RE: Pattern recognition

From: Chris Healey (chealey@unicom-inc.com)
Date: Tue Oct 07 2003 - 11:42:53 MDT


As I understand it, the requirement of banishing ambiguity in all cases
would invalidate a formal systems approach. Since a system cannot be both
consistent and unambiguous in all cases (ala Godel), the formality would
necessarily be broken to banish ambiguity.

Of course, that does not mean that the model will not almost entirely
approach consistency, but it does suggest a greater depth of complexity than
the simple approach would allow for. My guess would be that given a more
complex (inconsistent and unambiguous) system as a model, and given by
definition it is a correct model (highly tested, and not disproven through
empirical data), it would be relatively easy to generate an approximate
formal systems model (consistent and ambiguous) that is very often correct,
but unpredicatably wrong. Starting only with such a formal systems model
however, it would likely be relatively difficult, and perhaps intractable,
to perform the reverse operation, and generate the complex model.

I believe this embodies Yudkowsky's failure-mode of "collapsing the levels
of organization" which tend to appear in many complex systems, especially
intelligence.

-Chris

-----Original Message-----
From: owner-sl4@sl4.org [mailto:owner-sl4@sl4.org]On Behalf Of Perry E.
Metzger
Sent: Tuesday, October 07, 2003 1:01 PM
To: sl4@sl4.org
Subject: Re: Pattern recognition

"Ben Goertzel" <ben@goertzel.org> writes:
> Ambiguity exists in natural language because natural language involves
> *lossy* compression of thoughts and percepts. Lossy compression is
> necessary because to communicate our thoughts and percepts in detail would
> take waaaaay too much time and effort, given practical realities.
>
> Precise, unambiguous language exists: it's called formal logic.

However, formal logic can only speak about objects inside the formal
system. It is unclear how one can unambiguously produce mappings
between the formal system and the real world.

Perry



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT