Re: [MLUG - DISCUSSION] Re: [MLUG] AI - any suggestions?

["Igor Izyumin Jr." <EMAIL:PROTECTED>]

> the system you describe also has one other basic weakness - it'll fail
> miserably on simple tasks of the type:
> - given the fact that I have a truck, 4x4, a bicycle and a horse, how
> many cars do i have?
These are not simple tasks at all.  These are reasoning skills.  Therefore, 
this is really intellect that you are talking about.  But the whole thing is 
like this; if you want to process REAL language (not a subset of it, like the 
old Sierra adventure games did, where they basically looked for specific 
words) but meaningful language, of the type you find on this mailing list, 
for example, it is imperative that you have natural intellect.  If you look 
closely, many things in our language are implied, based on the fact that 
humans will throw out something that does not fit the rest of the things in 
their memory.  Basically, if I say that thing about the cars, the brain does 
some analysis and figures it out.  Following is my idea of how it does it.
"Given the fact" -> standard phrase that means that an idea follows
"I have" -> goes to memory, since it is irrelevant at this point
"truck, 4x4, bicycle, and a horse" -> goes to memory, since it is data
"how many" -> probably a question, means to count the objects in some data set
"cars" -> narrows down to what kind of objects
"do I have" -> connects to data set that was stored in memory
Now begins the analysis part.  The brain looks at the data set, and analyzes 
words, connects them to abstract ideas, which connect to images, feelings, 
whatever.  For example, "horse" is immediately excluded because it is 
connected to the image of a living animal and not a mechanical human-made 
chunk of metal that is a "car".  The bicycle is also thrown out, because a 
"car" is associated with something that's self-propelled, not human-powered.  
The truck passes the test, because it is associated with cars, even though 
it's not strictly a car.  The 4x4 is the hardest part, because it does not 
really mean much: it could be a 4x4 piece of wood or a 4x4 room or a number 
of other things.  However, the brain analyzes what the data set is composed 
of, until it finds a class to which all of the items belong.  Then, it finds 
that a 4x4 truck is the most likely idea that the "4x4" word is associated 
with.  Thus, it finds that the only two objects that fit the class of cars 
are the "4x4" and the "truck".  Then it counts the number of the things in 
the data set and finds out that it is two.
As you can see, it is VERY, VERY complex.  You have to have a complex neural 
net for that, and the most complex neural nets that exist now do not have 
anything close to this capability.  It requires a HUGE amount of memory, a 
HUGE amount of processing power, and a system to input feelings, perceptions, 
etc, as well as a system that finds similarities between data stored, as well 
as a system of garbage collection and compression by throwing away the 
specific details and referencing them to other ideas instead of storing a 
duplicate.  Then, the system would need to be fed an enourmous amount of 
information (just think how much data a human accumulates and stores by the 
age of 2 and how little the human can do intellectually at that point).  I 
would be surprised if I saw somebody create a system remotely close to this 
within the next 50 years.  And Michael appears to think that it can be done 
with a few perl scripts.
-- 
-- Igor
--
To unsubscribe, go to http://mlug.missouri.edu/members/edit.php

Archives are available at http://mlug.missouri.edu/list-archives/