W3C home > Mailing lists > Public > semantic-web@w3.org > May 2009

Re: [ontolog-forum] Research Illusion

From: John F. Sowa <sowa@bestweb.net>
Date: Sun, 10 May 2009 11:49:40 -0400
Message-ID: <4A06F794.1080707@bestweb.net>
To: "[ontolog-forum]" <ontolog-forum@ontolog.cim3.net>
CC: 'SW-forum' <semantic-web@w3.org>, Mustafa Jarrar <mjarrar@cs.ucy.ac.cy>, jeremy@topquadrant.com, Sören Auer <auer@informatik.uni-leipzig.de>, Pieter De Leenheer <pdeleenh@vub.ac.be>
Azamat,

There is a fundamental problem about evaluating new ideas of any
kind:  People always interpret new information in terms of their
previously established mental patterns and structures.

That point has several implications:

  1. Anything that fits previously established patterns will be
     quickly perceived, interpreted, accepted, and added to the
     old patterns.

  2. Anything that doesn't fit the old patterns will be "anomalous".
     It won't fit, it will create "cognitive dissonance", and it
     will be ignored or rejected.

  3. Even worse than outright rejection is the misinterpretation
     caused by forcing the new information into some older
     pattern that is inappropriate and misleading.

This is true of all kinds of learning from infancy to the most
sophisticated scientific research.  One of my colleagues at IBM
submitted a paper to a conference, and one of the reviewers
rejected it with the comment "I never saw anybody do anything
like that before."  Apparently, they wanted new research, but
only if it fit the old paradigms.

Eventually, the author managed to get the paper accepted by
different reviewers, and the paper became a minor classic of
its kind.  This is just one of many examples of "reviewer roulette",
which has plagued every branch of science and engineering.  For
the humanities, the problem is even worse because the criteria
for testing ideas by experiment are much harder to apply.

The same kinds of prejudices plague entire fields, not just
individual reviewers.  During the 1970s and '80s, another colleague
at IBM, Fred Jelinek, was the manager of a group that used statistics
to analyze natural languages.  In those days, the amount of data
they had to process was so large that they swamped a large IBM
mainframe.  So they had to run their programs at 3 o'clock in
the morning, when they could get enough computing power.

I remember that one of the researchers who worked for Fred had
developed a parser that used statistics to guide the choice of
which option to follow.  She submitted the paper to IJCAI in 1981,
and it was rejected with the statement "Statistics is not AI."

By the 1990s, personal computers were as powerful as the mainframes
of the early 1980s.  So the same kinds of techniques could be run
on PCs, and Fred Jelinek became a guru instead of a crank.  Now,
statistics is the so-called "mainstream", and papers are often
rejected if they don't use statistics.

Some comments on earlier comments:

> SA: "I have the vision that research communities' crowd intelligence could 
> be employed in the Web 2.0 style for deciding about research funding".
> 
> MB: "...we see people can vote resources...Allowing people to add 
> ontology-based annotations is just similar and would be another step 
> forward."
> JC: "Google scholar provides citation counts, which while still a fairly 
> rough measure, does include an idea of the importance of any piece of work."
> 
> PDeL: "I agree with the value of the wisdom of the crowd effect in many 
> cases, however it should be controlled somehow to prevent the emergence of 
> "foolishness of the crowd".
> 
> MP: "We second the idea of common standard ontologies for the semantic web
> use."

All of those techniques can be helpful, but none of them are magic.
They still won't overcome the fads of the "mainstream", and they still
can't distinguish a truly significant innovation from the latest fad
or somebody's pet idea "that just ain't so".

> AA: I incline to think that the "crowd intelligence" or "foolishness of the 
> crowd" may explain the nature of the "phenomenon", and a canonic world model 
> encoded as a machine-understandable common ontology standard of meanings 
> may allow to head off it at all.

Perhaps.  But my greatest fear is that the choice of standard is more
likely to be determined by the latest fad or by the organization that
has the most hype and money to throw at it.  To avoid embarrassing
the guilty, I won't mention specific examples.  But for anybody who
advocates a common standard ontology, I would say

     Be careful what you wish for.

If any readers think that they have an ideal ontology in mind, I'd
like to ask one question:  Do you believe that you have sufficient
hype and money to make your preference become the new mainstream?

John Sowa
Received on Sunday, 10 May 2009 15:50:18 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 21:45:29 GMT