[Fis] Fis Digest, Vol 105, Issue 12 Human Wisdom vs Meta-Intelligence

[Eric Werner]

Dear Lou,

Have been meaning to respond but was busy programming (when I get the 
mood it can be fun) and outings with family.

Consider notion of meta-AI ( defined as not only being to solve problems 
created by someone else, but to create new definitions for new concepts, 
to question and come up with problems that lower level AI can solve).

Initially I had a similar view as you that rational thought, deductive 
inference would be need to be implemented in other complimentary modules 
with their own logic unlike an LLM - a parrot that gives the illusion of 
thought and understanding.

But I am no longer so sure.

Note, the way you appear to define creative mathematics, there is still 
a human in the loop in that defines the rules and search space which are 
then given to the rational system (computer).

However, although we are the designers of the boundaries, frame of 
reference, possibility spaces and talking points for now, LLMs keep 
gaining competencies.

Once  a meta-AI starts designing the search space and the rules, then 
meta-AI systems will perhaps surpass humans in the process of generating 
mathematical and scientific theories.

The problem will then be how to get these systems to explain what they 
have created. If for example the meta-AI designs a system based on its 
new theory and it works, then we may be incapable of understanding how 
it works due to are limited mental capacities. Like us trying to explain 
to a baboon the design of the von Neumann computer architecture.

Again distinguishing between implementation-instantiation of a system 
versus its functional architecture, the LLM or some form of it with a 
meta-LLM  that is rationally equivalent or superior to human reasoning 
and creative thought may be possible. Like your reference to number 
theoretic limits, it  may be an empirical question that cannot proven or 
intuited by deductive formal methods let alone a priori.

Imagine an Artificial Spencer Brown (ASB) thought engine, surely you can 
imagine that.😉

Best,

Eric

P.S, I do not mind being tricked into arguing with myself, as such  
Socratic methods or Hegelian Dialectics leads to new approaches quickly 
and meta-rationally. But I would much rather argue with you over a beer 
or glass of wine. Then you may be wondering who tricked whom.


> (I am sending this again with message body made small.)
>
> Dear Eric,
> The boundary between our dreams and our actualities is vague.
> We do not yet actually have AI.
> And when we get it, it will not longer be artificial.
> AI = ~ AI.
> The present LLM’s are nowhere near doing creative mathematics.
> It is not enough to mimic rationality to do creative mathematics.
> When the rules are all given and a search space is specified,
>   then computers can look for and find mathematical proofs that humans would not find without them.
> This has been done and it will be done spectacularly in the future.
> This will be exciting but we (the mathematicians) are designers of these games.
> We will always be happy to see the machines go forward into more and more possibilities.
>
> The key concepts here are comprescence and coalescence.
> As we work with technologies we are no longer alongside them, we are coalesced with them.
> I use my glasses by putting them on and becoming the world view that happens in SEEING THROUGH them.
> And then “I” have lost “my” objectivity.
> It was never mine.
> Best,
> Lou
> P.S. Please note that I write in such a way that it is tempting to imagine arguing with my point of view.
> But I do not have the point of view. You have the point of view. And when you argue with “me” you are arguing with yourself.
> My intent is to write down points of view until they become absurd and turn into other points of view.
> I trick you into participating, but you should know that I am doing this.
> You trick me into responding.
> Knowing will accelerate the process.
>
>
>
-- 
/Dr. Eric Werner
Oxford Advanced Research Foundation
https://urldefense.com/v3/__https://oarf.org__;!!D9dNQwwGXtA!RFyN7hk0AlV9HeO8eEdZLAxEABvEOkEAaKm_0gyN88qHOtckzvh_1mFRw2mOP7PhNet2zRuCwRHNtlMhAXtVRB0$ 


/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listas.unizar.es/pipermail/fis/attachments/20231109/c79d0f3f/attachment.html>