Infants *start* with general learning skills - they have to extensively
discover for themselves how to do most things - control head, reach out, turn
over, sit up, crawl, walk - and also have to work out perceptually what the
objects they see are, and what they do... and what sounds are, and how they
form words, and how those words relate to objects - and how language works
it is this capacity to keep discovering ways of doing things, that is a major
motivation in their continually learning new activities - continually seeking
novelty, and getting bored with too repetitive activities
obviously an AGI needs some help.. but at the mo. all projects get *full* help/
*complete* instructions - IOW are merely dressed up versions of narrow AI
no one AFAIK is dealing with the issue of how do you produce a true
"goalseeking" agent who *can* discover things for itself? - an agent, that
like humans and animals, can *find* its way to its goals generally, as well as
to learning new activities, on its own initiative - rather than by following
instructions. (The full instruction method only works in artificial,
controlled environments and can't possibly work in the real, uncontrollable
world - where future conditions are highly unpredictable, even by the sagest
instructor). [Ben BTW strikes me as merely gesturing at all this].
There really can't be any serious argument about this - humans and animals
clearly learn all their activities with v. limited and largely general rather
than step-by-step instructions.
You may want to argue there is an underlying general program that effectively
specifies every step they must take (good luck) - but with respect to all their
specialist.particular activities, - think having a conversation, sex, writing a
post, an essay, fantasying, shopping, browsing the net, reading a newspaper -
etc etc. - you got and get v. little step-by-step instruction about these and
all your other activities
So AGI's require a fundamentally and massively different paradigm of
instruction to the program, comprehensive, step-by-step paradigm of narrow AI.
[The rock wall/toybox tests BTW are AGI activities, where it is *impossible* to
give full instructions, or produce a formula, whatever you may want to do].
From: rob levy
Sent: Wednesday, July 21, 2010 3:56 PM
To: agi
Subject: Re: [agi] Of definitions and tests of AGI
A "child" AGI should be expected to need help learning how to solve many
problems, and even be told what the steps are. But at some point it needs to
have developed general problem-solving skills. But I feel like this is all
stating the obvious.
On Tue, Jul 20, 2010 at 11:32 PM, Matt Mahoney <matmaho...@yahoo.com> wrote:
Mike, I think we all agree that we should not have to tell an AGI the steps
to solving problems. It should learn and figure it out, like the way that
people figure it out.
agi | Archives | Modify Your Subscription
-------------------------------------------
agi
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
https://www.listbox.com/member/?member_id=8660244&id_secret=8660244-6e7fb59c
Powered by Listbox: http://www.listbox.com
