Fringe of consciousness, importance spreading

[Ben Goertzel]

A somewhat interesting point came up in discussions w/ Misgana re ECAN today...

So far we have been working on two methods of attention allocation:

-- spreading importance from everything in the AttentionalFocus, frequently

-- spreading importance from everything in the Atomspace, but more
slowly (by sampling Atoms from the Atomspace with probability
proportional to STI, and then spreading from each Atom as it's
sampled)

But we have seen a need now for a third method

-- iterating through every X in the AttentionalFocus, and spreading
importance K links out from X (where, say, K=2 or 3)

The practical use-case that made this requirement clear was: rapidly
boosting the importance of question-answers, that contain words which
are in the AF.... The question-answers are ImplicationLinks, which
point to ListLinks, which point to words ... so to get STI to the
ImplicationLink, a word in the AF needs to send STI to the ListLink
which then needs to send STI to the ImplicationLink...

But more generally one can view this as an importance-spreading agent
aimed at spreading activity from the AttentionalFocus to the "Fringe"
of consciousness -- those things that one can feel at the back of
one's mind, on the edge of one's consciousness ... one is aware they
are there but not quite aware of what they are, but one can generally
with effort pull them into one's awareness...

Sometimes spreading enough STI into the Fringe will cause something in
the Fringe to get boosted into the AttentionalFocus ....

-- Ben

--
Ben Goertzel, PhD
http://goertzel.org

Super-benevolent super-intelligence is the thought the Global Brain is
currently struggling to form...

[Roman Treutlein]

Sounds like a great idea. But how exactly do you spread K links out? Do you just spread it from your Source to it's Targets and then use these Targets as a new source?

Or do you have something more complicated in mind where you spread the STI directly from the Source to an atom K links away?

[Matthew Ikle]

Interesting.

What I’m hearing is that while the original two methods likely achieve the requisite attention spreading they don’t do this fast enough.

Rather than introduce a new third method, it seems to me that we could simply replace the spreading of importance from everything in the AF with the spreading of importance K links out, where the value of K might be determined from context. The “third method” really just seems to me to be a generalized version of AF spreading rather than a new method. As Roman points out, the actual mechanics of how to spread K links out also needs to be determined.

—matt

> --
> You received this message because you are subscribed to the Google Groups "opencog" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to opencog+u...@googlegroups.com.
> To post to this group, send email to ope...@googlegroups.com.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CACYTDBdGNipWfnguuZY%2ByyGtnCxAKoXfqoTiYMSsGcqj0rQ5Mw%40mail.gmail.com.
> For more options, visit https://groups.google.com/d/optout.

[Ben Goertzel]

On Tue, Sep 13, 2016 at 12:49 AM, Roman Treutlein <lordm...@gmail.com> wrote:
> Do you just spread it from your Source to it's Targets and then use these
> Targets as a new source?

yah, that was my thinking... not too fancy but I don't see what else to do...

[Ben Goertzel]

> Rather than introduce a new third method, it seems to me that we could simply replace the spreading of importance from everything in the AF with the spreading of importance K links out, where the value of K might be determined from context.

I just worry that would result in the internal AF STI-spreading
happening too slowly, thus screwing up attractor-formation dynamics
inside the AF ...

After all, spreading 3 links out from X may take 100x or 1000x or 10Kx
more effort than spreading only from X, right?

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/1F463D95-D0D5-4B95-9D4D-092C0EEFDDFC%40gmail.com.

> For more options, visit https://groups.google.com/d/optout.

[Matthew Ikle]

Ahh — I was thinking that the default setting for K would be one but that this would change for specific contexts.

To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CACYTDBepQVSPMh__wu%2BR2yUjQmCT6ofUr0enxcJttGM0Xbamzg%40mail.gmail.com.

[Roman Treutlein]

Using this for all AF spreading with a default K of 1 and using a higher K with a certain probability. Seems like a cleaner solution then have 2 different spreading mechanisms for the AF.

[Ben Goertzel]

I'm still thinking on this...

I understand that it would be cleaner software design wise not to add
a third agent -- but I am worrying about the dynamics... it seems that
to get the dynamics right inside the AF we want STI to be coursing
around really fast within the AF, and that this will get screwed up if
the AF importance-spreading loop has to wait for spreading from the
fringe which may be a couple orders of magnitude slower than spreading
just from the AF ...

> https://groups.google.com/d/msgid/opencog/f7c5efcb-48b3-4e22-8553-a11e9f7e8cd7%40googlegroups.com.