storage, indexing, and retrieval of "everything."

Text-mining applications can use the horsepower our network of search
engines to FIND the topical corpus of information very quickly. The app can
then analyze the content of the retrieved set of information, and doing
whatever filtering may be appropriate. (To avoid all the 273,000 query hits
on Google!)

After retiring from 30+ years of library work, I've affiliated with Power
Text Solutions, Inc. Our current information service application is called
"iResearch Reporter."  (iRR)

iRR uses a complex combination of linguistic analysis, automatic synonym and
term variation identification, on-the-fly specialized vocabulary
identification, text extraction, concept clustering, and formatted summary
reports to provide "answers" from the communal network wisdom.

For samples of  the reports resulting from queries on the (tongue-in-check)
sample questions bandied about during the recent exchange please see:

HOW MANY ROADS MUST A MAN WALK DOWN?
iResearch-Reporter LITE search
*** Results from analysis of 10 Web documents...
http://iresearch-reporter.com/themes/me/321/02_08_07_18_37_output.html
*** Results from analysis of 60 Web documents...
http://iresearch-reporter.com/themes/me/322/02_08_07_26_49_output.html

ANSWER TO LIFE, UNIVERSE AND EVERYTHING?
iResearch-Reporter LITE search
*** Results from analysis of 10 Web documents...
http://iresearch-reporter.com/themes/me/320/02_08_07_10_32_output.html
*** Results from analysis of 60 Web documents...
http://iresearch-reporter.com/themes/me/319/02_08_06_55_33_output.html

Okay, these aren't serious answers to serious questions. But I suppose  you
can say they really do answer THOSE particular questions,

Check out the samples of answers to some REAL user questions at
http://www.irr-usa.com

That site presents detailed information on our LITE and PRO versions of our
information tool approach, There's also a link to the main site, which
offers a 24-hr demo of our LITE product.

(BTW, Web4Lib-ers, if you'd like to subscribe to iResearch Reporter, please
use the registration page on my www.irr-usa.com site.)

Cheers,
   --ernest
----------------------------
Ernest Perez, Ph.D.
Power Text Solutions, Inc.
http://www.irr-usa.com
ernest at irr-usa.com
_______________________________________________
Web4lib mailing list
Web4lib at webjunction.org
http://lists.webjunction.org/web4lib/