[MUD-Dev] Support for remote NPCs
Joel Kelso
joel at ee.uwa.edu.au
Fri Jul 10 10:55:42 CEST 1998
Nathan F Yospe wrote:
> On Thu, 9 Jul 1998 CJones at aagis.com wrote:
> <snip stuff about scripting/programming NPCs>
>
> This doesn't actually have to be online... and the mud *can* be compiled. I
> *know* I'm the only person here who doesn't have a programming language, as
> opposed to a building framework, for reasons that come down to "programming
> languages are too unsophisticated for my needs", but... well, think of this
> as a case of definition. A programming language needs to be able to define,
> among other things, the state of the world. That's unacceptable.
Well a guess that's true for a _programming_ language. But a "world
constructionlanguage" (or in fact a set of related languages) would be really
handy. A language
for assembling (or defining) bits of the world from components would be all the
more useful if you could define your own structures and embed bits of code in
it where
necessary.
> There's no
> way a programming language can work around this and still be usable, unless
> it is, essentially, an interface definition kit. That's what I have, and as
> it ended up being too unweildy to program with... I redesigned it as a tool
> that can eventually be set up with a GUI and a set of parameters... and the
> effective result of filling out a form, and assembling smaller things. With
> this in mind... if I were to say that this is quite the approach I hope for
> in time, when I'm running the mud on a public machine? Well... the point is
> this: Complex behaviors being evaluated by the mud are a drain, and there's
> no need to keep eight *large* neural nets running at the same time just for
> a single NPC... when that NPC, and a few hundred more like it, can run from
> a remote computer and interface just like a standard client, only without a
> text or graphical interface... in other words, STORE THE AIS ELSEWHERE! The
> critters from Creatures, which I believe Ling just mentioned, show what the
> potential of a neural net based AI can be. (No, I don't do the "chemicals",
> at least not in the AI. I do have instincts and such running on *both* NPCs
> and PCs, though, which amount to chemicals, in terms of effective weighting
> of input data... and the neuron count for a *normal* client is currently at
> about 12,000... the NPC would be even worse. I don't want that on the mud's
> host machine. Put the clients on remote machines... any remote machine... a
> PC here, a university account there... give PCs incentives to install these
> NPC executables and run them... and I end up with a lot more capability for
> simulated NPC inteligence. Add to that the tokenization of language... with
> the fact that the NPC can eventually learn to talk to PCs... and perhaps in
> time some of them will pass Turing tests.
Do you have an intermediate "event language" for events such as sights
andsounds in the MUD, which NPCs can use directly, or do NPCs have to
parse the descriptions that human players get ? If you're not really
interested
in natural languge learning (or MUD description text learning), this might be
a bit of a shortcut.
Joel Kelso
-- joel at ee.uwa.edu.au -------------------------------------------
"Wasn't it a four-year-old this time ? It makes me feel like standing
in the middle of the Atlantic with a megaphone and shouting:
'America! We know you're in there! Put the guns DOWN!'"
- Overheard Conversations
-- http://ciips.ee.uwa.edu.au/~joel ---------------------------------
More information about the mud-dev-archive
mailing list