Re: AI-Mind in Forth
- From: Jean-François Michaud <cometaj@xxxxxxxxxxx>
- Date: Mon, 18 Jun 2007 20:41:50 -0700
On Jun 18, 10:10 am, Frank <fjru...@xxxxxxxxx> wrote:
ATM,
Got a question for you regarding the logic on the AI generating a
thought. Correct me if I am wrong. An active noun is selected as the
subject. The sequel (verb) to the noun is now activated. The most
active verb is selected. The squeal to the verb (noun) is activated
and the most active noun is selected for the object of the verb.
If this is correct then the following is a possible:
Man eat food (Human input)
cat eat fish (Human input)
cat eat mouse (Human input)
mouse eat garbage (Human input)
man eat mouse ----- AI generated thought
man eat garbage ----- AI generated thought
Another problem I see is when a noun is used on both sides of a verb.
You get strange thoughts generated.
Idea is thought
thought is concept
man is father
generates --- thought is father & man is concept --- by the AI
Your comments and thoughts?
Frank
I don't personally think that useful AI can be achieved by selecting
active words in a list of words, no matter how those words become
active.
It seems that things are somewhat more complex; this model seems to me
to be fundamentally incomplete.
Modeled after the human brain (given the outstanding ability that
human beings have), neural networks offer the most potential, but they
still don't come out as a clear and effective method like one would
hope for because it is not yet well understood how the human brain
wires itself to structure information through perceptions and
language, current neural network models are not as useful as they
should be because they are not structurally complete.
One thing is certain though, language is not the first generator of
thoughts and so a method based on such an approach can not work.
Words are not the "stuff" of thoughts. Thinking is a fundamentally non
verbal process.
A 6 layers neocortex can approximate any computable function (this is
why we can learn to perform any task as well as we decide to).
What I feel is a more interresting take on AI is creating an AI based
on an evolving evaluation function (multiple evolving evaluation
functions; discrimination/rough categorization in multi-dimentional
space.
The human mind roughly discriminates information into "categories" as
much as possible; no verbalization is necessary for this process to
occur. Although information can be structured in as many ways as can
be imagined, each mind will have a tendency to *structure* information
in a very unique and very specific way. The boundaries of categories
generally won't be clearly defined, as ours aren't, even though we
like to pretend for ourselves and to others that they are.
Time (sequence) and space (localization) seem to be critical
components to human perception and acquisition/organization of
knowledge in 3 dimentional space and I think that those concepts
should be incorporated in the melting pot.
Simply, thoughts, in the mind of an AI, could be generated by
observing it's own internal model of what goes on in the world it's
exposed to, just as we do. Thoughts emerge in the system as an ability
to roughly estimate how event's unfold given raw sense data; or as the
mind is more advanced, given other thoughts.
The evaluation function needs to be exposed to massive amounts of
information through senses and needs to be able to interact with it's
environment so that it can converge towards something meaningful in
the environment in which it must evolve.
My 2 cents ;-).
Regards
Jean-Francois Michaud
.
- Follow-Ups:
- Re: AI-Mind in Forth
- From: pablo reda
- Re: AI-Mind in Forth
- References:
- AI-Mind in Forth
- From: Frank
- AI-Mind in Forth
- Prev by Date: Re: FORTH Shboom Manual WTB
- Next by Date: Re: AI-Mind in Forth
- Previous by thread: AI-Mind in Forth
- Next by thread: Re: AI-Mind in Forth
- Index(es):
Relevant Pages
|
Loading