Re: Qualia Question

"1Z" <peterdjones@xxxxxxxxx> wrote:
> Curt Welch wrote:

> > You brought up the idea of "mechanical" when you wrote:
> >
> > "There is no reason to insist that everything has to
> > be explained mechanically."
> >
> > I only used the word in response to your comment.
>
> I only said that because of Lester Zick!

That's what I get for commenting without trying to follow the real context!
:)

Like I said, I have never been able to understand Lester's definition of
mechanical.

> > We known for a fact
> > that different hunks of matter which make up human bodies, are still
> > functionally the same in terms of being intelligent.
>
> There is no evidence that the implementation (the biochemistry) varies
> much more widely than the function (intelligence).

True. But we only have one example of intelligence to study, so it's no
surprise that the one example we have to study (humans), have a lot in
common with each other. And since the device is so large and complex, and
our tools so primitive, it's no surprise we are still working on the
reverse engineering problem. Nothing found in the brain so far indicates
that anything odd is happening in there - it's just a big signal processing
device which is still too complex for us to figure out. But progress is
being made daily, and by all predictions, we will figure it out before too
long (a few hundred years at max). There's no indication so far that there
is anything there beyond our ability to figure out.

And since we don't understand the function of the only example we have, it
should be no surprise that we have not built a duplicate yet.

What do you think would happen if you took a standard PC back in time 100
years and asked the leading scientist and engineers of the day to try and
reverse engineer the device? They would have no clue what they were
looking at. The don't even know what a transistor is, and they don't have
any electronic tools that work at the frequencies our computers run at
today. The vacuum tube was only invented in 1904. All they could figure
out was that the device as producing a ton of RF noise. It would take them
50 years to figure out what the thing was doing. And that's a simple PC
designed by humans. They would have to first master all the technology it
was built from, and then create tools just to study it.

The fact that we have been looking at the brain in all this time is the
same type of problem. Here we have a technology we have not yet mastered
and we are slowly developing the knowledge needed to understand what we are
looking at.

> You are not comparign liek with like. Yes there are variations between
> people,
> but they are all variations on a theme. The kind of difference that
> exists
> between a silicon chip and cell is vastly freater than between people
> with
> O-type and A-type blood. It's a whole different ball-game.

Yes. Of course. But I've studied, designed, and built a ton of different
machines in my life. In all that study, and work, I found that the
implementation can easily change and still provide the same function.

One common function in simple machines is just size and shape. If you want
to duplicate the function of a screw, you have to get the size and shape
correct - there's just not much room for change.

But on the other end of the scale of machines, we find electronics, and
computers. Their functional requirements has to do with producing the
correct type of signal - i.e., the correct flow of electrons in and out of
an electron pipe (write). In these types of machines, the same function
can be duplicated with machines that have almost nothing in common in terms
of size and shape.

You can build a radio out of vacuum tubes, or transistors, and they both
work just fine even though the physical and logical operation of tubes is
completely different than transistors. You can use simple resistors and
capacitors to create a tone control which acts as a filter to shape the
sound. Or, you can build the radio all in the digital domain, so it
receives the radio signal in a digital format and does the tone control by
running mathematical signal processing algorithms on the numbers, instead
of doing the work in the analog domain.

What we know about building signal processing devices is that the hardware
implementation is irrelevant as long as it produces the correct signal, and
transforms the signals correctly in the end.

There's a lot we don't know about the brain, but the one thing we know
every well is that all our conscious intelligent behavior is created by the
brain which is acting as a signal processing device. And what we know
about signal processing devices is that it makes no difference what the
hardware is made from (tubes, transistors, or neurons) - if you process the
signal correctly, you get the same function.

You can encode sound signal in a wire for a phone call, or in a scratch on
a wax cylinder which is read with physical contact, or in the pits of CD
red by a laser. In the end, it still sounds like a person talking when the
signal is used to specify how the air gets moved.

With all the work we have done, we have built a huge body of machines which
attempt to duplicate intelligent behavior, and all of those devices, that
come at all close to duplicating different parts of intelligent human
behavior, have to date, been signal processing devices - and none of them
used neurons yet that didn't stop them from producing the correct signals.

All this adds up to a huge body of evidence telling us that the solution to
AI is simply a matter of building a single processing device with the right
function and we know for a fact that signal processing devices don't need
to be made out of neurons to work.

> > All humans and all machines are built from these same building blocks,
> > and the nature of what we can build is of course directly controlled by
> > the nature of the building blocks we build with. So of course the
> > nature of the building block is the foundation of all the functions a
> > machine build from these things can have in our universe. In the end,
> > the "function" is defined in terms of these same building blocks - so
> > you can't separate the function from the building blocks.
>
> OK. So you can't dulicate neurons in silcon. So AI is impossible ?

You seem to have missed my point. The building blocks everything is built
from is atoms, not neurons. Transistors and neurons and vacuum tubes are
all built from the same building blocks - atoms. We will be able to
duplicate the function of the brain using transistors. I'm quite sure of
that.

> > For example, a "machine which spits out H2O molecules" is a functional
> > description. But the function is defined in terms of the actions of
> > these building blocks.
> >
> > If you take a device like a network router on the internet, it's easy
> > to see how they are functionally equivalent. It makes no difference
> > what brand, or design, router, might be located in the network between
> > my computer and Google's computers, because I can access the web site
> > either way. They are functionally the same.
>
> Presumably you mean at a high level of description. You also seem
> to be saying that physical differences are functional differences at a
> low level
> of description.

Yes exactly.

When you say "high level" all you are really saying is, "when I leave out a
lot of the details". The details you include however are all defined
ultimately in physical terms because the meaning of all those terms were
build on words, which were build on words, which started with words which
described physical stuff. All our meaning is ultimately grounded back to
physical reality.

> > It's impossible to create a functional description which is not based
> > on the physico-chemical properties of the universe.
>
> Any kind of functional description ?

Yes. I think. Did my double negative confuse you?

I was trying to say that all functional descriptions are based on the
physico-chemical properties of the universe - including the functional
description we call a NAND gate.

> So you can't say that a TTL NAND
> gate and a CMOS NAND gate are functionally equivalent ?

Of course you can. But the meaning of NAND gate is defined in terms of
atoms. There is no other way to define it in this universe. The concept
of a "signal" only has meaning in terms of the interaction of physical
matter.

When you talk about a high level functional description like NAND gate, you
leave out the lower level functional description that ties it to physical
reality. We can specify the functional description of the NAND gate with a
simple truth table which seems to have no connection to physical reality.
A CMOS NAND get might for example be a 3.3V device and the TTL a 5 V
device, so that they are not in fact directly compatible. But this
encoding of logic values was left out of the truth table description of
what an NAND gate is.

However, it's impossible to build a real NAND gate without filling in the
missing functional pieces. You must fill in the missing functional parts
which were missing which tie the functional description to physical
reality.

Inversely, the only reason we were able to define the function of a NAND
gate in the first place was because we were describing the behavior of
physical matter. So, the description of a NAND gate is based on logic
values, and values can only exist in physical matter - so therefore the
functional description of NAND, is tied to, and depended on, the properties
of physical matter in our universe.

> Or an ultimately
> fine-grained functional description ?

Don't know what point you are getting there.

> > When I say only the
> > "function" is important, what I'm really saying is that only a subset
> > of the total properties of some configuration of matter is important.
> > Function is always defined in terms of, and is grounded to, the
> > properties of matter in this universe. It's impossible for us to
> > create a functional description which is not grounded in the physical
> > universe.
> >
> > So when I say only the function of a human is important, I'm not saying
> > anything we don't already know as a fact. We already know humans don't
> > have to be identical clones just to be intelligent. So we now that
> > only a subset of the behavior of the description of the matter we are
> > made from is important to make us "act intelligent".
> >
> > The only thing in question here is whether our physical make-up alone
> > is enough to explain human behavior, and human subjective experience.
> > I believe it is, and I think it's an obvious "fact" that it is, but I
> > don't have the simple evidence to make everyone else believe it yet.
>
> But that -- the absence of ghosts or souls -- doesn't justify AI. You
> need
> to argue that some level of functioning can be duplciated in a
> radically different material substrate.

I did that above for you with the discussion about our experience with
signal processing devices.

Did I mention I can build a signal processing device out of water or light,
or chemicals as well?

> > If we can
> > create robots that act human-like in all ways, including talking about
> > how things "feel" to them, including being confused about why it feels
> > like it does to them, then I think it will become obvious that humans
> > are just "machines" as well, and we will at that point know how to
> > describe what type of machine we are as well.
>
> That is approaching the problem from the wrong end. Why should we think
> such duplication is possible in the first place ?

Because I'm an engineer that wants to be the person who built the first
human level AI. Why I want to do it has everything to do with how I should
approach the problem.

The point is, we don't know if it's possible, but there is also no evidence
showing it's impossible. We just don't have enough data to answer the
question either way. Those of us that want it to be possible approach it
from my direction - we assume it's possible and will continue to do so as
long as we can explain away any data that suggests it is not possible. And
so far, that's been easy to do.

People that don't want it to be possible, or for whatever reason, have
simply chosen to believe it's not possible, will not change their mind
until someone can offer proof that it is possible. No one has that proof
yet. It will only come in the form of machines which act intelligent and
human in all ways. Until people on my side of the fence can do that, we
have no hope of winning an argument that AI is possible. Because the
ultimate defense from the other side is always - if it's possible, why
can't you build one?

When we get human-level machines running, then we will be faced with a new
level of debate - are they really conscious? Are they sentient? I think
that debate will be much like the evolution debate. Many people will just
not accept that the machines are conscious like humans are conscious. In
the end however, we will both know how to build these machines, and we will
have a complete understanding of how the brain works to show that it is
performing the same function as the machines. So there will be no doubt
about what we are. Like evolution, which still, after 100 years, is not
fully believed by more than half of the people in the world, it will
probably take a long time for society to accept what has been found and to
change the culture to embrace the idea.

> Insisting that people are
> entirely physical doesn't demonstrate that.

Right, it doesn't. I can't demonstrate it. I can only demonstrate that
there are answers to all the questions that are consistent with the point
of view that AI is possible. If you take my stance on explaining what we
are, it gives us answers to all the questions. The huge missing pieces of
the puzzle however is that we don't have a conscious intelligent machine
yet. Until we get that job done, everything else is just idle chat to a
non believer.

To a believer however - someone like me that wants to solve this problem,
it's not just idle chat - it's important data about the path we must take
to solve AI. If it's possible, everything must be physical.

> Physical things can have unique, unreproducible properties.

Like what? And what evidence do you have to support the idea that if we
cloned the device at the atomic level, that the clone would not have all
the same properties.

Give me an example of a signal processing device that is unique and
unreproducible.

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