Sergio is indeed completely off-base.
You, Alan, are getting v. warm when you say we need systems that can
autonomously learn - although it would be better to talk of autonomous
discovery learning (otherwise it implies to many that the system will learn
a pre-defined course of action, as distinct from *creating* one for itself
for the first time).
We would like robots that can like an infant discover for themselves how to
pick up different objects or walk on new,different terrains - and in
general find/discover their way around the world.
If your infant robot has to pick up a strange new object - a toy soldier
say - or walk on a water bed, or sofa for the first time- there are no sets.
He will not attempt to fit that object into a *set* of similar objects and
then search a *set* of pre-existing picking-up or walking actions. He will
have a loose idea of picking up, and adjust that to the real object as he
goes along/reaches out. Ditto he will have an idea of walking and adjust
that to the terrain.
There will be no systematic searching of, or application of, any
pre-existing set of options. Do infants' groping actions look like that is
happening? There are zillions of infant vids out there to test theories
here.
Real world problemsolving and action are not like a chess program, searching
a set of options -or a more sophisticated Bayesian program.
Sergio's and similar theories may be powerful for maths or logic, but they
have nothing to do with AGI (and the test Sergio is - where are the real
world problem applications? We haven't seen any - and won't).
--------------------------------------------------
From: "Alan Grimes" <[email protected]>
Sent: Friday, June 22, 2012 6:18 PM
To: "AGI" <[email protected]>
Subject: Re: [agi] Prediction Did Not Work (except in narrow ai.)
Mike Tintner wrote:
Sergio:But there is one thing that is infinite: Mike's infinite variety.
I KNOW, REPEAT, I KNOW! DON'T TELL ME THIS AGAIN! I know that there is
an infinite variety
Sergio,
You didn't get it. There are NO SETS in real world problems. Not
infinite sets.None. Try naming one where there is a set - from
scientific to tech. problems to everyday practical problems: What shall
have I lunch for today? What shall I watch on TV or the net tonight?Or:
What are we going to do re the Euro crisis? How can we cure cancer? Set,
please.
Mike:
Sergio is not completely off-base here. Set theory is extremely
powerful, so powerful that much if not all of the rest of mathematics
can be expressed in terms of set theory.
I now need Sergio to do two things:
1. Propose a strategy to convert a realistic sensory perception, such as
a pixel matrix or a tono-spatial map of binaural hearing into the kind
of set he proposes.
2. Show that his algorithm can produce useful results when fed input
from 1. (ie, the recognition of a shape or the identification of a
sound, etc...)
There seems to be a very good chance that he will be able to pull this
off. But then the burden of pruf is on him...
That said, I wish to hell he'd stop professing to know stuff about
neurology, such as that his algorithm has nearly perfect explanatory
power over all 200 types of neurons... There are a lot of critically
important neural sub-systems in the brain. If you study the overall
architecture of the system, you will see that there are roughly four
different layers of organization, each one *MODULATES* a simpler
behavior implemented by the next lower level. So no, the cortex does not
produce behaviors, it merely modulates and organizes behaviors produced
by lower levels, such as the thalamus, hypothalamus, and spinal nuclei.
So in order to have an AGI system capable of doing anything useful, you
MUST build it as a complete cybernetic agent. No sub-set of an AGI mind
will be able to solve anything beyond toy problems. You MUST HAVE A
COMPLETE AGENT WITH AUTONOMOUS LEARNING!!! There's no partial solution.
Either it is sufficiently autonomous to learn (and hence act) on its own
or you will be spending the rest of your life spoon-feeding it
information.
--
E T F
N H E
D E D
Powers are not rights.
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/6952829-59a2eca5
Modify Your Subscription:
https://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com
