"RecilS" The__Hand at hotmail.com wrote in news:comp.ai 18 Jul 2000:
> I'm going to briefly (hopefully) explain my approach
> to computer learning here and then ask for some pointers
> so please, please bear with me.
>> I'm writing this in Visual Basic (Not exactly the best choice
> but I'm more comfortable with that than c++ or anything else)
You might look at the Mind.VB free AI source code on-line at
http://www.virtualentity.com/mind/vb/ -- a port of Mind.Forth AI.
> So far my program maintains an INI file with lists of words,
> their adjectives and things like that.
ATM: It may take a more complex algorithm to conceptualize
the words. See http://www.geocities.com/mentifex/tutorial.html
> The program parses sentences and statements to find 'is',
> 'are', 'like' and similar key words and then parses out
> the noun and it's respective adjective. It stores these
> in the ini file and tells you "I have learned
> that [blank] is [blank]" or whatever is appropriate.
ATM: If the words are not truly conceptualized, then the
expression of the pronoun "I" is meaningless.
> It also checks for conditional arguements (AKA "If grass
> is green then rain must have fallen" would not teach the
> program that grass is green, but that it is green only
> if rain has fallen).
>> Anyway, data gathering is the program's strong point.
> It can learn things in this way very quickly and then
> load it's entire array of data upon loading.
ATM: Maybe it is like the http://www.cycorp.com ontology.
> Now the way I want to use this to involve the program
> in conversation is to have it chat about a topic to
> the extent of it's knowledge and then propose a
> similar topic, much like a real life conversation.
> AKA - If you are talking about sailing, even if you
> never mention water or ocean or anything, the program
> will realize that sailing is done on the ocean
> and ask you what kind of fish you like to eat.
>> I am very confident that I can do this and hence
> eliminate the randomness of computer conversation.
> The program will actually be able to have a continuous
> string of conversation and at the same time, note facts
> for future use.
>> So I really just want opinions.
> Do you think I am approaching this correctly?
ATM: You are making a start (bravo!), but you may need
to develop a much deeper theoretical framework, such as
> I've only talked with my program for about 5 minutes and
> it already has a list of about 50 nouns and their respective
> adjectives and about 20 things that it considers 'good' or 'bad'.
> That's another thing it does, it has a preprogrammed list
> of good and bad and all their synonyms so if you say
> 'grass is bad' it will list grass under the 'BAD'
> section of the ini file along with the bad adjective used.
>> Tell me what you think, THANKS!
Merely listing grass under the 'BAD' section may not be
enough to create manipulable concepts of grass and 'bad'.
You may need to create the software analogs of long nerve
fibers that can conceptualize words by holding all their
incoming and outgoing associative tags at a logical punctum.
> - Doug
>>[ comp.ai is moderated. To submit, just post and be patient, or if ]
>[ that fails mail your article to <comp-ai at moderators.isc.org>, and ]
>[ ask your news administrator to fix the problems with your system. ]