Ron,
Sounds like you're calling for something not a million miles from what I've
been calling for.
[Obviously I'm a techno-idiot and so have only been very loosely,
philosophically outlining what I mean, but neverthless that can be useful,
because it does point in a new direction - and calls for designing a new
kind of machine - which upsets everyone and turns them abusive because they
don't want to have to think about that, they just want to work with the
machines they've already got, even if those machines don't work. .].
Essentially, you may be saying like me, that creative analogy - which is the
absolute heart of creativity, and what AGI has to achieve - works basically
by what you might call "physical analogy."
How did you see/ or come up with the idea that "B" is like "13." ?
And since those two figures are rather close, let's take some further apart.
How is it that when you look at what is actually a bicycle seat and
handlebars:
http://cn.cl2000.com/history/beida/ysts/image18/jpg/02.jpg
you are able to seat a bull's head, something like:
http://www.chu.cam.ac.uk/images/departments/classics_bulls_head_rhyton.jpg
How is it that you can look at an ink blot
http://www.bbc.co.uk/schools/victorians/images/school/learning/slideshow/ink_blot.jpg
and see a fish?
Or a cloud:
http://lintrups.dk/images/Diverse/hiroshima_mushroom_cloud.jpg
and see a mushroom?
What you're doing is what lies at the heart of arguably all creative analogy
and metaphor. It's what enables you to understand the verbal metaphor
"mushroom cloud", or to understand the words, "the clouds cried" because you
can see that raindrops are like tears, or see that someone is "bull-headed"
from both the way they set their head rigidly like a bull does, and proceed
either in a physical charge, like a bull does; or to see that someone "eats
like a pig" from the way they stick their head into a plate and chomp away,
compared with the way a pig sticks its head into a trough and eats..
What we can say with great confidence, is that this cognitive process does
not, and cannot work by any digital analysis - because that relies on
dissecting things into their PARTS. You can't dissect the *features/parts*
of a cloud and the features/parts of a mushroom and observe their
similarity. Ditto with all the other examples. You can't dissect the parts
of that ink blot and dissect the parts of a fish, and observe a likeness.
[You can't therefore form two verbal networks designating their
features/parts - pace Gentner, Minsky et al - and recognize the similarity
of the objects by comparing those feature networks]
Why? Because THERE ARE NO SIMILARITIES BETWEEN THE PARTS of the objects.
If you look, the only similarities are between the wholes - or forms of the
objects - and the LOOSE FORMS at that - and esp. though not exclusively,
their LOOSE OUTLINES. It's only the loose outline of that h-bomb cloud that
is like the loose outline of a mushroom, and ditto only the loose outline of
that blot that is like the loose outline of a fish, or the loose outline of
a bicycle seat that is like the loose outline of a bull-s head..
Those outlines have to be loose, because if you look too closely at them
again, the similarity vanishes. That cloud has fluffy edges to its outline,
the mushroom has none. The bicycle seat is triangular-ish while the bull-s
head.is oval-ish - quite a difference.
All this will bother the hell out of both a digital, analytic program and
machine trying to compare the parts/features of those objects. But it
doesn't bother your brain at all. You can see the actually rather-distant
similarities with great facility and speed because you're working with the
wholes.
So how is it done mechancally?
"Physical analogy." Your brain, or any comparable machine if it's to be
successful, has to work with the wholes rather than the parts. It has, in
effect, to physically overlay the outlines of those objects. Literally,
physically (and not metaphorically as per current AI terminology) map their
maps onto each other and see if they loosely fit.
And the brain also has to make those maps/outlines FLUID and FLEXIBLE - it
treats them, I suggest, as if they were outlines seen through water - highly
squishy and squishable. And it's quite prepared to SCULPT those outlines and
chop off bits here, and add bits there. It's only looking for a loose,
sometimes extremely loose fit.
Thus you can look at the highly intricate and serrated outline of the map of
Italy and the relatively simple and smooth outline of a boot - which are
actually radically different in many respects - and nevertheless see the
rough similarity. Your brain has squished those shapes considerably to match
each other.
Physical analogy. Free-form- matching.
(We could also use the very common term here for this kind of thinking -
which has a literal, physical truth -this is literally "figurative
thinking" - working with the figures, the outlines of objects).
Remember the brain, neuroscience tells us, is full of flexible maps. They
are fundamental to its workings. When you make any movement, for example -
reach out your hand to grasp something and shape it like a cup - you will
form a flexible, fluid cup shape/outline - altering/shrinking/expanding it,
maybe even losing or adding a finger or thumb, as you get nearer the object
and adjust to its precise outlines. Fluid maps/outlines are essential to
direct the operation of a robotic body in the real world, which to
continually flexible adjust/squish its shape.
But no computer is capable of this yet. [AFAIK]. They can't 1. overlap
forms/outlines directly/. And they can't 2. "SQUISH" those shapes - alter
them fluidly and flexibly.. truly plastically. A computer can obviously
achieve somewhat similar morphing effects by mathematical means - but they
can only proceed formulaically - using formulas based on the *parts*. And
this has to be free form matching of the wholes/*outlines* - as if you were
squishing plasticine.
So we have to start designing a machine- possibly some new version - of the
current machine. that can do these things.
{Is this, Ron, anything like what you mean]
P.S. Note that physical analogy - free-form matching - is with great
probability not only at the heart of creative analogy and metaphor, but also
visual object recognition. The brain has to continually compare objects of
radically different shapes to visually recognize them as being of the same
kind - as being basically say the same form of squashed "ball", or
squished "face", or very diversely squished "amoeba" or, well, endlessly
squished "plasticine" itself..
----- Original Message -----
From: "Ronald C. Blue" <ronb...@u2ai.us>
To: <agi@v2.listbox.com>
Sent: Sunday, January 11, 2009 9:29 AM
Subject: Re: [agi] Identity & abstraction
I would agree that the ABC example is an analogy. Generally speaking I am
quickly successful in explaining how you can model the brain in electronic
to people with backgrounds in analog electronics. The historical efforts
in this direction of associationism and opponent process go all the way
back to Aristole. Interesting observations reveal the opponent process
nature of color. Example stare at the picture of the American flag
http://www.brainviews.com/abFiles/IntOpponent.htm
in a dimly lighted room for 45 seconds then look at at any gray area in
the room and you will see the colors switch. The opponent process for
color are blue-yellow, red-green, and black-white.
People who played with an opponent-process model of leaning reads like a
list of who's who in psychology including Pavlov. They all dropped the
model because it was not simple.
Einstein said make your theories as simple as necessary to explain the
data. Simple doe not mean so the average American can understand it.
Illusion are clues on what the brain is doing. What the brain is doing
can be model in an AGI machine, Even computers can be programed to
experience illusions or violations of the
programmed expectatons. Example:
Marshall, J.A. & Alley, R.K. (1993, October). A Self-Organizing Neural
Network that Learns to Detect and Represent Visual Depth from Occlusion
Events. [In Bowyer K.W. & Hall L. (Eds.)] Proceedings of the AAAI Fall
Symposium on Machine Learning and Computer Vision, Research Triangle, N.C.
p70-74.
Your stated goal is the development of an AGI machine. I am telling you
in my opinion that it can not be done in a programming environment but it
can be done using opponent process circuits. We can not stop a child open
his head and list his programs for our review and simple understanding.
Sadly this is also true for analogy phase state opponent processing
machines. Children are not controlable and neither are analogy phase state
opponent processing machines. The current goal is developing a
programming control system to interface with an analogy phase state
opponent processing machine. After spending $200,000 we have been stuck at
this problem level for 18 years. We had the AGI but no interface to
traditional computations. At this time the current progress is promising
that the two procedures can be made to cooperate with each other.
You now have enough information to start your thinking.
Ron
http://u2ai.us
-------------------------------------------
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/?&;
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/
Modify Your Subscription:
https://www.listbox.com/member/?member_id=8660244&id_secret=126863270-d7b0b0
Powered by Listbox: http://www.listbox.com
