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

Jason Resch jasonresch at gmail.com
Sun Apr 16 04:45:28 UTC 2023


On Sat, Apr 15, 2023, 11:02 PM Gordon Swobe via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

>
>
> On Sat, Apr 15, 2023 at 3:04 AM Ben Zaiboc via extropy-chat <
> extropy-chat at lists.extropy.org> wrote:
>
>
>> The *associations* of words are what contain, not just the seeds of, but
>> their actual, meanings.
>>
>
>
> Please explain to me how the association of one word with another or
> others tells me anything about the meaning of any if I know the meaning of
> none.
>

>From the patterns:
Words following a, an, the, are objects/nouns.

Numbers are often listed in a certain sequence (e.g. in ascending order),
often used before nouns (can infer it gives a count). We note that one/1 is
special as we don't see a plural form of nouns used when dealing with one,
but do for multiples. We note the identity and common occurrence between
none and zero. We thereby come to learn the meaning of 'none'

Words that can stand alone as sentences are verbs. Only certain nouns
perform certain verbs, and with certain frequencies. This defines a
particular repitoire of possible and probable actions available to any
noun. We learn which nouns are most capable (e.g. 'people') and which are
least capable (e.g. 'rocks')

Words like 'have' define parts. Words like 'is' define identity or
membership within a broader category. We can begin building a mapping
between all nouns by noting which appear before 'is a' and which appear
after 'is a'. E.g., a cat is a mammal. A mammal is an animal.

We begin building a semantic network of all these relationships, even if at
first we don't know the meanings of any verb or noun, we know the words
associated with what each noun can do, what categories it belongs to, and
what parts they have. Once a few pieces fill in, the meaning of the rest of
the map comes into focus.

We find that when we trace the 'is a' relation ladder all the way to the
top starting from any noun, and we tend to always reach 'thing'. We thus
can conclude the meaning of these words is the most general, nondescript
category of object. So we know what 'thing' means. Likewise if we trace 'is
made of' relationships down, we always end up with 'atom' or 'particle',
and this we could conclude this word is the most elemental thing from which
every 'thing' is made.

Certain prepositions or adjectives connect certain nouns in certain ways,
for example by defining spatial relations: inside, near, above, below, etc.
We'll always see certain words connected by inside but almost never the
other way around: "the man is in the house", but not "the house is in the
man". Each example of a pattern like this provides an inkling of some
relationship and property of the objects involved. When we find numerical
descriptions relating to 'size' we find a pattern that larger objects are
almost never said to be 'inside' objects with a greater numerical size. We
might infer then they inside relates the spatial relation of one object
having a smaller size and located within another. We learn the word 'small'
and 'large' from numerical associations in the descriptions of words like
length width, height, which often appear together. We find that when volume
is mentioned as well it is the the product of these three. So we understand
the meaning of volume as a product of these properties, each of which 'is
a' dimension.

I hope this shows how the patterns in language, if analyzed in detail,
reveal a great deal, and enough to puzzle out the probable meaning of a few
words at first and then each new word learned gives clues to solving all
the other words related to it, until all words can be understood provided
enough example usages.

Jason



> This should be obvious. Look at the word 'wicked' for example.
>>
>
>
> I can infer the meaning of wicked only if it associated with other words
> of which I know the meaning.
>
> -gts
>
>> _______________________________________________
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230416/452ae3f9/attachment.htm>


More information about the extropy-chat mailing list