From: Andrei Alexandrescu (andrei_at_metalanguage.com)
Date: Wed Oct 22 2003 - 15:04:22 PDT
"Two Theses of Knowledge Representation: Language Restrictions,
Taxonomic Classification and the Utility of Representation Services"
Jon Doyle & Ramesh s Patil
It makes little sense to post a review *after* the deadline and
*after* the paper has been discussed in the class, but the spectrum of
the homework deadline looming, added to the cold terror of a paper
deadline getting closer, has obscured this duty.
(Thinking that this review won't be read makes one wonder about the
tree falling in the forrest.)
This paper marked a landmark. It put a question mark above an entire
nascent discipline, showing that its current best methods, idioms, and
ways of measurement are inappropriate when applying them to the real
world. To this day, the KR community is struggling to address the
fundamental issues raised by Doyle and Patil.
Another important point raised by the paper was that in artificial
reasoning in particular, unreliable but usually remarkable performance
is preferable to consistently, reliably mediocre performance. This is
the case with human intelligence, and this is *the opposite* of most
other fields of CS.
Flaws - the paper is a bit long winded and the style used might
reflect some trauma that the authors might have had while trying to
get KR-ONE to work.
Whether or not KR systems, and in which form, are suitable for
implementing sensible artificial reasoning is the most important open
question that the paper authors indebted us with.
Andrei
This archive was generated by hypermail 2.1.6 : Wed Oct 22 2003 - 15:03:58 PDT