Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!Germany.EU.net!zib-berlin.de!easix!news.uni-stuttgart.de!rz.uni-karlsruhe.de!stepsun.uni-kl.de!uklirb.informatik.uni-kl.de!schmid
From: schmid@informatik.uni-kl.de (Klaus Schmid)
Subject: Re: Re2: Progress of CYC
Message-ID: <1995Feb13.133805@informatik.uni-kl.de>
Sender: news@uklirb.informatik.uni-kl.de (Unix-News-System)
Nntp-Posting-Host: skolem.informatik.uni-kl.de
Organization: University of Kaiserslautern
Date: Mon, 13 Feb 1995 12:38:05 GMT
Lines: 62

walid@research.att.com wrote:

> CYC and similar projects that claim to have figured "a" way
> of dealing with the representation of common snese knowledge
> are doomed to fail since they are based on a very simplistic
> model of cognition. The problems that such systems face are:
> - ad hoc representation (we still have NO CLUE as to what are
>   concepts, let alone how they are represented)
>
>  we have no CLUE as to what sort of semantic primitives do
>  we have, if any. 
>
>  the ontologies of such representations are COMPLETELY ad hoc.
>
>  learning (esp. language learning) is still a dark tunnel.
>
>  cognitive models that account for time constraints and
>  constraints on working memory are still in their infancy.

It may be true that no adequate model of cognition exists which could
provide a base for the CYC-system (saying we have NO CLUE is an exageration,
I think). However you are making the unsubstantiated assumption that
a project like CYC should be based on such a model. However any model 
which allows such a system for arriving at reasonable answers when asked 
every-day questions will do.

According, to your argumentation there could never have been a working
expert system, because we have no (exact) model of information processing.

I think, the basic assumption  underlying projects like CYC is, that there
are many ways for achieving a goal. One must not use the same methods 
humans use for achieving a goal. However, such knowledge may be helpful.
If that assumption is not correct, then we all have to turn to 
(refined) neural network models in the end ;-)

>  So before we get too excited and make the same mistakes that
>  AI did years ago (by overselling itself) lets be humble.
>  Before we claim we can represent millions of knowledge
>  units (KUs) of the world, lets us try to figure out
>  how two concepts are related to each other in our cognitive
>  universe (how and why does one concept trigger another
>  in certain contexts, and what is actually stored there
>  in long-term memories?) Minsky's society of mind (his
>  little agents...) might be a starting point. As far as
>  CYC - hopefully AI will not put down, AGAIN, becuase it
>  made BIG claims and oversold itslef yet another time!

As far as CYC is concerned I get the impression that more came out of
this project then from most other AI projects up to now. (More effort
was spent on this project, too.) Still, it may fall short of reaching 
some goals, but it seems to satisfy others, not initially proposed
when the project started.

Research in cognition will help (and has helped) AI in reaching
some goals, however, it is not the only way.

Klaus


----
Klaus Schmid 
schmid@informatik.uni-kl.de
