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

[Gordon Swobe]

On Mon, Apr 17, 2023 at 5:17 PM Ben Zaiboc via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

>
> So, what about 'grounding'? These object models in our brains are really
> the 'things' that we are referring to when we say 'potato' or 'apple'. You
> could say that the words are 'grounded' in the object models. But they are
> in our brains!
>

Yes. I agree, and this is not the first time I have needed to clarify
myself on this point. When I first mentioned referents here on ExI some
weeks ago, I made the mistake of assuming that everyone would know what I
meant, and so my language was not as precise as it ought to have been.
Since then, I have tried to be more precise and to clarify that ultimately
referents exist, as you say, in our brains. I wrote for example that we
have referents in our memories and in our dreams; that even if we dream of
pink unicorns, they are still referents. Optical illusions, hallucinations,
abstract ideas, intuitively known mathematical truths, all these purely
subjective phenomena are referents no less so than the direct perception of
an apple.

With respect to symbol grounding, we need to have some kind of referent in
mind to ground whatever symbol we are reading or writing or speaking or
hearing. Otherwise, it is meaningless gibberish. The problem for the LLM is
that from its perspective, every word in the entire corpus is a bit of
meaningless gibberish. It can do no more than analyze how all the bits of
gibberish relate to one another in terms of patterns and statistics. As it
turns out, with a large enough corpus and enough processing power and some
additional training by humans, this is enough to put on a pretty good show.

-gts
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230417/080ba48c/attachment.htm>