Todor,
This is a cool way of answering problems.
Basically: "there is only one way of solving this problem {putting bricks/toys
on top of each other to build a tower] It's obvious. It's so simple. I, Todor,
have solved it. If only you'd read my work, Chapter X, you'd realise this..."
Unfortunately, there are more or less infinite ways of laying and positioning
one brick on top of another (and positioning your body, head, eyes, and arms
while doing so) - and indeed knocking the tower over - esp. if you are an
infant with poor motor control - as you can see if you check out the reality:
http://www.youtube.com/watch?v=dxfPuEK4t08
http://www.youtube.com/watch?v=2cCmzHMAlZU
http://www.youtube.com/watch?v=dgmDivV0UWI
http://www.youtube.com/watch?v=yMC58x6Y7xk
http://www.youtube.com/watch?v=9C9YjQKwfaw
If you think the infants are iterating a pattern of bricklaying, please explain
what it is - incl. the precise physical positions and "coordinates" of the
bricks on top of each other - *and* the starting points on the floor/elsewhere
from wh. they are picked up. .
Actually, the infants are doing what you're doing below - improvising/making it
up - building a crazy argument [as in "crazy walk"} - as you go along - taking
it one step/brick/wild assertion at a time, higgledy-piggledy - and not
iterating any consistent pattern of behaviour.
And that is indeed the foundation of an **intelligent** AGI.
This OTOH is narrow AI brick-building- and iteration - and while v. useful
also v. unintelligent :
http://www.youtube.com/watch?v=XjHJY7_BzOI
http://spectrum.ieee.org/automaton/robotics/diy/video-watch-flying-robots-build-a-6-meter-tower
From: Todor Arnaudov
Sent: Thursday, July 12, 2012 10:37 PM
To: AGI
Subject: Re: [agi] Real World/ Creative Reasoning - what no one gets about AGI
Mike, thanks for the link to the video (nice selection), but I don't think you
get anything from my thousands of sentence explanations and definitions...
>This is a baby *creating* a course of action -
>http://www.youtube.com/watch?v=yMC58x6Y7xk
>*creating* the building of a tower of bricks,
>not *following* a preplanned course of action. Watch how he lays one brick on
>top of another. Consider the final structure. That's exploration not iteration.
Well, this is exploration for a short time, then it is iteration. :) I you were
capable to understand my sensori-motor explanations, this perfectly matches my
previous explanations.
It's a mere adjusting of coordinates, and the bricks are layed on the top of
one-another, because of the following very simple laws (which is easily
derivable from scratch - it's just the simplest, i.e. the one that comes first
to mind, there aren't a lot of other options):
- In the initial state the bricks just lay on the ground one one row?
- The baby may catch them and move them, that's not "smart" enough.
- The only possible "intelligent" action over two bricks is to put the second
over the first one. You may put it on the sides too - yeah, but that's less
"interesting", see below to see why. (One other reason is, because it's done
first and may be done many times. Then it gets boring - too easy and
predictable.)
- You can hold only one brick in a hand, and brain/highest cognitive
capabilities have a point of integration, the attention/eyes follow/fixate on
just one brick in one hand.
- Obviously the interesting parameters is the height, because it can be changed
when the bricks are moved in a very specific manner (unlike the random motion
that produces other results), which can be specified formally as:
- The spatial coordinates of the bottom of the second brick should match the
coordinates of the top of the first brick.
The operation, "halting criteria" are as simple as matching (they are default
criteria), the baby tries to match some parameters, and literary "halting" -
when putting a brick on the another and pushing, the baby feels a force that
stops it moving it further, that's like a "ground". When there's a force
against the motion, a very straight "decision" is to stop pushing, it's a basic
reaction, and it's learned easily - attempts to push a brick to the basic
ground. Which in low level is:
-- Motor commands move in a direction against the plane of the ground.
-- Proprioceptive feedback that the hand doesn't move further, while in all
other situation, when there's not an obstacle, when there's motor command to
move in a given direction, the hand moves. That's how "obstacles" are
generalized - you push, but your hand and the object in it doesn't move further.
This is how the top of the brick is recognized and generalized as a "higher
ground", another plane.
Using generalization, the baby attempts to perform the same operations that on
the basic ground.
As mentioned - MATCHING and feeling the RESISTANCE are the simplest and most
elementary criteria, which perfectly match the situation and the basic
cognitive capabilities - the baby moves another brick to coordinates given by
another brick.
"Why exactly those bricks, and not others?"
-- Because those were closer to the baby, because she saw those in the moment
when she decided to catch a brick (and then her attention is "locked" for some
time and the other bricks are ignored until that one is felt in the hand) etc.
...
And if you did understand my other explanations on
vision/geometry/3D-transformations, you wouldn't be that impressed that the
baby notices that some parameters change when the hand is aligned with the
objects and its coordinates are adjusted.
Another thing that is very obvious, regarding how the baby recognizes that it
should lay the bricks one on the another, is that no matter how she moves the
bricks on the plane, the height of the topmost point of them stays the same.
The only way to change this parameter further - the height - is to put another
brick. Initially all the bricks are on the same plane, they have the same
height.
If you put the brick on another one which is on the ground, the total height
would just match the first one (that's also interesting in the beginning,
because now these heights match each-other height parameter).
>The basic principles are the same that adults use to create searches on
>amazon, ebay, or anywhere on the net - and that underlie every real world
>activity, period.None of
>them are algorithmic.
Yes, the principles: maximize MATCH, optimize paramaers, get "bored" by too
predictable input etc. are the same, but the levels of generality of a search
in Amazon and this are so huge. And you're wrong in your classification - a
search in Amazon is not "real world", this is an abstract natural language and
social world, crystallized in stereotypes and social scenarios which were done
thousands of times.
"Real world" is a world of low generality such as of this baby, and the
"creativity" in that domains is obvious and elementary - sorry if you can't see
it.
In the Amazon etc. examples it's obvious either, but the "instruments" and
operations are in higher level coordinate systems, with other types of
"coordinate adjustmenets" and other types of searches for maximizing the match
between parameters.
> Which part of your AGI course covers creativity/ real world reasoning? Which
> other AGI course covers creativity/ real world reasoning? None.
Actually in most parts... :-D Since the focus is on sensori-motor generalizing
hierarchies, as I explained 1000 times so far, every action of mind is
eventually reducible to coordinate adjustments/transformations (a baby stacking
bricks, or a student typing on a keyboard a query in google, or doing
whatever). All that is "real world reasoning", because it's derived from
sensory inputs and the output is ultimately in "motor outputs" - coordinate
transformations in the lowest-level Universe space.
Read my works and see any talk of Schmidhuber about creativity and
interestingness.
-- Todor
http://research.twenkid.com
http://artificial-mind.blogspot.com
On Tue, Jul 10, 2012 at 5:52 PM, Todor Arnaudov <[email protected]> wrote:
Mike,
It's you who doesn't get elementary things.
Talking about high level crystallized activities that 20-30 year-old humans
do (and they are rather logical than "fluid"), do you have any idea how many
transformations happened in the brain for that time.
You should give examples of babies or little children playing, not natural
language related activities of 30-year old researchers.
What do you know about creativity? Show us a short story, a creative drawing,
a film, some dance move, musical piece (not that all those are quite
"mechanical" like everything).
>Correct me -
>
>you're doing what everyone else does here, and thinking of : "how can I find
a *standard* way of solving this problem?" -
>
>so first you try to solve the particular problem, and second you try to turn
it into a standard way of solving problems like this.
>What you've missed is : this is a **creative** problem, a **one-off** - as
are all real world problems,..
>By definition, **you didn't/don't have a pre-existing method of solving it**
- and no standard method is possible..
>When you go to search for "good books on AI", you know something about
searching/research and also about AI, but you don't have a standard search
method or frame.
3294039430390 people around the world reach to the same "non-standard"
methods.
And in what year do you live, Mike, currently people have a decade or more of
using search-engines, they don't invent it.
Searching in a library or an encyclopedia is not that much different. The
first time it's "invented", then it's applied.
Limited working-memory and the pruning of search is also how advertizing and
economy works, why the options on the top are selected and the rest is usually
ignored.
People don't check everything, but what's near-by, what fills up their
working memory and is strong enough.
Then they stop searching and buy the washing powder brand which is repeated
the most on the TV, or click what's on top.
As of "the best neuroscience researcher", it's the same search of who's "on
top" on some list, for whom the superlatives are said, the most publications in
"prestigious" sources, etc.
All those are very high level concepts, but I know you don't make a
difference between abstract and specific, you use to talk about specifics and
call it general - the set of all different chairs.
>You have to **create** a method on the fly - in effect, create and modify a
"quasi-program"/plan-of-action as you go along. That is actually what you did -
your first (as distinct from your second) course of action.
The first time you do.
>Whatever search methods you used *won't* be directly applicable to further
searches. In the real world, every problem situation is **different, ** even if
also similar to others. ****Real world reasoning IS creative reasoning*****
You don't get, that it is different depending on the POV and the resolution
and level of generality. We're talking about language related problems which
are of high generality.
>Your next research task may be *who are the leading authorities on
neuroscience?" or: "what are the main branches of neuroscience?" For each of
these, you will have to create new
>methods of search. What you learned re AI may help, but only somewhat.
That's the nature of living in the real world .
What new methods Mike, the guys who do those searches use to be 20-30 years
old, they have read enough of textbooks, if not something else.
That information is available at a glance in such literature, even literary
written as "branches" etc.
>No one in AGI understands that the function of an AGI - real world agent -
is to deal with creative problems. Not standard rational problems for which
methods of solution exist. But new >problems for which no method exists.
What do you understand Mike, sometimes I think I'm not discussing with a homo
sapiens sapiens, you don't get a thing of what I'm explaining to you.
>AGI isn't about running pre-existing-plans-of-action to solve rational
problems, it's about forging them to solve creative problems. About a machine
not running , but creating a search.
Eventually everything is "pre-existing", the molecules of your body are
preexisting anything you do consciously, and all of the actions mind can do
with the body are "preexisting" it just selects from them.
Sorry you lack capacity to get it, it seems not all homo sapiens can.
--
-- Todor "Tosh" Arnaudov
http://research.twenkid.com
http://artificial-mind.blogspot.com
--
-- Todor "Tosh" Arnaudov
http://research.twenkid.com
http://artificial-mind.blogspot.com
AGI | Archives | Modify Your Subscription
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com
