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

Giovanni Santostasi gsantostasi at gmail.com
Wed Apr 19 02:04:44 UTC 2023

```Brent,
Please reply to all so the others on the list can see our communication.

>
>
> *1) If you had the type of description of Giovanni's redness you look for,
> do you expect to see my redness yourself? Please answer yes or not and
> elaborate. *
>

*I tried to answer this, let me answer this specific question.  Once they
know what your redness is, then they would start showing me different
colors, until they see Giovani's redness in my brain.  If they show me a
red patch, and they see Giovani's redness, then YES, we will know that my
redness is the same as yours.  But if they see Giovani's redness, when they
show me a patch of green, they will know that my redness is Giovani's
greenness.*
This is where we are butting heads, literally and figuratively.

We cannot do that. How can we find out what Giovanni's redness is? I gave
you a scenario of how to do that and in fact how it is done with similar
experiments. You show many subjects the same stimuli (a red piece of paper
or whatever) and record the response. ONE PERSON IS NOT ENOUGH !!!

I think this is the crux of the matter. The way we do science is
by collecting the statistical properties of a phenomenon. We can only say
what is the response to red in average over a large sample of subjects.
Let's simplify and say there is a particular parameter (it is going to be
more complicated than this in the real experiment) but let's simplify and
say there is a particular parameter, like your glutamate concentration in a
region of the brain or whatever that activates when people are exposed to
an object that reflects the color red. It would look something like this
when I collected enough data:

[image: RednessA1.png]
What this graph represents in the x graph is the concentration of the stuff
that the brain produces when it is exposed to red and the y axis represents
the fraction of people that responded with a given concentration. It is a
range and it is a Gaussian distribution (in reality probably more
complicated like a lognormal distribution but this is not essential right
now).

When I show the color green then I get another distribution (more glutamate
but less serotonin or whatever). Where is Giovanni's response? It is one of
these dots but it is not useful by itself because if I reproduce the same
amount of glutamate in you, you WILL NOT see my red.

This is because each physiology is different even if redness was due to the
concentration of a particular neurotransmitter (that absolutely is not) you
cannot reproduce the same experience by reproducing this particular
physiological condition in somebody else. The graph shows we respond
differently to the same stimuli.

Also from the graph, you see that when shown the red color some people
react like some other people react when they see green. It doesn't mean
that they would see green, seeing green is responding to the stimuli green
by definition.

Let's say somebody's response to red is on the right side of the red curve,
but when they see green their response is also on the right side of the
green curve. If somebody sees green and their response is in the
overlapping region with red they are not seeing red but green, it is their
green. It is a relative uncommon response but it is still red and green for
this person, but it doesn't mean the red of this person is like the
greeness of somebody else. That is nonsense.

What this shows is that "redness" and "greenness" are not precise things
but a range of possible responses. Our understanding of green is in this
distribution, this is how we understand a phenomenon. The distribution is
the phenomenon. If you want to summarize it then you can use an average, or
to be more precise an average and a standard deviation or you can give
other higher stats like the skewness if the distribution is not symmetric
and so on.

How do you use this in practical applications like making a color-blind
person see red given we have only distributions over a population? We can
still do experiments on the individual and see what is his particular
response, if it is away from the average we can try to add or reduce
glutamate (again redness has nothing to do with glutamate and it is not due
to a single physical factor but I'm going along with your suggested cause
of redness to simplify things) and see how the individual responds. This is
how we design drugs, for example, we design them on the average response of
a population and then hope the individuals will respond within a range.
I feel that your entire idea to reproduce a person's experience of redness
in another is based on a deep misunderstanding of how science works. In
general, scientific knowledge is based on a large N of subjects or trials.
Bigger the N, the better.

Now there is some space for what is called N of 1 which means following
what an individual response is to something like a drug, a diet, stimuli,
or whatever. This could be done by tracking the person in time and seeing
how their physiology reacts as a function of time to a drug or maybe dosage
or something else. But still whatever you conclude will be very particular
to this person and cannot be generalized. It would be able to be used to
reproduce the same response in another individual.

Really scientific knowledge is statistical in nature. Even things like
electrons are not based on the study of a single electron but billions. The
same is true for anything regarding the brain because we usually study the
activity of billions of neurons at once when we fMRI or EEG.

Please look at the graph above and see why this example shows that your
idea of how science can understand redness doesn't make any sense at all.

Giovanni

On Tue, Apr 18, 2023 at 6:12 PM Brent Allsop <brent.allsop at gmail.com> wrote:

>
>
> On Tue, Apr 18, 2023 at 6:41 PM Giovanni Santostasi <gsantostasi at gmail.com>
> wrote:
>
>>
>> *RQT is predicting that as soon as experimentalists and neuroscientists
>> start seriously observing the brain in a non property blind way*I have
>> no idea what it means and again you are using randomization of words you
>> used before, that doesn't help with communication.
>>
>  Sorry, I'm trying to adjust my terminology to you, the target audience.
> I believe you indicated you prefer the term "property" to "quality" so I'm
> Using that.
>
>> 1) If you had the type of description of Giovanni's redness you look for,
>> do you expect to see my redness yourself? Please answer yes or not and
>> elaborate.
>>
> I tried to answer this, let me answer this specific question.  Once they
> know what your redness is, then they would start showing me different
> colors, until they see Giovani's redness in my brain.  If they show me a
> red patch, and they see Giovani's redness, then YES, we will know that my
> redness is the same as yours.  But if they see Giovani's redness, when they
> show me a patch of green, they will know that my redness is Giovani's
> greenness.
>
>
>
>> 2) Do you understand that current science is based on modeling, and
>> models are always not 1 to 1 with the phenomenon they try to model? And do
>> you understand that is done on purpose because that is what modeling is all
>>
>
> I tried to prove to you I understand this.  The idea is to come up with a
> model which would correspond to people's subjective qualities.  It would
> probably be a one to many model, since there are likely more than just
> glutamate, which results in a redness quality.  The dictionary which
> specifies the true color qualities of things would be a model of subjective
> qualities.
>
>
>> I'm trying to improve communication so I would like to see if there is
>> some common ground to build upon. Please answer my questions because from
>> your answer I can then give a more meaningful reply and also it would help
>> me understand better your position.
>>
>
> Yes, thanks for your patience with this.
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230418/c72ca7a5/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: RednessA1.png
Type: image/png
Size: 45042 bytes
Desc: not available
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230418/c72ca7a5/attachment-0001.png>
```