I now have the feeling that a constrained grammar, where new equally
constrained words and grammatical constructions can be introduced, won't
work to produce the base for a good AGI modelling program.  Conceptual
multiplicity (like ambiguity) will manifest itself from the attempt to
refer to characteristics of the internal model that is to be
constructed. Perhaps when combined with the ability to peek into the model
the constrained grammar will help to direct the program to disambiguate.
So now I am thinking about combining these two ideas that I have recently
talked about.  This seems possible.

One problem that I just thought about is that if the constrained grammar is
capable of helping the program to disambiguate a reference then the program
might lose the potential for the flexibility of thought that seems
necessary.  However, the constrained grammar could be used to refer to the
multiple paths that one might take relative to some point of
consideration.  Furthermore, if the entire problem is that the combinations
of multiplicity or ambiguity are too robust to be constrained then there is
no reason to think that the program might not be able to discover that
concepts may be multiplistic just because I want to use a constrained
grammar to disambiguate a statement or an idea.

So I conclude that since multiplicity is (presumably) a function of
intelligence then it does not make sense to try to limit the AGI program so
that it avoids the complications that combinatorial multiplicity can
entail.  However, it is possible, and it does make sense, to use a
parallel IO mechanism that is designed to be highly structured and which
might allow for some disambiguation and structured 'explanation' to assist
in directing the AGI program.

Of course there have been efforts to combine 'labelling' or other forms of
language with IO mediums like vision or robotics and none of these have
been very notable for their effectiveness in producing true AGI.  My method
is different in that it will allow the programmer/teacher to observe the
kinds of 'thoughts' that the program is considering when dealing with a
subject matter, and it will be designed to allow the user to provide much
more detailed directions to the program.  Finally, I am not making the
presumption that by combining simple language with some other kind of IO
media that intelligence will simply emerge.  I am working from the slightly
more insightful position that multiplicity is an pervasive problem when
working with combinatorial 'concepts' in a computer program and this is the
problem that has to be overcome at certain strategic moments in order for
the program to begin to 'think'.

Jim Bromer



-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com

Reply via email to