[ExI] Computers, qualia, 'symbol grounding' (@Gordon)

Brent Allsop brent.allsop at gmail.com
Sun Apr 2 21:59:54 UTC 2023

On Sun, Apr 2, 2023 at 12:54 PM Will Steinberg via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> Mr. Groks The Sensorium, you keep claiming that ChatGPT hasn't 'solved'
> the 'symbol grounding problem' but I have yet to see any evidence for this,
> only evidence that ChatGPT is unlikely to experience the same qualia that
> we experience.  But I have seen no proof that the AI has NO qualia with
> which to ground symbols with, and if you did have that proof, you would
> become a very famous philosopher.

If you have never experienced the new color 'grue' you are still able to be
abstractly aware of a 'grue' quality.  You could objectively observe and
describe all the causal properties of grue.  At that point, you would be
like an abstract computer, and you can know everything abstractly.
Now, when you take that grue stuff, and computationally bind it into your
consciousness, you will finally be able to directly experience it, and
finally know what all your abstract descriptions of greness are merely
describing. Your definition of grue will finally be grounded, and you will
be able to say: 'oh THAT is what grueness is like."   It's not a matter of
'proving' that before you experience it, you are different.  It is simply a
grounded definition.

How do you know that qualia aren't fungible?

Redness is simply a physical quality of something.

> Was Hellen Keller a p-zombie just because she didn't have grounded symbols
> for sight and sound?

See above grue example, to understand how as far as grue goes, you are like
Hellen Keller, a computer, being different from someone that has a grounded
definition of grueness.

How do you know that it's not possible to build a model of the world using
> only whatever qualia computers experience as the base?

You can represent red things in the world with a redness quality.  Or you
can use a grenness quality.  Or you can use the abstract word red.  But the
abstract word red, is not a quality, it is only an abstract word.
You can build a model of the world, using any and all of these.  The
different models just won't be qualitatively like each other.

You seem to believe that if you reverse engineer language, you are left
> with a bunch of empty spaces for qualia, and that self-consciousness is
> dependent on these atomic experiences.

I prefer the term 'elemental' to atomic.  After all, some people predict
that qualities are of something at the sub atomic, or quantum level
<https://canonizer.com/topic/88-Theories-of-Consciousness/20-Orch-OR>.  The
'elemental' level is simply whatever physical level is required to fully
describe a composite conscious colored experience.  There could be an
infinite amount of physics below redness, but you need not model below the
elemental level to fully describe elemental redness.

> What's to say that any qualia can't take the spots of the ones we used to
> develop language?  We can communicate with people who are deaf and blind
> from birth.  Even someone who had none of the external senses that we have,
> but a single bit of input/output of some kind, could communicate with us.
> Imagine for a second there are aliens which only perceive the world
> through magnetic fields.  We have no possible way to reckon the qualia for
> these fields, but we CAN produce the fields, and measure them.  And with
> this we could both send and receive magnetic fields.  You might say that
> without known constants to both refer to, we could never talk with these
> beings, but is it true?  Can you say without the shadow of a doubt that
> qualia cannot be inferred from the entirety of language?  After all, at the
> end of the day, past the sensory organs everything is condensed into
> electrochemical signals, same as language.  So wouldn't you perhaps think,
> with utter knowledge of one side of that equation, that it could even be
> simple to reconstruct the other?

You're missing the point.  Redness is simply a physical property of
something in the world.  You simply computationally bind whatever that is,
into their consciousness, then you tell them: "THAT is what I use to
represent red information."
Or, if they already use redness to represent something, say green, then you
could simply say: "My redness is like your greenness, both of which we call
red."   The point being, you simply need to define your symbols in a
physically grounded way.

If I was able to perfectly recreate a human eye and brain, and knew the
> neurophysocal content of a 'standard' red quale, would I not be able to
> make that brain experience the red quale?

Only if you use whatever physics has a redness quality.  Otherwise, no,
although you could  use some other physics to 'code' for that, as long as
you had a grounded dictionary so you could know what that code represented.

>   Do you think it is possible that access to the relations between all
> language, ever, could enable one to reconstruct the workings of the
> sensorium, and then infer qualia from there?  What if the entity in
> question not only had this ability, but also experienced its own types of
> qualia? (You do not know whether this is the case.)  Would that make it
> even easier to reverse engineer?
> I simply think--or rather, I would say I KNOW--that you can't possibly
> know whether a system, of which you do not know whether experiences any
> qualia or not, using an inference tool on language of which you have no
> personal access to verify whether can reconstruct qualia, and which
> actually, not even the people who make it understand fully what is going
> on, is conscious of itself.
> Btw, is that even what you are arguing?  You seem to be jumping back and
> forth between the argument that ChatGPT has no qualia (which again, you
> can't know) and the argument that it has no awareness of itself (which
> again, again, you can't know).  These are very different arguments; the
> first is the most important unsolved problem in philosophy.

You are wrong, you can know this.  There are the 1. weak, 2. strong, and 3.
strongest form of knowing this.  See the "Ways to EFf the Ineffable"
section in the "Physicists Don't Understand Color

> This is really getting into the weeds of the subject and I don't think you
> should speak so surely on the matter. These problems are the hardest
> problems in all of philosophy, neuroscience, theory of mind. There are
> NUMEROUS thought experiments that at the very least bring the sureness of
> your opinion below 100%.

There is evidence of a consensus supporting RQT
and all these people are predicting this isn't a hard problem at all, it's
just a color quality problem <https://canonizer.com/videos/consciousness>.
And all we need to solve this problem is physically grounded definitions
for the names of physical qualities (not the qualities things seem to be)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230402/52c2ff5d/attachment-0001.htm>

More information about the extropy-chat mailing list