[ExI] GPT-4 on its inability to solve the symbol grounding problem

Jason Resch jasonresch at gmail.com
Sat Apr 8 21:42:07 UTC 2023

On Sat, Apr 8, 2023, 3:54 PM Brent Allsop via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> Great, thanks so much for this!!
> Yes, Gordon, and Adrian are getting it, even extending the idea to other
> implications, like things may appear to be grey in dark light, even though
> they can know it is still red.
> And Jason understands the fundamental issue, he just has some remaining
> questions.
> The fundamental idea is that the redness you experience is just a
> physical fact.

What is the reason for including the word physical here before fact? I
could understand if you said it is just a fact, or a brute fact, or an
unexplained fact, or even an unexplainable fact. But I by saying it's a
"physical fact" adds to my confusion.

Redness is a property of your knowledge of a strawberry.

I agree with this. In that the manner in which a quale feels depends on the
structure and potential of one's mind to accept and relate different sets
of knowledge.

And properties can represent information.

It's not very clear to me what this sentence means, but I would say
redness, like all qualia, are informative.

> Jason was asking about the dictionary.
> This has to do with the way we represent information in a substrate
> independent way.

Information can be encoded in many ways I agree: groves in a record, holes
in cardboard, magnetic tape, optical disks, magnetized zones on a spinning
plate, charges in a flash drive, etc.

For example, anything that is not a redness property can still represent
> redness, if you have a dictionary.

I would say any information channel can be used to inform a system in a way
that it would interpret to mean red. Consciousness requires not only
information, but a system to be informed. The information can take any
form, and if the computational theory of mind is correct, then the system
to be informed can also take any form (due to the substrate independence of
information processing systems): electro mechanical, relay switches, vacuum
tubes, integrated circuits, photonic circuits, ropes and pulleys, marbles
and wood, water and pipes, etc.

+5 volts can be thought of as representing red, a punch in a paper tape can
> be thought of as representing red.  Even a greenness property can be
> thought of as representing red, if you have a dictionary that tells you
> this.  But the transducing system that detects the +5 volts operates as the
> dictionary, and it sends the correct meaning to whatever next property will
> be representing that same 1 not 0 or red not green meaning.

I can agree with this. (If I interpret it correctly that you you are only
pointing to the substrate independence of information.)

> A transducer can interpret +5 volts to represent 1.  Then a series of
> transducers detecting voltages on wires can represent a string of 1s and 0s.
> A dictionary can specify that a particular string of 1s and 0s represents
> the letters "Red."
> A further dictionary can say that the referent of the word 'Red' is the
> quality your brain uses to represent red things with.
> On Sat, Apr 8, 2023 at 1:02 PM Jason Resch via extropy-chat <
> extropy-chat at lists.extropy.org> wrote:
>> My questions related to the diagram:
>> 1. Why is the strawberry gray, is it supposed to be gray to signal that
>> photons are colorless? Are all three seeing the same strawberry?
> They are all looking at the same strawberry.  The fact that it is grey is
> because all we know of the object is that it reflects 700 nm light.  But
> this abstract description of that behavior of the strawberry tells us
> nothing about what it is like.  It is something in your brain that has your
> redness property, not the strawberry reflecting 700 nm light.  The
> strawberry just seems to be red, because our brain falsely colors our
> conscious knowledge of it to be red, to tell us that is the one we want to
> pick.  Nobody knows the colorness quality of anything out there.  All we
> know are the false colors things seem to be.  All we need to do to know the
> true colors of things, is discover which of all our descriptions of stuff
> in the brain, is a description of redness, so we will then have the
> required dictionary.

I might suggest an improved visual then:
Showing in one scene all three surrounding the same gray strawberry, having
three lines, with each going from the strawberry to each of their eyes with
a label 700nm light, and have the human who sees green be wearing some
funky looking glasses/electronics.

You can ask Dalle-2 or Midjourney to draw this for you.

> 2. The image file is called "functionally equal machines", but how are
>> they functionally equal when they each have a different mental state from
>> the same stimulus?
> They all know the strawberry reflects 700 nm light.  And they will all
> tell you the strawberry is red.  And they can all be equally intelligent.
> They just represent their information in different ways.

You might then also add a speech bubble, to clarify that all three are
saying "I see a red strawberry" (assuming the second person is meant to, as
if they have been wearing the inverter since birth).

>> 3. Why is the same person seeing a green strawberry? Is it meant be the
>> same person or a different person with inverted qualia?
> The only difference between the first two, is the second one has a red /
> green signal inverter between his retina and the optic nerve.  This changes
> the dictionary of which property it uses to represent red strawberries
> with.  The first one represents red knowledge with redness, the second one,
> because of the inverted dictionary, represents red knowledge with greenness.

Okay that makes sense. I think then adding a speech bubble would help, as
would some indication of the person having a signal inverter in their brain.

>> 4. What do you mean by a dictionary conveying the meaning of red?
>> Dictionaries say nothing of the quale of red. They can only refer to things
>> that look red, but we have no proof people even see colors the same as each
>> other.
> Let me know if what I've said above doesn't answer this question.

I think I understand now, but I think dictionary is a word apt to cause
confusion. It might be clearer to say: within each mind there is a process
able to convert a stimulus to a particular
perception/quale/experience/mental state (whatever you think is the best
term). And then you can say different minds may have different mappings
between stimuli and mental states such that the same stimulus can lead two
different minds to have entirely different
perceptions/qualia/experiences/mental states.

> And finally, there is more to it than just there are different ways to
> represent information (1: directly on physical properties, or 2: in a
> substrate independent way, which requires a dictionary)

This makes zero sense to me. How do you know physics isn't at the bottom,
nothing other than information? It seems to me you think different physical
objects somehow directly relate to qualia. But this is disproven by
Chalmers dancing qualia argument applied to two switchable but functionally
equivalent neural circuits made from different material substrates. No
difference in experience can be reported or noticed, even if switching and
inverting color qualia every second. A qualia change that can't be noticed
is a contradiction in terms (in my view). I think if you see that the
substrate independence of information does not end at the optic nerve but
follows all the way through to all the nerves of the entire brain, you will
see my point.

> There are also different ways of doing computation.  First, there is
> computing directly on physical qualities, using some kind of waves
> <http://www.izhikevich.org/publications/pwc.pdf>, or quantum entanglement
> that computationally binds the qualities into one composite qualitative
> computational experience.  Second, there is computing with abstract binary
> 1s, and 0s (it doesn't matter what physical properties are representing the
> 1s or the 0s because you always have a transducing dictionary which tells
> you which is which.

If you are saying there are many ways to build computers and the substrate
makes no difference to what they can compute, then we agree.

  And you computationally bind groups of 1s and 0s in a CPU where registers
> can be computationally bound with discrete logic gates.  Both of these
> systems can do computation, and both can be intelligent.  But what they are
> like, and the way they compute are very different.
> One is phenomenally conscious, and the other is, though equally
> intelligent, abstractly so.

Why can't both be phenomenally conscious?

Could you please:
1. Define consciousness
2. Define phenomenal consciousness
3. Define abstract consciousness

This would help immensely.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230408/85853811/attachment.htm>

More information about the extropy-chat mailing list