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

Ben Zaiboc ben at zaiboc.net
Tue Apr 18 12:28:10 UTC 2023


On 18/04/2023 12:26, Brent Allsop wrote:

 > Oh, so a picture is composed of all the properties of the electrons 
in all the wires, representing all the pixels of a picture.

No.

The properties of the electrons, such as spin, mass, charge, etc., 
aren't really relevant, except as they relate to the ability of the 
electrons to carry bits of information.

They don't have to be electrons, they could be pulses of pressure in 
tiny pipes filled with gas, they could be protons, or magnetic signals, 
or electromagnetic waves or vibrations in a beer-cans-and-string 
computer, etc. All of those things will have different properties, but 
they can all carry information, which is the important thing.

The picture is built up of the information carried by the electrons (or 
whatever), assembled into an information structure (a model) by whatever 
system is capable of doing so. 'a picture' is a totally abstract thing, 
which can be represented by any number of arrangements of matter like 
numbers in a table, beads on string, packets of charge in an array of 
capacitors, tiny LEDs on a screen, waves of depolarisation in a bundle 
of axons (arranged into spike trains), and so on.

Even pixels aren't necessary, and aren't the lowest-level components of 
pictures. Pixels are just a part of the way we normally create pictures 
in computer systems. No pixels in a painting. You could, if you wanted, 
call the individual rods and cones in the retina 'pixels', but that's 
not really what they are.

It's all about information. Everything else is just the embodiment of 
the information, and is infinitely variable.

 > These "models in our heads" are made of real world things, which have 
qualities we directly apprehend in an infallible way, via the 
computational binding.

They have to be /embodied/ in something, but it doesn't really matter 
what, as long as the system works. They are /made of/ information, not  
'things' (unless you want to call an information pattern a thing, in 
which case they are a thing). I don't know what you mean by 'qualities 
we directly apprehend', but it sounds distinctly dodgy to me. Literally 
speaking, nothing can be directly apprehended, and as I've said above, 
the qualities of the things used to embody the information hardly 
matter. Again, it's all about information.

And nothing is infallible!

Ben
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230418/0e8e30c1/attachment.htm>


More information about the extropy-chat mailing list