Re: Symbolic AI, a waste of time?

"bob the builder" <brulsmurf@xxxxxxxxxxx> wrote:
Don Geddis wrote:

"bob the builder" <@hotmail.com> wrote on 7 Jul 2006 07:15:
I came to the following conclusion: The best way to realize strong AI
is to stay very close to the human implementation.

All of our devices are so far from human-level performance, that it's
premature to conclude that you know the proper engineering path.

Agree, maybe i only know the path most likely leading to strong AI. :P

Airplanes are not built like birds.

So they say, but i think this argument is working against you. You
could say that a bird can fly beacause it has "fly-ness". This fly-ness
can be seen on an abstract level. We most define this fly-ness and use
it to create an airplane.

The wright brothers didnt take this approach. They stayed very close to
birds. Ofcourse a plane doesnt need to leg eggs and mate with other
airplanes. But the design of an airplane stayed as close as possible to
the bird design. I would even go as far that if people stayed even
closer to the bird design there would have been hanggliders 1000 years
ago.

The Chinese had kites over 2000 years ago. Maybe they had some that could
lift a man?

Modren hang gliders depend on materials like aluminum and carbon fiber
composites, and synthetic fabrics to get the weight to lift ratios low
enough. I don't know if the needed materiars were aviable in the world
1000 years ago.

However, doing a bit of searching on the net, I found that Otto Lilienthal
in the late 1800's was the pioner in hang glider design. His book is what
drove the Wright Brothers. He built and flew hang gliders for years but
after thousands of flights, he was eventually killed in a crash.

Part of his book is on line and I found this about his design:

http://www.lilienthal-museum.de/olma/edokvo41.htm

7. It is possible to construct a practical apparatus with an
area of 10 sq. m. and a weight of about 15 kg., employing
willow canes and some textile covering.

So it looks like he used willow canes and textile of some type.

Maybe more research by the Chinese 2000 years ago might have led to a hang
glider made out of silk and bamboo? I suspect however that the progress
Otto made was based on a foundation of mathematics and physics that just
didn't exist 1000 years ago and trial and error building without the
research he did to understand the aerodynamics probably wouldn't have
gotten very far.

Interesting also to this conversation, Lilienthal heavily copied, and was
motivated, by birds. If you look at some of the photographs, you see his
later designed looked a lot like bird wings:

http://www.lilienthal-museum.de/olma/images/f089relo.jpg

But, one of his failings was that he thought flapping wings was going to be
the key to extended flight. This is where he went too far in his attempt
to stick to the natures designs.

Nature is guaranteed to produce well optimized designs. But it's optimized
to a different set of constraints than what man has to work with. So some
aspects can end up the same, and some are going to end up different, for
well optimized designs based on the constraints that man has to work with.

Knowing where to follow nature and where to ignore it is the key.
Unfortunally, until you fully master the technology, you won't know the
answer to that.

It is possible (or even likely) that the different characteristics of
transistors vs. neurons could lead to a different architecture for
implementing human-level intelligence in the two substrates.

Indeed , even likely. Just as the motorcycle engine in the first
airplane caused a different architecture. But because we are forced to
make some choices that differ from the human design doesnt mean we
should let go all together. Although AI will run on a piece of silicon,
It will closely resemble its carbonbased counterpart.

Ah, when you said "stay very close to the human implementation" I had no
clue you were talking silicon. Building a brain out of transistors isn't
close to nature at all in my book.

Or (in other words) the architecture must model neurons, DNA etc. The
Symbolic AI approach to AI is a mistake. This because of the
intimate realation between funtion and implementation.
Anyone begs to differ?

I beg to differ.

Thank you.

You're overselling "the intimate relation between function and
implementation."

You've ignored that it is possible to implement layers of abstraction
(like the network layers in the internet), where a given layer only
depends on some, but not all, characteristics of the layer below.

Its possible, in theory, yes. I would never say there is one path to
AI. But i do think there is one path thats 300 years shorter than the
others.

Yes, and time will tell us what that path is. :)

The key I think is to first correctly understand the problem domain. For
flight, that was understanding the physics and dynamics of flight
(structures moving through the air). Nature gave important pointers that
suggested long "wing" structures would be an optimal system but it required
a lot of research and testing on different shapes and structures to
understand what, of the things man could build, would work best. And the
harder discovers, were the ones that nature didn't give us any help with -
the fact that the best solution for a man made machine was to use a
spinning wing instead of a flapping wing. Finding the solutions when we
have no one like nature to show us the way are aways the hardest to find.

For AI, there's still no clear answer to what the problem domain is. I
personally have a fairly clear answer, but my answer is only one of many
that people are exploring and there are no doubt a few more important
discovers that need to be made (like the spinning wing) which are just not
yet here yet.

One of the hardest things to extract from the brain is it's learning
algorithm. The brain not only grows itself, but it no doubt uses some of
these same growth technologies to constantly re-program itself (to learn).
Understanding how it changes its structures in response to experience is
the most important missing link to the puzzle and one of the hardest to
unravel. It's just very hard to get inside a brain and watch how it
changes.

If we could uncover it's full structure, and it's learning algorithms, we
would and could quickly use them in our own designs. Copying nature would
be a good way to go - if only nature made its design available to us!

It's kinda like trying to copy the design of birds without ever seeing one
- just being told that these animals that can fly exist. We know what the
brain can do by watching how it makes a human body move. But the brain
itself, is hidden - and is going to require a lot more hard work to
uncover.

So, for those of us playing with silicon trying to mimic its power, we have
a lot to figure out on our own for now with all too little guidance from
nature to follow.

I've always like following the idea of a network of signal processing
devices. But is this a flapping wing I'm following that doesn't optimize
as well to silicon as other techniques would? I not long ago switched to
the using pulse signals, again following nature, because they seem to offer
advantages that I didn't see at first. I thought for a very long time that
the pulse signal format was just an optimize that made sense for nature for
reasons such as minimizing energy consumption and one which didn't make
sense for transistors. But I now believe they have other advantages for
the problem.

Discovering which clues from nature to follow, and which not to follow, and
trying to get more information about the brain so we have clues to follow,
is what the puzzle is all about.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: Re: Propping up the theory of Evolution
    ... And so you say, well I am going to give it a better brain, so it can ... be sentient, and so then it is going to be able to use language, so I ... then attribute all the power of God to nature. ... There is no design. ...
    (talk.origins)
  • Re: Ostfront or fancy projects: German aircraft designers in 1944-1945
    ... look for a successful design that did that looks similar ... The J-35 engine was too large to fit in the P.1101 ... German personalities and the sequence ... As an example the X-5 could not land if the wing sweep was greater ...
    (rec.aviation.military)
  • Re: 52/1 Performance in a 15M ship at half the weight.
    ... Cole’s Sailplane factory is in Bend Oregon. ... Company, Cirrus Design, Lancair, and Adam Aircraft. ... Designs that can be made light with small wing areas offer improved ...
    (rec.aviation.soaring)
  • Re: 52/1 Performance in a 15M ship at half the weight.
    ... wasn't suitable for soaring. ... Cole’s Sailplane factory is in Bend Oregon. ... Company, Cirrus Design, Lancair, and Adam Aircraft. ... Designs that can be made light with small wing areas offer improved ...
    (rec.aviation.soaring)
  • Re: Ostfront or fancy projects: German aircraft designers in 1944-1945
    ... look for a successful design that did that looks similar ... Lets see now the claim is the X-5 ... German personalities and the sequence ... Messeschmitts swept wing development or because you don't report it ...
    (rec.aviation.military)