John Scanlon wrote:
Richard, could you describe your algorithms in a general way (I'm not
asking for any proprietary information), so I could see if they would
fit into my concept of a KBMS?
Imagine a system in which the units of knowledge ("elements") are like
artificial neurons, only more complex and able to have two states:
either dormant, or active. When active they are in a space
("foreground") where they have the ability to see a limited distance
around them. The function of these elements is to capture regularities
in the arrangement of other active elements and/or to cause operations
to occur on other elements. The semantics of an element is ambiguous
between encoding a patterns in lower level elements and triggering the
activation of higher level elements. Element activity is driven by
several processes of relaxation that operate simultaneously ... these
relaxation pressures are grounded in different places: some originate
in the motivation system, some are governed by the attentional focus,
some by the edge of the foreground where sensory input arrives, some by
the output side of the foreground.
There is some structure to the foreground space, determined by feedback
loops and buffers associated with the language processing mechanism:
generally speaking a given area of the foreground will play host to
elements at a particular level of abstraction, and sometimes also to a
particular domain.
The element structure that develops as a result of a given language
input will be governed in a non-deterministic way by the coalition of
relaxation and activation 'forces' that act on those elements. Elements
are not, of course, manipulated by any symbol processor outside of the
elements themselves. Elements do not necessarily stand in a one-one
relationship with the things that we refer to as "concepts" or "symbols".
This description is sketchy because it has to be so brief: it is a
condensation of an architecture derived from the human cognitive system.
It is also a sketch of a _class_ of architectures, since there are
many degrees of freedom not pinned down in the above.
Given that elements develop their structure as a result of their own
learning mechanisms, the idea of tacking a language interface onto such
an architecture would, as far as I can see, be without meaning.
Hope that makes enough sense to answer your question.
Richard Loosemore.
----- Original Message ----- From: "John Scanlon" <[EMAIL PROTECTED]>
To: <agi@v2.listbox.com>
Sent: Monday, November 06, 2006 5:04 PM
Subject: Re: [agi] The concept of a KBMS
Richard Loosemore wrote:
When you say that it "provides ... a general AI shell, within which
any AI algorithms can be experimented with ...", I find myself
exasperated [tho' not to worry, I am exasperated a lot ;-) ] -- it
does not provide a language shell within which *any* of the
algorithms that I work with can be experimented with. There is not
even a shred of relationship between the kind of
'simplified-natural-language' systems and interfaces being discussed
on this list right now, and the entire class of intelligent systems
that I would consider the most viable candidates for cognition.
So, within the confines of certain conventional stuff talked about in
whatever they teach in CS Artificial Intelligence courses these days,
maybe these interfaces would help ... but it would be overstating the
case to say that this includes all AI designs.
I am certainly exasperated with most AI work also. Giving someone
a Ph.D in artificial intelligence is like giving someone a degree in
aeronautical engineering before the Wright brothers.
If you could provide a basic description of your algorithms I might
be able to show how they could be tested within my concept of a KBMS.
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
