Jarrad,
There are no probabilities here - they only exist in the fantasies of AGI
textbooks & Ben's mind - & artificial narrow AI exercises;
Where on earth are you going to find a) probabilities and b) a list of the
possible substances that might "probably" be on the pic. ?
Raindrops, ink blots, semen, urine, plasticine, cells, specks of dust, earth
particles. Raindrops 0.7 probability, semen, 0.05 probability, specks of
dust 0.22 probability. Oh puh-lease
The list of substances is infinite, and probabilities impossible to
identify/compute.
Ditto with every real world scene.
Logic & maths (& any kind of precision) simply do not apply - because there
are no a) sets and b) uniform shapes/elements.
Quick *shape*-matching and guesswork (rather than systematic consideration
of logical options) are how I suggest the brain works. AGI is rough and
dirty, not clean and precise like narrow AI.
Something more along the lines of .." that particular shape matches loosely
with other liquid...water... rain..drops I have seen..." "this whole set
of shapes/drops matches loosely with other sets/configurations of raindrops
I have seen - and is "pretty likely" (rather than 0.72) to be such"... And
the brain won't like a dumb logical program go through every shape and set
of dots there but be satisfied with a general impression, (much as the
brain doesn't go through every chess move but works on quick consideration
of a few visual configurations of chess pieces).
That is a horrific approach from an artificial world, logical AI POV - but
it's real world smart, and logic simply doesn't apply in the real world,
(not without an AGI human mind to apply it).
--------------------------------------------------
From: "Jarrad Hope" <[email protected]>
Sent: Friday, June 08, 2012 7:46 AM
To: "AGI" <[email protected]>
Subject: Re: [agi] The Visual Alphabet
Hi,
I'm very new to this list, AGI and AI in general, and probably have
less of an understanding than Mike.
From what I've read - for the water example, say we took that photo
and gave it to an AGI. Wouldn't PLN and some visual processing be able
to infer (based on past experience) that all that blue is probably
sky, theres multiple shapes that appear to be on a plane (inferring
the plane is transparent, probably glass) - would bring up a bunch of
nodes of things that could be related/attached to glass and sky one of
them being water - from other experiences/knowledge they're would be
networks that would branch out to water being particles/h20/fluid
dynamics etc etc (should probably cut off the algorithm before this
point).. and therefore all those distinctly different shapes are
probably some kind of fluid, probably water (unless in the context
there is a higher probability of it being, i dunno, urine?)
AFAIK the shape recognition itself - as in all those shapes belonging
to water - is almost irrelevant - because theres more conceptual data
that can be inferred from the image, from previous learning. Building
up a huge dataset of various shapes of everything seems like it will
lead to an infinite dataset which would be impractical to traverse and
will for sure produce collisions.
The above also applies to your dancing example, PLN looks like it's
capable of abstracting away a dance move from a human body, - if it
saw a human body dancing it would infer all humans can mimic that
dance move, including the ones that dont have 2 arms and 2 legs - the
AGI could relate this atom-set to all mammals - all animals - if it
traverses further up its 'human concept' tree.
Would be interesting to see how it could prove an inanimate object is
capable of that dance move - without having seen it before.
So I can personally see how AGI is able to grasp what's in an image,
at least with a 'highish' probability. As long as it has enough
previous experience/knowledge.
I feel like Mike is coming from a purely Visual Processing stance -
and in that sense I agree with him, no amount of vanilla image
processing is going to work.
Please correct me if I'm wrong - I'm learning as much as possible for
the AGI Summer School in Iceland
Regards,
Jarrad
On Fri, Jun 8, 2012 at 5:37 AM, Mike Tintner <[email protected]>
wrote:
P.S. To explain further: I said that the brain(-and-body) understands
changes of form by robotic/embodied principles. So how is it that you
are
able to understand a body in any new dance shape as a human body? Note
that
all kinds of weird new dance moves we've never seen before are
continually
being invented.
You understand those shapes by literally simulating them with your body -
asking: could my body take that new shape? Your body does not need to go
all the way to simulate new body movements - it can prefigure them, for
it
is all the time prefiguring and checking prospective movements. (In this
way
your brain-body system tells you that your injured foot cannot make a
certain normally habitual movement, BEFORE you ever make it). If the
human
brain-and-body can do this, robots can and will have to do this, (if only
eventually).
There is a great deal of evidence that this kind of simulation actually
happens.
We understand how other kinds of body, like water or lava, move by also
simulating them with our body, comparing them to how we move.
To the classic AI mind, this sounds at first v. complex and wild, but
actually it's vastly more parsimonious than a blind, fantastically
convoluted geometric approach, which hasn't worked and can't work.
I'm suggesting we have to really "grasp" visual shapes to understand them
and their many transformations.
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed:
https://www.listbox.com/member/archive/rss/303/22581513-9fe46a1c
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/303/6952829-59a2eca5
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
