🚀 go-pugleaf

Don't blame me, I voted for tycchow@riesz.mit.edu (Timothy Y. Chow).
>Let me see if I've got this right.  Many of us have speculated at some time
>or another whether or not the color green "looks the same" to other people
>as it does to us.  Maybe other people, when they see a red cape, see what I
>see when I see green, but call it red.

You're right that there's no way to measure the quality of "redness"
perceived by others, but like the consciousness of others, this issue
can be resolved by communicating with those others.

Consider a lemon on a purple tablecloth.  I can point to the lemon and
say, "That's yellow," and you'll agree, because whatever it is that you're
perceiving, you've been taught to name it "yellow."  I can point to the
table cloth and say it's purple, and you'll agree for the same reason.
But if I say, "Look at that yellow lemon and see how it blends in to that
purple tablecloth; you can barely see it there," you'll disagree (I hope),
because the qualities that you perceive as yellow and purple *clash*
with each other.  They do for me, too.  So we know that the qualities
we perceive as yellow and purple bear the same relationship to one another
within both of our minds -- which means they're either the same, or so
orthogonal as to make differences unimportant.

When I was about twelve, I'd been thinking about this issue, and my
brother happened to mention that he'd always thought the rainbow had
a discontinuity in it:  "Red, orange, yellow--that's a logical
progression.  But then comes *green*, which doesn't look anything
like them; it's like the rainbow just makes a left turn out of
nowhere."  When I realized that I had the same thought I knew with
certainty that we perceived colors in, for all intents and purposes,
the same way.

>Consider the Turing test.  Suppose a machine passes the Turing test.  Is
>it thereby conscious?  In principle it seems that a very simple but
>extremely detailed computer program---namely, one that has a vast number
>of "if-then" statements covering every possible stimulus that it might
>encounter---could pass the Turing test, but would leave many uncertain
>about calling it "conscious."

I doubt if that machine could fool anyone for more than a minute.
Suppose it contains the if-then statement, "IF stimulus == 'How old
are you?' then response = '24, why?'".

So the conversation progresses:

ME: How old are you?
IT: 24, why?
ME: Oh, just wondering.
IT: I see.
ME: How old are you?
IT: 24, why?
ME: How old are you?
IT: 24, why?
ME: How old are you?
IT: 24, why?

In reality, if I said the same five things to a person, it'd go like this:

ME: How old are you?
THEM: 24, why?
ME: Oh, just wondering.
THEM: Ah. My birthday was just last week, actually.
ME: How old are you?
THEM: You just asked me that. 24, remember?
ME: How old are you?
THEM: [knocking on forehead] Hello? Anybody home?
ME: How old are you?
THEM: [becoming concerned] You wait here--I'll get a doctor.

The point is that a real consciousness doesn't have a set of answers
to any given stimulus; the answers change with experience.  Human beings
display flexibility, an ability to learn, a sense of humor, and hundreds
of other qualities that even the most elaborate lookup table imaginable
could never duplicate.  A machine that displayed all the same qualities
would be, in my opinion, conscious; in any case, it would possess something
so perfectly orthogonal to consciousness as to render pointless any
speculation about whether it *really* was.

>The above are the intuitions underlying Dave Chalmers' argument, I think.
>A complete catalogue of physical facts about vision might still leave us
>wondering whether or not other people's visual experiences are the "same"
>as ours.  Similarly, even if no human being can devise a test that will
>distinguish between Dave Chalmers and the if-then monster, and even if we
>have a complete listing of the if-then monster, we might still be unsure
>about whether the if-then monster is conscious, even though it assures us
>that it is.

It seems to me that there's more to this than you're seeing.  A complete
listing of a Turing program would not be sufficient to know how such a
program would react to a given situation after it'd been running for
ten minutes; it would have accumulated experience in that time which
could not have been predicted by reading the source, which would affect
its responses.  The listing is simply the structure; the structure
plus the current state is what you need to examine.  You might be
able to freeze the system, coredump it, analyze every register and
memory location, hook up the computer's sensory inputs to a data
cable which would provide it with a precisely controlled sequence
of bits, fix its random number generator to only feed it numbers
from a particular sequence that you control, and use all this
information to predict its *next* response--but this wouldn't prove
any point.  You can only determine the consciousness of a system by
*communicating* with it, and what I've just described is not communi-
cation.  You might as well "prove" that I'm not conscious by pointing
out that whenever you subject me to ten thousand volts, I scream.

-- 
"This coffee is my blood, this breakfast burrito is my flesh.  And
 it cost me 89 cents, so get your goddamned own." - Jeff Lester, "Saint 11"