Knowledge Base Or Bust

.. drclue at drclue.net
Mon Dec 17 17:56:21 EST 2001


Owen Nieuwenhuyse wrote:
> 
> Arthur T. Murray <uj797 at victoria.tc.ca> wrote in message
> news:3c1cda50 at news.victoria.tc.ca...
> > "Owen Nieuwenhuyse" <nieuweo at ezysurf.co.nz> wrote on 16.DEC.2001:
> > > Arthur T. Murray <uj797 at victoria.tc.ca> wrote in message [...]
> >
> > >> Human:  cats eat
> > >> Robot:  CATS EAT BUGS
> > >>         10   41  10
> > >>
> > >> The AI gives the wrong answer in the fourth exchange because the
> > >> concept of "bugs" has retained too high an activation.  [...]
> >
> > > ON:
> > > What do you use these "activation levels" for?
> > ATM:
> > The activation-levels determine which concepts -- by dint of high
> > activation -- will be included in a sentence of thought generated
> > by the interaction of Chomskyan syntax and a conceptual mindcore.
> ON:
> That sounds somewhat neural-net.
> 
> I am more interested in expert-system style processing, where
> sets of rules and axioms (similar to your subject-verb-object statements)
> which are activated by a current condition,
> are executed, returning a result.
> The result is generally a chain of interlocking rules and
> statements that is then pruned to match the input question.


Well, that would be an almost apples and oranges relation.
"expert systems" are not much more than simple selectors
in a hard-wired context.

> 
> You would get a similar effect by sweeping a file or files containing text
> versions
> of statements looking for key terms, then iterating through the returned
> list,
> dynamically linking to subroutines expressing these same statements
> in "program" form.
> The "sweep and iterate" cycle is repeated until results resembling the
> specification for output are found, or until you "time-out".

I guess the crux of the difference is in assigning spikes
in data patterns vs. hardcoding rules.


> 
> The general principle is rather like that used by CYC.
> 
> Are there any useful knowledge bases around that fit with this method?
> 
> I have been working on a sample I/O dialog with some described resolution
> methods. Anyone want to compare notes?
> Are there any examples of highly sophisticated dialogs in use?
> (other than in chat-bots).

I've been involved in several projects where foreign data sets
were automaticly joind to existing databases by routines that
had no prior knowledge of the incomming data.

> 
> I am also looking at weighted qualifications and how they would affect
> contrasting or conflicting arguments for interpretation.
> 
> An example would be:
> birds(80%)(adult)(live) can(80%) fly(by themselves)
> exceptions would be:
> 1) excluding flightless birds.
> 2) excluding injured or restrained birds.
> 3) excluding birds which are too fat-ie large turkeys,chickens.
> 
> Plus, of course, the specified qualifications.
> 
> (this looks a bit like your activation levels)
> 
> Has anyone managed to get a scheme like this to work successfully?

I have made working code of this flavor, but the approach I use
is highly data driven.

-- 
--=<> Dr. Clue (A.K.A. Ian A. Storms) <>=--
--=<[]>=- http://www.drclue.net
--=<[]>=- C++ HTML JavaScript DHTML CGI TCP/IP SQL JAVA VRML NSAPI 
--=<[]>=- http://www.drclue.net/F1.cgi/HTML/HTML.html	(My famous
HTML/CGI guide.)
--=<[]>=- http://www.drclue.net/beta			(My X-BROWSER DHTML library.)




More information about the Neur-sci mailing list