On Tue, Mar 25, 2008 at 4:42 PM, Vladimir Nesov <[EMAIL PROTECTED]> wrote:



> Simple systems can be computationally universal, so it's not an issue
> in itself. On the other hand, no learning algorithm is universal,
> there are always distributions that given algorithms will learn
> miserably. The problem is to find a learning algorithm/representation
> that has the right kind of bias to implement human-like performance.
>
> It's more or less clear that such representation needs to have
> higher-level concepts that refine interactions between lower-level
> concepts and are learned incrementally, built on existing concepts.
> Association-like processes can port existing high-level circuits to
> novel tasks for which they were not originally learned, which allows
> some measure of general knowledge.
>
> As I see it, the issue you are trying to solve is the porting of
> structured high-level competencies. Which looks equivalent to the
> general problem of association-building between structured
> representations. Is it roughly a correct characterization of what you
>  are talking about?
>
> Vladimir Nesov
> [EMAIL PROTECTED]
>

Can you give some more indication about what you mean by "porting of
structured high-level competencies" and the "problem of association-building
between structured representations"?

I do not know where you got the phrase "porting" from since I have only seen
it in reference to porting code from one machine to another.  I assume that
you are using it as a kind of metaphor, or the application of an idea very
similar to 'porting' to AGI.

Let's suppose that I claim that Ed bumped into me.  Right away we can see
that the word-concept "bumped" has some effect on any ideas you might have
about Ed, me and Ed and me.  My claim here is that the effect of the
interaction of ideas goes beyond semantics into the realm of ideas proper.
If it turned out that I got into Ed's way (perhaps intentionally) then one
might wonder if the claim that Ed bumped into me was a correct or adequate
description of what happened.  On the other hand, such detail might not be
interesting or necessary in some other conversation, so the effect of the
idea of 'bumping' and the idea of 'getting in the way of' may or may not be
of interest in all conversatations about the event.  Furthermore, the idea
of 'getting in the way of' may not be relevant to some examinations of what
happened, as in the case where a judge might want to focus on whether or not
the bumping actually took place.  From this kind of focus, the question of
whether or not I got in Ed's way might then become evidence of whether or
not the bump actually took place, but it would not otherwise be relevant to
the judge's examination of the incident.

Presentations like the one that I just made have been made often before.
What I am saying is that the effect of the application of different ideas
may be more clearly deliniated in stories like this, and that process can be
seen as a generalization of form that may be used with representations to
help show what kind of structure would be needed to create and maintain such
complexes of potential relations between ideas.

While I do not know the details of how I might go about to create a program
to build structure like that, the view that it is only a 'porting of
structure' implies that the method might be applied in some simple manner.
While it can be applied in a simple manner to a simple model, my interest in
the idea is that I could also take the idea further in more complicated
models.

The point that the method can be used in a simplistic, constrained model is
significant because the potential problem is so complex that constrained
models may be used to study details that would be impossible in more dynamic
learning models.

Jim Bromer

-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription: 
http://www.listbox.com/member/?member_id=8660244&id_secret=98558129-0bdb63
Powered by Listbox: http://www.listbox.com

Reply via email to