http://www.wired.com/wiredscience/2009/04/newtonai/
http://www.wired.com/wiredscience/2009/04/robotscientist/
http://www.wired.com/wiredscience/2009/01/retailrobots/

Now if that ever changes, and they can demonstrate moral judgment...
then perhaps we will see a new form of government run by machines
(Maybe they'll even replace us eventually-or more probably a synthesis
might occur)

Computers don't yet have intelligence (albeit programmers like to use
the term "AI") but they are still  far more than "glorified
calculators" (I really regret saying that)  While a computer does
calculate (using much the same transistor based logic gate technology)
there are significant conceptual differences over a basic calculator
that make a world of difference.  Although the component technology
between calculators and PCs is similar- conceptually a calculator's
functionality is closer to Herman Hollerith's 19th century punch card
machine than modern computers

Putting aside some newer calculators are like small computers.... the
big difference between a computer and basic calculator is that a
computer can have its programming easily adjusted (i.e. software)
whereas a basic calculator usually is limited a small range of
hardwired tasks

Theoretically one could create something akin to a calculator that
does everything a computer does but creating a distinct piece of
hardware for every function would be very expensive (and thus
impractical). What makes the computer fundamentally different is its
versatility. Rather than relying on dedicated hardware (or later
firmware)... software makes the task of adding functionality possible
(which is light years ahead of taking out a soldering iron every time
we want to do something new)

--
Les Cargill

There is of course constant feedback via reason also. I might desire
that money should grow on trees (money being essentially the modern
expression of food, shelter, etc...) However, when I observe the world
I see it doesn't. Thus I become disappointed... a feeling... which
adjusts my purpose which makes me use reason to find the necessities
of life in some other way.  Or perhaps I am given an education so I
don't need to check if money grows on trees  and can work for it.
Again though...why do I accept that education in the first place?
Because I was spanked or rewarded as a child not because I was born
fully mature and able to rationalize why I needed an education.

Unfortunately we lack a detailed scientific description for feelings.
Which means we lack the accompanying algorithms to describe a key
aspect of human intelligence. Which means we can't write code for it.
We can create programs that simulate behaviors (e.g. we can imitate
the behavior of some types of insects or get a program to play chess)
but the big goal line is human-like AI. The problem with current AI is
that it has some reason (of the analytic variety) but it has no
motives. It can't have motives because it has no feelings. Our
approach to AI has been to provide individual motives (a task or
behavior). Unfortunately since there so many possible permutations of
tasks human intelligence must deal with trying to write them all
becomes as unwieldy as trying to write computer code with a soldering
iron.

The reason why an operating system is not intelligent is not because
it can't reason analytically but because we have to give it explicit
instructions to do something (via software) If we didn't it would just
sit them humming along doing essentially nothing for eternity. We
could (with time) create a computer that appeared intelligent by
gradually adding a massive number of programmed scenarios that gave it
appearance of intelligence (our approach at the moment)  but this imo
would not be the essence of intelligence. (although that's still a
philosophical statement because for now we can't say for sure that's
not how humans work).

There is a big if here. Can feelings be described in an analytic
fashion also? Assuming the answer is yes (and I am correct about
missing feelings being the problem)..... I have a hunch that  the
solution to AI is somewhere to be found in Mandelbrot's approach (of
fractal fame).

Feelings are these complex behaviors that sometimes appear almost
random. Mandelbrot  saw large seemingly random shapes as the result of
very small much simpler patterns that through repetition formed into
those more complex shapes. Besides accurately describing things like
the growth of trees and forests... this approach has turned out to
have very practical applications in computer programming  (e.g. the
inspiration behind modern anti-aliasing received its inspiration from
Mandelbrot as well as the polygon approach to generating 3D graphics
engines) .

The specific reason why I think a smaller description to AI will be
the "solution" (rather than just trying to program every permutation
of behavior) is because of the way human life starts off, We were all
a tiny fertilized egg at some point. Our DNA is even smaller. Putting
aside theories that the universe started off as a tiny singularity...
because of our DNA's tiny size it appears likely there are physics
limitations as to how much information our DNA is  capable of
containing. Yet despite that we still ended up with our immensely
complicated bodily functions and intelligence.

So my bet is the code to write general purpose AI intelligence is
rather small. We just don't know what that algo is yet  (or possibly
algos).

--
Les Cargill

Joshua Greene is on the cutting edge of essentially a new field that
is trying to bridge science and philosophy. (neurophilosophy). What
makes Mr. Greene's research unique is that he decided to examine how
moral reasoning works on a biological level. The ability to image a
living brain is a relatively new technology. Greene uses MRI while
posing all sorts of different types of moral questions to subjects.
His research while not some panacea of ultimate philosophy does gives
some concrete scientific evidence that emotions can play a role in
moral judgment.  (which really is self-evident given people tend to
focus on behaviors that bring them pleasure not pain)

http://tinyurl.com/y8bbzw3

If I had to guess I think the trick to better understanding feelings
will be getting down to the nitty gritty resolution of something like
a neuron and how it interacts with other neurons  And just as our
brain can be broken down to a collection of smaller similar neurons...
neurons can themselves can be broken down into smaller repetitive
structures. This is just one more reason why I suggest Mandelbrot's
approach to complex shapes seems most likely to create a believable
AI. It would mimic the way our brains appear to actually work.

We'll probably see androids at some point in this century. We are
already seeing technology like Sony's AIBO (which allowed hobbyists to
import and write their own code to create behaviors) It's only a
matter of time before that gets married to technology like Honda's
ASIMO and a ubiquitous operating system with an API is created so 3rd
parties can add software.

However, IMO (as a long time programmer) trying to code every type of
behavior is a tedious approach and not representative of real
intelligence. There will still be uncanny valleys where we'll easily
be able to spot the difference. And even if we eventually can't, Alan
Turing's test for an artificial intelligence is not a good measure of
intelligence alone. The test's focus is on whether a computer program
can fool someone for some arbitrary period of time.... not whether the
program is reasoning the way humans do. It's sort of like the
difference between a diamond and cubic zirconia. Or comparing a car to
a horse and wagon, They may appear the same on the surface or the
behavior might be similar in some respect, but on closer inspection
their are huge internal differences.

This isn't to say that the behavioral/task-oriented approach to
programming doesn't have its own rewards (which is really what the
modern computer represents) but I don't see it that intelligence as
analogous to human intelligence yet despite that the term "artificial
intelligence" is often used. I would assert to be intelligent like
humans by definition requires the intelligence have something that
mimics feelings (since that's one of the key properties of every human
brain)

Of course even we create a feeling computer the argument can be made
it is still only a golem. However, if it logically uses the exact same
principles as a human brain the question becomes much murkier.   What
would make make silicon based intelligence less real than a carbon
based intelligence? Is the essence of "intelligence" the specific
material used or the mathematics behind it? Given we share much the
same genetic materials as a frog or potato... I would say the logic
rather than materials is the defining property of intelligence.
(Although this doesn't not necessarily mean that the intelligence is
"alive" which is a distinct issue)

I realize he's using this to study what regions of the brain
do what - to 'reverse engineer" the brain, but the structure here
is sub-rational.

Ultimately, I'm pretty well convinced that Mandelbrot is
simply about data reduction, simplifying representation. It
doesn't solve the gross problem at all - the gross problem
being that adding all these maps up doesn't yield
a coherent whole. The coherence is - surprise - because
of the sheer number of iterations of evolutionary power.

We're going "nuanced", but Nature is brute force...

*yet.

--
Les Cargill

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

 > This is just one more reason why I suggest Mandelbrot's
 > approach to complex shapes seems most likely to create a believable
 > AI. It would mimic the way our brains appear to actually work.

??

-RKN

"Feelings are these complex behaviors that sometimes appear almost
random.  Mandelbrot  saw large seemingly random shapes as the result
of very small much simpler patterns that through repetition formed
into those more complex shapes. Besides accurately describing things
like the growth of trees and forests... this approach has turned out
to have very practical applications in computer programming  (e.g.
modern anti-aliasing received its inspiration from Mandelbrot as well
as the polygon approach for generating 3D graphics)"
...

What I'm saying is that just because feelings don't appear logical
doesn't necessarily mean there isn't a precise formula behind them. I
suspect the seeming randomness of feelings is the result of repetition
of very simple analytic principles that end up forming much more
complex behavior.  The reason why we aren't all clones of each other
is because from instance-to-instance we are subject to different
experiences from the environment around us (which can adjust our
behavior and apparently can even shape us on a microscopic biological
level ala phenotypes)

Going back to my polygon example.. most programmers don't care how we
generate 3D graphics (unless creating a 3D engine or some lame flash
movie) . It's essentially a black box and we just use some API or
software to draw the shape and let OpenGL or the DirectX worry about
it. In practice, what's under the hood when you see rich 3d games
graphics today (e.g. xbox) is almost always polygon based.   The
reason why we use polygons (and related polygon mesh) is because  the
(most popular) alternatives of Nurbs and ray tracing while excellent
for prerender film and photo realistic stills are usually
substantially slower to process. (and there is also something called
sub-division surfaces which is sort of a cross between polygons and
the beziers employed by Nurbs but its not my area of expertise).  A 3D
shape generated by polygons is really just a collection of tiny
repeating polygons that create the illusion of being a complex shape

Likewise... there (now) are AI researchers out there that are taking
the same programming approach I'm described.. but towards
intelligence.  Instead of trying to build some vast intelligence via
one giant complex program.... they are trying to build it from many
many tiny simple programs that interact with one another. Rather than
telling them specific complex behaviors, they observe complex
behaviors that naturally arise via their interactions (its really of
like a zoologist or biologist might)  I've heard the term "neural net"
applied to describe this approach to AI sometimes. I don't know how
much research goes into specifically "feelings" (which is orders of
magnitude more complex) but it seems reasonable if someone got the
algos right feelings (or something that appears to mimic feelings)
would be a natural consequence.

Of course this is all hypothetical and just my own feel for the best
approach towards AI research. No one knows for sure yet.

Since 'morality' is a guide to the values a living creature requires, one
must ask how something that is not living, can possibly relate to those
values. Our desire to live is simply the result of millions of years of
weeding out all those that didn't care to live. Emotions such as fear, or
aggression or empathy loyalty and love, are the result of survival of living
beings, and the death of the remainder.

Once volition is involved, it is possible to rationally examine the
appropriateness of our emotions. Getting back to machines and emotions,
where is the link between their artificial feelings, and the requirements of
survival of a human? In what rational way, could such a machine have what we
call a moral code?
--
Arnold

However, my chief motivating factor is still emotion on some level. I
don't want the unpleasantness of a shorter death. (or diabetes, or
being fat). Even cases of "fasting" are emotions at work. While it
might seem someone is denying themselves pleasure... they are doing it
because they see the promise of a greater pleasure later on ("heaven")

Mal and I were briefly discussing this on another thread. I brought up
the point that artificial intelligence might potentially represent a
future form of government for man. Suppose machines became capable of
running an economy far more efficiently than men (which I believe
would require moral judgment.... which I believe in turn requires
something approximating feelings). Suppose these machines could
produce goods, technology and services that we ourselves could not
manage by ourselves. Would we chose to still run our economies or
would we hand them over to benevolent dictator machines?

It took humans hundreds of millions of years of evolution to get as
advanced as we are. If we want to speed up the process with silicon
based intelligence copying what's already there is probably the best
method. One of the complaint about neural nets at the moment (as
applied to AI) is that there isn't enough information of how the brain
works yet to accurately simulate it (which makes the research rather
random and somewhat fruitless beyond insect-like behavior)

--
Arnold

 From here you segued into Mandelbrot sets, which
are self-similar on magnification, but neurons
aren't.

Or perhaps you merely meant that many *copies* of
certain biomolecules that constitute certain structural
components of a neuron are necessary to make those
components, which *is* true, and if that's what you
meant I'll move along. But I'd still consider your
verbiage quirky.

-RKN

Where I think some semantic confusion lays between us is I see
something can be repetitive and similar without being perfectly
identical. (although something can of course be repetitive and
identical as well). I don't typically like using the word identical
for the specific reason anything can be argued different if someone
chooses to ignore the essence of a particular  argument.  For instance
even an atom of the same element has particles in different positions
or speed at any given moment from another atom of the same element
(which would invalidate Mandelbrot sets in nature if one took it to
the absolute definition of perfectly identical as opposed to my
similar)

http://en.wikipedia.org/wiki/Fractal#In_nature

For example, a dendrite "tree" is not dissimilar to the growth pattern
of trees (thus the words "dendrite"and "tree" to describe them). Both
human tissue and trees have been compared to Mandelbrots sets (by
experts in each respective field)

"In the present study we apply fractal analysis to this unsolved
problem and calculate the fractal dimension for each dendritic arbour
of a neuron. We will hereby prove that by application of fractal
analysis to the dendritic arbours of these cells whilst ignoring other
neuronal attributes allows for clear discrimination of only three cell
types."

http://www.sciencedirect.com/science?_ob=ArticleURL&_udi¶T0G-4PD4XHV-2&_user &_rdoc=1&_fmt=&_orig=search&_sort=d&_docanchor=&view=c&_searchStrId 85000721&_rerunOrigin=google&_acctÀ00050221&_version=1&_urlVersion=0&_useri d &md5 57abc43e3fd11d222c674e9edf5fd6

http://books.google.ca/books?id=t9l9GdAt95gC&pg=PA3&lpg=PA3&dq=mandel... BMQ6AEwBA#v=onepage&q=mandelbrot%20trees%20living&f=false

http://www.google.ca/#hl=en&ei±X4SpL2AoWj8Aaf_4X0CQ&sa=X&oi=spell&resnum=0&ct=result&cd=1&ved AYQBSgA&q=mandelbrot+set+dendrite+neuron&spell=1&fp=f616e92003ac2179

http://www.google.ca/search?hl=en&sig=yFIZXYF1ETpKTt0AxOcMMHpvjA8&q=m...
http://www.google.ca/#hl=en&ei=_mX4SvCYCZHTlAexrf3xCg&sa=X&oi=spell&r... AYQBSgA&q=mandelbrot+sets+polygons+3d+graphics&spell=1&fp=f616e92003ac2179

I continue to stand by all my points. but if you feel it could have
been worded better that's a fair enough criticism.

Pretty quirky verbiage if you ask me.

You know what they say, "A concept which describes
everything distinguishes nothing."

-RKN

Again though, that's far from the only thing I'm saying here. Patterns
can also be stacked on top of other patterns. And running with the
case of specifically fractals, while natural fractals can't really be
described as "perfectly identical"....self "similar" fractals  do
exist in the natural world.  This is exactly why we can create forests
that look organic and realistic using software based on simple rules.
(landscape is another great example) The list of natural shapes that
follow fractal approximations is quite extensive.

http://library.thinkquest.org/26242/full/ap/ap.html

Fractal analysis is a quite common approach for trying to determining
the shape of tiny biological structures.  It's "pretty quirky
verbiage" to suggest fractals have nothing in common with neurons when
plenty of examples exist that contradict your speculative
assumption :) A Google search of the terms +"fractal analysis"
+neurons gave me more than 19,000 hits (the pluses meaning both terms
are present).

http://www.google.ca/#hl=en&q=%2Bneuron+%2B%22fractal+analysis%22&met...

A search of +fractals +neurons returned 1.9 million hits.

http://www.google.ca/#hl=en&source=hp&q=%2Bfractals+%2Bneurons&btnG=G...

.....

The general theme of my thesis is that the complexity of intelligence
is really a collection of smaller simpler rules stacked on top of one
another that both repeat (on enormous scales) and interact with one
another (as opposed to some giant super algorithm thats supposed to
account for every scenario humans encounter).  This programming
approach appears logical to me since that's how our body seems to
work. We are collections of different kinds of repeating cells that
form more complex structures (e.g. a skin, a brain, a skeleton, a
lung, a heart, and finally a human).

(From my perspective)  the reason why AI's currently don't
"understand" is because those smaller simple rules aren't being
followed (not that I'm saying I know what those rules are).  Trying to
make a computer understand at the moment is  like trying to get a frog
to understand. Frogs can't understand because some of the underlaying
simple repetitive rules are different than that of a human (which ends
up as them with the properties of frogs as opposed to humans).

Likewise... an AI will never truly understand like a human unless we
first understand our own simple repeating principles. Software today
can only mimick creativity in narrow bands of task-oriented behavior.
While we could create something that appeared intelligent using the
current approach (given enough lines of code)  IMO even if one
eventually passed the Turing test with flying colours it wouldn't be
an iota more intelligent than any computer is today (in the human
sense).

Incidentally...this approach to AI isn't my own theory. Plenty of
researchers for decades have taken this avenue.  It isn't popular in
mainstream programming yet for one simple reason. We usually can't get
a computer to do much useful using this method.  However, the reason
why that is seems fairly evident. Even if the rules are simple, given
the number of possible mathematical permutations possible unless we
know our own set of rules first... the probability is extremely low
we'll randomly stumble upon the code for intelligence beyond very
simple tasks. (although if the horsepower of advanced quantum
computing ever arrives...we might be able to use the trial-and-error
simulation approach)

I stated  "A search of +fractals +neurons returned 1.9 million hits".
The number is roughly 1.2 million hits. (The difference isn't relevant
but it gets under my skin when I see other people post inaccurate
figures).

Indeed...

... looking at the abstract referred to by the first
link in this list, it appears the authors applied
fractal analysis to the study of action potential
patterns in certain neurons, *not* to model the
structure (shape) of those neurons. Do you understand
the difference? I do.

-RKN

e.g " Your wording indicated - to me anyway - a misunderstanding of
the structure of a neuron, as if to say that if  you contunually
divided the neuron you'd see its structure "repeated" in the smaller
pieces. Which
isn't true of a neuron, or any other cell type. "

e.g.  "From here you segued into Mandelbrot sets, which are self-
similar on magnification, but neurons
aren't. "

e.g. " So on your view, since all matter is ultimately made of atoms,
therefore all matter is
constituted of "smaller repetitive structures. Pretty quirky verbiage
if you ask me."

e.g. "You know what they say, "A concept which describes everything
distinguishes nothing."

In either case...  isn't the fact they are using fractal analysis
(that's explicitly used to determine repeating patterns)  my whole
point here? Was I arguing how a computer chip should be "shaped"....
or was I using fractals as an argument that using simple repeating
code to build AI software makes sense because it duplicates aspects of
our brain? (and aspects of our biology in general)

And if you Google some more you'll see fractals can also be used to
determine the  shape of  neurons (or at least some aspects of them)
You'll find plenty of references to the shape of dendritic trees/
arbors  and axon branching being compared to fractals.

"In the present study we apply fractal analysis to this unsolved
problem and calculate the fractal dimension for each dendritic arbour
of a neuron. We will hereby prove that by application of fractal
analysis to the dendritic arbours of these cells whilst ignoring other
neuronal attributes allows for clear discrimination of only three cell
types."

http://www.sciencedirect.com/science?_ob=ArticleURL&_udi¶T0G-4PD4XHV-2&_user
&_rdoc=1&_fmt=&_orig=search&_sort=d&_docanchor=&view=c&_searchStrId
85000721&_rerunOrigin=google&_acctÀ00050221&_version=1&_urlVersion=0&_useri d
&md5 57abc43e3fd11d222c674e9edf5fd6

"A fractal shape can be completely described by a sin-
gle parameter, df, the fractal dimension (the mass of a
fractal object inside a radius r scales as r 0. Objects in
nature are fractal over a finite range of length scales r,
typically a factor of 10 or so. Here we apply fractal
analysis to retinal neurons in vivo and in vitro.
we find that neutons are fractal objects over roughly a decade in r.
We also suggest three possible diffusion-limited pro-
cesses that could be related to the fractal shapes observed.....

http://74.125.95.132/search?q=cache:CAJHDq1GOeMJ:polymer.bu.edu/hes/a... &hl=en&ct=clnk&gl=ca

"Yet another possibility might involve the creation of a fractal
architecture in which multiple copies of the synfire neuronal pool
arise as a result of the fractal geometry of axonal branching
(Bieberich 2002). "
http://cogprints.org/4432/1/single_neuron_theory.htm

My guess is there are probably more aspects in a neuron that are
fractal (although I cannot confirm as this isn't my area of
expertise). There are also many other aspects of our brains have also
been compared to fractals.

http://epub.ub.uni-muenchen.de/1675/)

http://www.pnas.org/content/103/51/19518.abstract

http://www.google.ca/#hl=en&q=%2Bneuron+%2B%22fractal+analysis%22+%2B...

http://www.google.ca/#hl=en&q=Brain+fractal&start &sa=N&fp=f616e92003ac2179

These concepts have been applied to existing computer neural
networks....using fractals patterned after dendrites and axons. (So
again it ain't just me saying this.)

http://en.wikipedia.org/wiki/Neural_network#Overview
http://www.fractal.org/Life-Science-Technology/Publications/Fractal-N...
https://mitpress.mit.edu/books/ARBHH/arbib-toc.html

Lets suppose we want to design a software heart that appears to beat
like a human. If we took the approach of just a pulsing object that
only has the appearance of a heart practical any new factors that were
introduced to test whether our heart accurately simulated a real heart-
would fail. We would be forced to code in behaviors  one-by-one to
simulate reactions.

On the other hand, a heart is made up of all sorts of repeating cells.
Thus if we designed software that  simulated a heart on a cellular and
physics level the simulation (if written correctly) would result in a
heart capable of beating naturally. (as well as reacting to other
stimuli as a human heart might). Behaviors wouldn't need to be
explicitly coded.

I'm applying the same concept to intelligence. Why machines can't
think well is because we are building them with superficial hearts. We
could keep adding behaviors (what we currently do) but this isn't the
way a heart works. The more sensible way (if we wish to simulate
actual intelligence rather than just mimic behaviors one-by-one) would
be to duplicate how our brain works on a microscopic level. This way
human-like behavioral tendencies would arise as a natural consequence
rather than each have to be explicitly coded.

Or at least this is the theory any how. In practice the behavioral
approach is much more practical and common because it solves immediate
problems, Over time more and more behaviors are being added to
computers so that one of these days we are probably going to have have
computers that seem intelligent. but the principles of that
"intelligence" function fundamentally differently than humans. A real
AI isn't only one that passes a Turing test but one where behaviors
aren't explicitly coded.

There are two big challenges in this approach

a. Can we create a software brain with similar structure to a human
one?  This is extremely challenging but I see no reason why not
(eventually).

b. Can we define the underlaying physics accurately enough for that
brain to function like a human? This one is a wild card. Since we lack
an absolute understanding of physics perhaps our brain would never end
up completely in the range of a human-like behavior.

The correct approach is to first discover how the mind works.
The challenge is not to duplicate the brain but to recreate the mind.

Epistemology anyone?

RoL

If we took only the high level behavioral approach to programming even
if we had a machine that passed the Turing test we'd still forever be
wondering if it was really "intelligent" (in the human sense).  What
would distinguish it as suddenly "Intelligent" versus prior
generations that simply had less lines of code?  A piece of software
that was programmed to work on micro principles similar to the
brain... that subsequently started to behave like humans do without
being instructed how to do so... would be a far more convincing
argument we understood the finer mechanisms of consciousness..

In any case, we appear to be saying much the same thing in terms of
ultimate goal despite different descriptions.