I thought of a simple example. Although it might not seem very
interesting in the 21st century it still can stand as an introductory
example of conceptual cross-categorization method. I need a word (or
text-fragment) look up table for my project. (A lexicon or dictionary
that related some word to a data object containing a 'concepts'
related to the word. I would like it to be able to efficiently detect
any word (that the program had learned) that contained a particular
character string fragment. For example, I would like it to be able to
quickly find any word that contained some string of one, two or three
letters. By repeating words in the dictionary an alphabetized-like
collection -an ordered collection- can be created that will collect
all the words containing the letter or letters. This would narrow the
search space. So the point is that by moving away from the mind-set of
always trying to find the most efficient way of storing data we can
significantly speed up the search time needed to find data. This is
not a perfect system but I believe it can be used effectively.

I realize that the idea of a trade-off between memory usage and
process time to find data is not new. And a text-based AI program just
looks like a 50 year old concept to some people. but the consensus has
formed that an AGI program will need to use discrete methods, fuzzy
reasoning, deep search, deep search derivatives and deep learning. So
rethinking the old AI strategies to see what we can do in a new
computing environment makes sense in this context.

I will try to work out some more details to show how a somewhat
similar method might be used to detect shapes in visual data.
Jim Bromer


On Tue, Feb 2, 2016 at 2:44 PM, Jim Bromer <[email protected]> wrote:
> I have been unable to make any headway into Logical Satisfiability so
> I have no evidence that the Lord had ever given me any kind of
> direction on the problem. I once thought that there was a very slight
> chance that He had and I then figured that if I was able -against all
> reasonable odds- to make some head way on the problem it could be
> taken as a sublime example of rational evidence in support of
> faith-based belief. I tried what I think are a few novel approaches
> but all of the potential solutions were in np (they were inefficient).
>
> I then decided to work on a simple categorization theory which I
> believe would be useful in discrete AI/AGI. Fundamental categorization
> is a logical problem and I again realized that it is therefore in np..
> This is a very serious problem because it would mean that you would be
> up against the bounds of worse-case computational feasibility with 40
> variables or so. (This is especially true of expected computational
> time feasibility on home computers).
>
> Categorization algorithms can be used to refer to collections of
> references. We can use logical relations about membership in indexes
> of references to other data objects and the logical relationships can
> be used to denote various conceptual relationships. Because a
> fundamental categorization methodology would be in np and since I have
> no expectation of finding an efficient solution to logical
> satisfiability I realized that I would need to use extensive
> cross-references and cross-categorizations if I wanted *a dreamed of*
> AGI program to be able to use cross-categorization efficiently. This
> is an idea I have been working on for years, but I just realized that
> it would not only be useful in an AGI program but it is probably
> necessary as well (for discrete methods). It is a necessary work
> around that I believe can be used to avoid the logical satisfiabiity
> problem.
>
> I will try to find a few simple examples.
> Jim Bromer


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to