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

Jason Resch jasonresch at gmail.com
Fri Apr 7 12:47:38 UTC 2023


On Fri, Apr 7, 2023 at 1:33 AM Gordon Swobe <gordon.swobe at gmail.com> wrote:

> On Fri, Apr 7, 2023 at 12:16 AM Jason Resch via extropy-chat <
> extropy-chat at lists.extropy.org> wrote:
>
> I believe that if GPT relaly believes it is not conscious, then it must be
> conscious, as one has to be conscious in order to believe anything.
> Likewise one has to be conscious to know. You said it "knows how it was
> itself designed". You also said that GPT "understands" AI. To me, knowing,
> understanding, and believing all imply consciousness, just as much as
> feeling, perceiving, and thinking do.
>
> As I wrote in this thread a couple of days ago, I need to remember to put
> those terms in scare quotes lest I appear to be making the same mistake I
> am railing against. I do not believe that GPT actually believes or knows or
> understands anything whatsoever, but "believes" and "knows" and
> "understands" are convenient shorthand.
>

How is "belief" different from 'belief'?
How is "know" different from 'know'?
How is "understand" different from 'understand'?

If there is no difference that makes a difference, they are identical. The
use of square quotes then, is merely an escape to say it is acting as if it
understands without understanding, or it is acting like it knows without
knowing. But there are simple tests for knowledge and understanding (e.g.
asking questions). To only be able to "know" the capital of France is
Paris, and to say that this is somehow different *really* knowing the
capital of France is Paris, is as troubling as saying something only
"multiplies two numbers" it isn't *really multiplying two numbers*, it is
only giving us the appearance that it does, when it returns the correct
product for any two multiplicands we supply it.

The idea of using square quotes for these terms reminds me of how Daniel
Dennet described as p-zombies with second-order beliefs, using words like
"thinkZ" to describe zombie thinking-without-really thinking:

"Zimboes thinkZ they are conscious, thinkZ they have qualia, thinkZ they
suffer pains – they are just 'wrong' (according to this lamentable
tradition), in ways that neither they nor we could ever discover!"


I find it a retreat -- a way of using language to describe what we plainly
see, while remaining in denial of the implications of the observed behavior.

Do you believe that a process can only "multiply two numbers" without
really multiplying two numbers?

Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230407/a635d540/attachment.htm>


More information about the extropy-chat mailing list