Paul
Werbos,
I
agree fully with your comments on the nature of
"intelligence"
Clearly, "intelligence" is a natural language concept -- it's fuzzy,
ambiguous, and it was created by humans for a certain collection of
purposes.
It was
not created for dealing with other species loosely as intelligent as humans, nor
for dealing with programs like Deep Blue, Novamente, or Paul Prueitt's eC/cA
system... or the internet...
When I
create a notion like "the ability to achieve complex goals in complex
environments" and call it "a definition of intelligence", I don't mean to imply
that this definition fully captures the intuitive natural language notion of
"intelligence." It's just a more crisply defined notion that captures A
LOT of the natural language notion of intelligence, and is A LOT more convenient
to work with.... Shane Legg has proposed to call my notion of intelligence
"cybernance" and that's fine with me too.
However, when Paul Pruiett says that "no computational system can ever be
intelligent", what I want to know is what HE means by the word
"intelligence." Does he believe this proposition holds under my
notion of cybernance/intelligence, i.e. would he say "no computational system
can ever be nearly as cybernant as humans"? Does he mean "no computational
system can ever be as intelligent as humans under the ambiguous natural language
notion of intelligence"? Or something else??
Ben
G
-----Original Message-----At 07:49 AM 11/07/2002 -0500, Paul Prueitt wrote:
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Werbos, Dr. Paul J.
Sent: Thursday, November 07, 2002 10:57 AM
To: Paul Prueitt; [EMAIL PROTECTED]
Cc: NaturesPattern
Subject: [agi] Re: on the notion that intelligence can be defined
Ben,
The definitions of intelligence might be derived from the notions of life given by Maturana and Varela. Autopoietic is the term used there.
Hi, folks!
As I understand it, my role here is mainly just as a quiet occasional observer -- which is fine, since I don't generally
have time to do justice to all the threads here.
But this question -- "Is it possible to define intelligence?" -- is somewhat worrisome.
I am not a specialist in the natural language side of things -- but it is an elementary, important fact of life that
words like "intelligence" are just words -- just grunts or noises -- until or unless WE CHOOSE to attach
some definition to them.
BEFORE there is SOME kind of definition, it is just a grunt, like any other kind of grunt,
and there is no magical reason why one can or cannot attach a definition to this grunt, like any other grunt.
We cannot hope to address the question of "defining intelligence" unless this foundation is
clear and assumed and maintained.
I have heard many debates about "Do you have the RIGHT definition of intelligence?" which are utterly
unanchored in this foundation. The question to start with may be: "What do we mean by a RIGHT definition
anyway? What are we looking for?"
There are roughly two or three reasonable sorts of responses --
1. Definitions are arbitrary in principle, but it would be desirable to pick a definition which corresponds to something
in reality... which fits the higher-order capabilities observed in the brains of real animals...
2. It would be REALLY nice to pick a definition which is also sharp enough that it can be realistically applied
to mathematical design -- specifies what we want well enough that it is meaningful to struggle to achieve it.
Fuzzy as this is, it is enough to capture the main need here, in my view.
Many people developing ""intelligent can-openers" would argue that this is a chauvanistic attempt to
deprive them of their right to market their can-openers as they see fit. There are many reasons why
I would not bow down to that complaint.
=== But, time being limited... I would move on. Just one final thought... my point here is that we
should try to avoid getting bogged down in the kind of debate or rumination which boils down to a fight over semantics.
