Re: "true" AI Hardware Development

jalegris@xxxxxxxxxxxx wrote:
Curt Welch wrote:
jalegris@xxxxxxxxxxxx wrote:
Curt Welch wrote:

But cars are artifacts to begin with, so coming up with a functional
equivalent is straightforward. Natural phenomena resist functional
interpretations.

All of modern medicine and biology says that's just not true.

Natural phenomena is functionally complex, but so far, has shown no
resistance to functional description at all. Life is being very well
explained functionally in terms of its atomic make up.

Try making a car that is functionally equivalent to a
horse.

We don't have the tools to do engineering at the atomic level. That's
why we can't build machines that are functionally equivalent to a
horse. But we understand a hell of a lot about what a horse is in
functional terms down the the atomic level. What we still don't know
is far greater than what we do know, but there's no indication that any
of it is unknowable or that there's any "magic" stuff there that is
going to prevent us from a complete functional understand the
functional understanding of every atom in the body of a horse (or a
human).

But intelligence is assumed to be different - it is the only natural
phenomenon that is assumed to be equivalent to its own simulation.
Why should this be?

Because so far, we have found nothing but signal processing going on in
the brain. And if that's all that is happening, we know for a fact
that we can build functionally equivalent machines. Until someone
points to something that can't be explained in signal processing terms,
it's perfectly valid to stick with the assumption that the brain, and
all the behavior it creates, is nothing more than the actions of a
signal processing machine.

We know that it is not a digital signal processing machine, so there
are interesting questions to be answered about whether the same
function can practically be duplicated in a digital computer. We just
don't know the answer to that yet. However, I for one have no problem
switching to analog signal processing techniques if it will give us a
practical advantage over digital in producing intelligent behavior. So
far however, I've not seen any indication that it would be required
(but I can think of a few examples of why it would).

I think it's because AI buffs implicitly accept the Computational
Theory of Mind - a Turing machine is the only thing that actually is
functionally equivalent to its own simulation.

You hang up on the concept of a simulation is misguided in my view.

And analog AM radio is not a simulation. I can build two different
analog AM radios and they will be different. Neither will be a perfect
simulation of the other. Even if they are build from the exact same
schematic, the components that make them up will have very different
characteristics. The matching resisters in each radio will have
different resistance values. The transistors will have different gain
and voltage drop and resistance characteristics. The speakers will
have different voice coil resistance and the cones will be a different
weight. Every part will be different. But yet, both manage to work
just fine as AM radios.

Likewise, no two humans are the same. We have billions of examples of
intelligent machines, with no two being the same. Yet we have no
problem creating a functional class and putting all these different
machines into the same class (intelligent humans).

But brains don't just
process information - they also process matter.

What on earth do you mean by that?

Do you understand that all computers are physical machines that work by
pushing physical matter around (i.e., they are all 100% mechanical
machines)? How is a brain any different?

A "truly intelligent" simulation is going to need some real meat.

Meat is made up of electrons and protons. Computers are made up of
electrons and protons. It is already the same "meat" at the core.

Humans are just machines made up of electrons and protons with arms
that have a central signal processing control system that makes the
arms move. We already know how to build machines out of electrons and
protons with a central signal processing core to make its arms move in
intelligent ways.

There are many people that are hung up on the belief that humans are
somehow special (we posses some non-physical magic about us - be it a
soul, or feelings, or emotions, or consciousness - it goes by lots of
different names). It's a concept built into our culture in more ways
than I can count. But with all the advances in science, we now know
that there is just no evidence to support these beliefs. Humans are
just complex machines which we have the power to functionally describe
and understand completely. Everything science has uncovered confirms
this. Many people resist this idea, but science has also uncovered
enough information about human behavior to explain why humans resist
this idea. It's simply because it scares them to believe it. It scares
them to accept the idea that they are nothing more than machines
created by a natural process of evolution. And this just means we are
conditioned to find find behaviors that make us stronger - and if
telling ourselves lies about how special we are makes us stronger -
then we do it.

I just don't buy any of those stories people like to tell about how
special humans are.

There is still much we don't know about humans. So there is room for
something to be uncovered to show that we are in fact more special than
we currently realize. But until that data is uncovered, I'm not going
to choose to believe it. People that are scared of facing the facts
about what we are however, seem to cling to that possibility as their
last life raft of salvation. I don't.

The human body is a wonderfully complex piece of engineering. It can
self reproduce and survive on it's own out in a forest. That's
absolutely amazing and something we haven't even come close to in our
machines. But the brain in fact is not all that amazing. It's not
much more amazing than our computers. The internet is a machine many
times more complex than any human brain. We have already build signal
processing machine far more complex than anything nature built. The
brain just doesn't impress me nearly as much as the entire human body
and DNA based life in general.

One day soon, (within a few decades) we are going to figure out enough
about how to build these types of signal processing machines that the
human brain is going to start to look out right stupid. Humans already
can't touch the processing power of computers. They do many things
with speed and accuracy untouchable by humans. And soon, there will be
NOTHING that the human brain can do which can't be done orders of
magnitude better by man made machines.

And when that happens, all this silly debate about how special humans
are, is going to finally stop (at least it will be gone from the
educated segment of society).


I am not saying that humans are so special, but human artifacts
certainly are. They are special because, unlike most naturally occuring
phenomena, they can be effectively specified (leading to replication)
by functional descriptions. Proof: language is independent of the
transmission medium, therefore it is pure (or almost pure) information,
therefore if a description of an object can be communicated by language
such that others can replicate it then it is a functional description.

So, if wad up a ball of string and dunk it in different paints and throw it
on a tree, you think that's an artifact that will be easier to functionally
explain than something created by evolution? Do you think you could
specify and replicate it? Not in your life. Yet it's a human artifact.

Yes, we tend to build things that are simple to understand. We do it to
maximize the tools usefulness to us. The more complex it is, the less
useful it is to us. Evolution didn't need to "understand" what it was
creating, and as such, it tends to create things far more complex then the
things we like to create.

Art is one large example of work created by evolution. An artist will
evolve the work towards it's final form without any need for creating
simplicity to make it easy to understand. And as such, most art created by
humans has all the same levels of complexity as life.

In general, the above is true for artifacts, but false for natural
objects. Why?

Because you are confused.

Because we humans were not privy to the design of nature.

Exactly like one human would not be privy to the design of another humans
art work. Which is why it's just the same as nature.

We can try to find out through science, which provides approximate
functional descriptions but, as is abundantly evident, scientific
characterizations of macroscopic natural phenomena are generally rather
crude. You say that modern medicine and biology are providing
functional characterizations, but they are not functional enough for us
to build working organs (unless you consider a plastic heart to be the
functional equivalent of a real one, by which standard AI has
apparently already succeeded - it's brittle and applicable only to
special cases).

The fact that we can't build a heart has nothing to do with a lack of
functional specification. It's a lack of tools for building molecular
level machines. Our hands aren't that small. That's the only thing stopping
us.

We know, down to the atom, how a large part of the cells work. But we have
no tools for building our own cells. Until someone events better tools for
manipulating atoms, we won't be able to duplicate it. This has nothing to
do with it being unable to be functionally specified.

Physics, chemistry and molecular biology are notable exceptions.
Functional characterizations at the microscopic level seem to work -
and that's just my point. A successful simulation of intelligence is
going to have to get right down there and account for all the messy
goings-on that computationalists choose to dismiss.

We don't have to duplicate the complexity of DNA based cells in order to
build a machine to dig a ditch. Why would you think we would need it to
build a machine to type intelligent messages on Usenet? There's really no
evidence to support the idea that we would have to duplicate the brain's
implementation at that level of abstraction in order to create intelligent
machines.

Though, once again, some people think that humans are so "different" from
normal machines (like computers) that there must still be something huge
missing. I'm just not one of them.

So, this is one way
we can get an accurate functional description of intelligence - get
right down to the meat of the matter (literally).

Sure. If we understood the entire human body down to the molecular level,
we would understand intelligence. At least I believe we wouldn't need to
go any smaller. But I also happen to believe that would be overkill.

Computers are easily reduced to information-theoretic functional
descriptions because we humans designed them as such.

True. We designed them to be very simple so we could program them.

That's what
engineering is - the production of language-based (and math-based)
functional descriptions.

Mostly, but not completely. There's a lot of practical trial-and-error
engineering going on all the time as well. People constantly build things
that work, but yet they have no clue why they work (or think they
understand why it works when in fact they don't - which they learn later).
That's normally considered bad engineering, but it's done all the time.
For example, you put in a bracket, and if it breaks, you put in a larger
one. You never once had to stop and actually calculate the force and
stress on the piece or actually document in a blue-print the
characteristics of the bracket you ended up with. It's engineering by
evolution and produces the same type of complexity evolution produces when
it "builds" life.

But, as I say, the brain doesn't just process
information - it processes matter. Pieces of structured matter move
around and contribute more than information - they contribute energy,
and chemical properties. You cannot reduce the operation of neurons to
just information processing unless you reduce chemistry and physics
too.

Well, as I said, all machines are mechanical. It's impossible to build one
that is not 100% mechanical. So the brain doesn't have anything special
about it in that regard. Saying that it processes matter is just stupid
because all machines, including all computers, process matter. Trying to
pretend the brain is special because it does the same thing every machine
does is just silly.

The information processing language we use to describe computers is, as you
said, just a functional specification language. Is it one which is good
enough to specify a car? Nope. Can we build cars? Sure. How do we do
this? We just use other functional specification languages. We have many
to choose from. We draw blueprints and talk about the static and dynamic
forces at work on the physical parts of the machine. We also have a large
language of electronic circuit behavior to work with. And more for the
description of the interaction of chemicals. No one is limiting us to
using computer software to specify the operation of an intelligent machine.
However, so far, nothing we have found in the brain indicates that we would
need something other than what computers can do. So until we find
something happening in the brain that can't be done with a computer, and
which seems to be needed to create intelligent behavior, we will continue
to use the tool that best fits the job - the computer. Once we find _any_
evidence it's not the best tool, we will simply switch and pick up the tool
that is correct.

When dealing with a mystery, we always tend to make the problem harder than
it turns out to be. We make up all sorts of complex answers to explain why
we haven't figured out the mystery. But in the end, most mysteries turn
out to be dead simple. Most the time, we just failed to look in the right
place. People think there must be something complex going on to explain
the brain and and feel like if it were simple, we would have figured it out
by now. I think it is simple, and one reason we haven't figured it out, is
because most people keep looking for something complex to explain it - they
are looking in the wrong place.

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

Relevant Pages

  • Re: "true" AI Hardware Development
    ... we can't build machines that are functionally equivalent to a horse. ... we have found nothing but signal processing going on in the ... matching resisters in each radio will have different resistance values. ... no two humans are the same. ...
    (comp.ai.philosophy)
  • Re: Robot survival instincts
    ... But if we killed the humans, ... I don't really buy your premise that machines are going to take over ... that will turn control of the earth, ... maximize the odds of our genes surviving and nothing else. ...
    (comp.ai.philosophy)
  • Re: Merging with machines inevitable, scientists say
    ... I see the odds of the machines taking over being exceptionally slim. ... Humans are programmed by evolution with the goal of surviving. ... build AIs with the goal of surviving. ... advanced AIs for the sole purpose of taking over control from the humans. ...
    (comp.ai.philosophy)
  • Re: Robot survival instincts
    ... Reinforcement learning machines have no desire to ... Even Asimov's robots had survival instincts. ... these machines so they do the best job while not being a threat to humans, ... Linux survives because it exists in an environment that ...
    (comp.ai.philosophy)
  • Re: "true" AI Hardware Development
    ... we can't build machines that are functionally equivalent to a horse. ... we have found nothing but signal processing going on in the ... matching resisters in each radio will have different resistance values. ... no two humans are the same. ...
    (comp.ai.philosophy)