[ExI] Symbol Grounding

[Giovanni Santostasi]

*Actually, you went further than that, arguing that linguistics is not even
the correct discipline.  But you were supposedly refuting my recent
argument which is entirely about what linguistics — the science of language
— can inform us about language models.*

We are trying to tell Gordon and Brent that to understand something as
alien as AIs mind we need a multidisciplinary approach. Gordon doesn't care
because he wants to fixate on some obsolete science such as linguistics
that at most is based on some anthropological knowledge of how language is
used by humans. Linguistics is worse than psychology because at least some
psychology is based on real experiments and relies on scientific data.
Linguistics is more philosophy than science.

Gordon relies on Linguists for his arguments simply because he found
something in it that "CLEARLY" distinguish humans from machines, this damn
business of grounding. He repeats like a mantra the linguistics positions
that meaning is derived from grounding. There is no scientific evidence
that this is the case, no experiment, and no multidisciplinary analysis of
this process. Like almost anything in philosophy is somebody's opinion that
somehow was adopted by a bunch of followers and become dogma.

If one wants to understand the phenomenon of language not just in humans
but in other animals and in fact even artificial systems then one needs a
multidisciplinary effort. For sure, neuroscience is one of the most
relevant sciences to this deeper understanding. What neuroscience has to
say on this topic is not just "interesting" but fundamental. Neuroscience
is not as precise as physics but it is science and its understanding of the
world is based on scientific evidence. We do need a new science of
linguistics that is not just some philosophical, arbitrary musing.
One can study the production of meaning in the human brain using
neuroscience and giving arguments based on what we know about information
processing in the brain, sensory perception, and so on. Using this approach
then we can make parallels with what machines do, given we tried to
recreate the same architecture of the brain via the neural networks.

In fact, if anything experiments we do with LLMs will show more and more
how inadequate current linguistics is. I think a lot of resistance from
linguists like Bender comes exactly from this, they feel threatened by
these models of language because they force us to reevaluate what language
is.
So Gordon ignoring neuroscience and any evidence presented to him (like my
experiments testing cognitive abilities of GPT-4) just shows how for him is
just ideology at this point and how he found something to ground himself to
justify the cognitive dissonance he is dealing with (that machines indeed
can understand, including GPT-4, and actually be conscious).

Giovanni












On Wed, Apr 26, 2023 at 1:03 PM efc--- via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

>
> On Wed, 26 Apr 2023, Gordon Swobe via extropy-chat wrote:
>
> > Computers have no human brains, or sense organs for that matter which
> are required for symbols to be grounded to be their referents,
> > and so the question remains how a language model running on a digital
> computer could possibly know the meanings of words in the
> > corpus. But you say I am the one dodging the question.
>
> I would suggest that the grounding is transitive. By the fact that humans,
> and
> human generated data, who have these organs and groundings, have created
> the AI:s, the AI:s have inherited the groundings.
>
> When the AI:s are then hooked up with senses, these inherited groundings
> will then be double checked and possibly revised if needed.
>
> Best regards,
> Daniel
>
>
> _______________________________________________
> 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/20230426/39ba20c3/attachment.htm>