On Fri, 18 May 2001, Henry Swanepoel pressed some keys and this came out:
> Our idea was to use some form of neural network encoded into the driver
> (as native methods) that could be hooked on to NPC objects in the DB
> with suitable values given to each to specify it's behavioural class.
> Because of the complexity, the neural net would have to be made very
> modular to allow it to be used most effectively in the DB environment.
I don't believe a neural network alone will be very intelligent - the
self-organisation involved is far too "anything goes" to be very meaningful. It
*is* possible to build a intelligence from an NN, but to have it emerge on its
own is terribly unlikely - probably impossible with most/all training methods
used.
This isn't to tell you not to do it - just to give you some advice from
someone who's been there. It may save some headaches.
> We have, however, not yet passed the initial idea stage yet, as there
> are other smaller issues that I've been trying to sort out first, hoping
> to get the bulk of the world running before we actually start mucking
> around with the driver and the hassle with developing this network
> (which I have no experience with - I will have to learn very fast. ;)
http://www.dontveter.com/bpr/bpr.html - this should help you a lot.
> Languages shouldn't be a problem. English is technically my second
> language, but I normally don't have a hassle with that. As long as we
> can correspond in English (or Afrikaans ;)...
Heh, yeah. It's the creatures themselves that will likely develop their
own language - unless, perhaps, if humans teach them to speak (I didn't realize
that possibility) - but the subject matter must be consistent with the VR -
talking about cars in a medieval environment just won't do - *unless* it can be
related to the environment somehow ("well, a car is *like* a horse and wagon -
without the wagon...").
--
Steve
|