[Fis] Theory of Meaning -> General Intell. -> Super Intell. (HELP!)

[Marcus Abundis]

Karl –
Again, thank you for your note.

Gordana –
Yes, I can understand your surprise in that I have not explored ChatGPT. My
foremost reason is that I am skeptical of statistical views of
‘intelligence’ – that is not how Nature works, instead requiring functional
analysis. Also, in reviewing papers for ICLR 2025, I was offered a ‘trial’
ChatGPT bot for improving paper reviews. My impression was the ‘bot’ simply
added a few flattering comments on my ‘good insights’, and then fed back my
own views, with differently arranged but similar terms. It did inspire me
to alter my review a bit to make it more ‘friendly/encouraging’ – but the
essence was unchanged. Thus, I found it somewhat useful, but not radically
so.

Also, I just now tried using ChatGPT’s writing: siribot.cn, Academic
Assistant Pro – I gave it my paper’s abstract to see what improvements it
might offer. The result was disappointing and did not even reach the level
of utility I saw for the bot I used for my ICLR review. It significantly
obscured the abstract, with what might be considered a ‘traditional
academic’ framing. Still, I will explore a bit more before abandoning
ChatGPT wholly.

My overall impression of statistical models of intelligence is that they
can present countless opportunities – say for example ‘flying monkeys’ –
but no matter how many times a ‘bot’ presents/writes/draws a ‘flying
monkey’ it will never produce a ‘flying monkey’. The bot cannot detail the
ontology, epistemology, genomics, etc. of ‘flying monkeys’, such that we
ever have any actual ‘intelligence’ about flying monkeys.

I believe such ‘statistical black-box bots’ merely extend Shannon’s
‘statistical surprise’, moving from entropically recombined discrete
characters, onto recombined words, phrases, sentences, (growing in
complexity), etc. until outputs finally resemble something looking like
‘intelligence’. The varied statistical (proprietary) filters one develops
and uses along this extended statistical chain is the only thing making
this possible, but that is not *actual* intelligence.

Jason – on your thought experiment note. This problem of useful ‘thought
experiments’ is the crux of the matter. For example, how does one from 200
years ago detail today’s experience of ‘modern air travel’ and all that it
entails – akin to Arthur C. Clark’s “sufficiently advanced technology
indistinguishable from magic”? What you ask for goes beyond implied ‘simple
tools’. Looking backward we easily see this (innovation is plainly
evident), but looking forward is entirely different and defies simple
description.

Karssimir – no, I was not suggesting collaboration, but suggesting your
‘definition’ could benefit from critical review here on FIS. My own
interests long-ago moved onto deeper issues of intelligence (recent AI
workshop submissions), beyond what you examine. For example, Howard’s note
on emergence points to just one missing aspect one needs to cover in a
comprehensive definition of information.

Marcus
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listas.unizar.es/pipermail/fis/attachments/20250311/28eecccd/attachment.html>