Re: it's not just the box dummy - it's just a house of cards. -- MIDRANGE-L

Alexei Pytel wrote

> 
> Attempt to emulate how human brain works is quite exciting. The only
> problem is - do you REALLY want computer to behave like human.
> Humans are subjected to mistakes, misjudgement etc. Similarly natural
> language is subjected to misinterpretation.
> It's a long story, so I would not go deeper.
> The only question I would like to ask - do you want your intelligent
> computer to wipe out your savings account just because it "thinks" it's
> worth doing.


Alexei 

Yes it is exciting, but I am not trying to clone the brain - nature
achieves this very well! 

My ambition is more down to earth - to find a much simpler and more
productive way of creating and maintaining applications, using some ideas
that I have developed from the study of one part of the functionality of
the brain, rather than by studyijng its mechanisms.

I am not proposing that at this stage we can create applications with such
random thoughts.
If I had a savings account and this happened, I would explode!

> 
> Let alone the issue of maintenance of "integrated data and rules"
> database! Since it will evolve with time, very soon nobody will be able
> to understand what's in it. We regularly complain about difficulties of
> fixing large undocumented programs. These will be dwarfed by a task of
> finding a flaw in a terabyte data-and-rules database.
> And what about recovery if smth goes wrong?

Actually people can understand it because it is written in their own
terminology and language and every part is always in its context. If I look
at a small program subroutine, I might understand each instruction, but
unless it comes with considerable documentation, I will not know what it
does because it has no context.  If someone gives me the context, i.e. the
program that calls it, I might have a better chance, but this may be
written in a program language that I do not understand, and probably
lacking in documentation but even if, once again, I do understand every
instruction, I still do not understand at that point what the files are,
what other programs create, update, read and report on those files, etc.,
and how the one program fits into the whole system.  In a Neural Database,
you can only get to a particular part through the context, and you can
never look at it in isolation, and therefore never out of context.  The
neural database is therefore self documenting.  

The whole purpose of the Neural Database is to simplify the whole process
of development and maintenance dramatically - not to complicate it.  It
really achieves it.  I have had people create simple applications in half
an hour with no training and not even a demo of the product, yet their work
was automatically integrated with all previous definitions.

To my surprise, I have found that the rules part of the Neural database is
much more compact than thousands of programs and thousands of files.  We
are talking about only relatively low megabytes for quite large problem
domains.  Although ruels and data are integrated, you can look at the rules
in isolation.  Flaws are very easy to find - much easier than debugging
code.  With the Neural database, we only have to be sure that each small
part is correct.  We do not need to try and consider the whole, because the
Neural Database provides its own integration, so therefore, if the parts
are correct, we can be sure that the whole, which may be too complex for us
to understand as a whole, is also correct.  If you journal and use
commitment control, then you do not have recovery problems.  The Neural
Database can be spread over multiple files if you wish.  We use 8 for
operational reasons, although these all have identical structures.  The
last time I had to restore any data was in /38 days when a disk went.  I
have never had to restore a file because of data corruption

> 
> I've seen a lot of ideas in my life which were beautiful while in a lab,
> but turned horrible when applied to smth more practical.
> This made me somewhat cautious.


I entirely sympathise but this has been used outside a lab and it really
works.  The short learning curve means that you can test the concept
without taking a lot of time and with minimal risk

If you like a paper on Connectionism (not a product description), the
theory behind it all, please ask

Best wishes,

Rob Dixon

+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---