[MUD-Dev] believable NPCs (was Natural Language Generation)

[J C Lawrence]

There's a 10 hour go-to-work gap in the middle of this email, during
which time I forgot much of my original plans the message.  My apologies
for the resulting discontinuity.

On Tue, 25 May 2004 20:24:25 -0400
John Buehler <johnbue at msn.com> wrote:
> J C Lawrence writes:

>> From the player perspective in terms of the actual goals that player
>> can be _seen_ to pursue (social, cultural, in-game, etc), what is the
>> actual function of an NPC?

>> If form follows function, then NPCs should be defined by the value
>> systems they satisfy for players.  Not the value systems we think
>> players have, or the value systems that we think are cute, fuzzy,
>> attractive, neat etc, but the ones that actually generate and
>> maintain player value and interest.

> The actual function of an NPC is to provide a motive reason for doing
> anything in the game.

That defines all NPCs as vendors.

Without attempting a formal taxonomy I tend to view NPCs as filling one
of the following roles (in no particular order):

  1) Cannon fodder.

    This is the NPC as opponent, stepping stone, enemy AI, big boss, etc
    model.  By implication it variously defines either treadmills and
    king-of-the-mountain mechanics.  More critically the cannon fodder
    version of NPCs almost by definition have poor emotional content
    except by external OOB association, and are almost necessarily
    one-dimensional.  The advantage of the model is that cannon fodder
    provide an endless series of well defined and easily discovered
    goals (aka targets) for players (cf Everquest).  As an aside, I've
    failed to find this model interesting for some years now, and in the
    AI forms am always greatly put off by the semi-uncanny valley
    aspects of the AI attempting human qualities when it is so obviously
    mechanical.  I like the prospect of playing with humans.  I don't
    like the prospect of playing with machines.

  2) Lemmings.

    Cf John Arras' GenMUD NPC populations (which, as one of those
    intellectually delighted engineers I *really* like).  In this model
    the players are surrounded and largely dwarfed by larger, more
    populous, and more effective (at changing world state) NPC
    populations.  To a large extent the model defines a player who is
    both a bit-part actor in a larger (variously unknown) story, and
    driftwood caught in the maelstrom.  The lemmings provide a context
    against which the player acts.  The player can affect the lemming
    behaviour, but the impedance between machine lemmings and humans is
    high enough that the relationship becomes one either of disengaged
    tarbaby manipulation, or driftwood in the maelstrom.  Both forms
    have poor emotional involvement and player retention factors.

  3) Infrastructure

    They are flexion points that either haven't been occupied by
    players, or are (thought to be) insufficiently interesting for
    players to occupy.  Basically implementations of game state
    transforms which require some sort of thematic external
    justification and identity.  Usually just means some sort of state
    transition machine which bears a "name".  The common case is often
    also a vendor redux.

  4) Vendors

    The standard models vend data or items, advanced models vend
    activities or (coordination) opportunities.

  5) Scenery.

    They're there, they (may) move, but you can't do anything to or with
    them.

Outside of the scenery case, the common aspect seems to be that they
have an identity, and can thus be (uniquely) identified (not necessarily
repetitively).  In the simplest form this means that they bear a name,
even if the name is implicit rather than explicit (cf Big Boss).
Identities are important as they allow the causes/sources of effects to
be, well, identified, and they thus provide anchors for emotional
relations and investments.  Without identities you can't get that.

> The steps beyond that are to have progressively smarter NPCs.

Here's a point where we part.  Supporting thought process:

  The basic motivation and interest of any person is to create an
  effect.

  Creating an effect on self is better than creating no effect.

  Creating an effect on an object is Okay.

  Creating an effect on someone else is good.

  Exchanging created effects with another is best.

Effects of course can be of any type.  They don't have to be griefers.

As the players are communicating indirectly via computers, all the
actually physical effects are out.  What's left is what can be
communicated (pretty much) in text and simple in-world actions (cf
semaphores).  We can simulate physical effects on virtual objects, but
such are weaker than directly creating or being involved in, or sharing
emotional state changes in other players.  The better/stronger effects
are those that directly manipulate the intellectual or emotional state
of the player.

Intellectual effects essentially sums to data transfer.  Such deltas are
typically uninteresting unless they are also accompanied by emotional
state changes.  ("The door is blue" is a generically uninteresting data
transfer) Emotional state changes require empathy by the recipient for
the cause of the emotional effect (may be different from the source of
the communication).  Devolving story into the simpler form, "drama" (cf
earlier comments on human emotion and reaction), this of course harks
back to Seller's recent comments on emotional involvement, earlier
discussions with Lee Sheldon and others regarding whimsy and
retrospective drama/emergent_story etc.

<insert long logical bridge that should be obvious>

So we're back to pulling the player's emotional strings and creating
(sorry I forget the term from Lazarro's talk) a strongly cyclical low
frequency graph of the emotional state of the player during play.  But,
as our games are intensely multiplayer we don't in fact control the
content of the communication channel to the player (the player
themselves and the other players primarily do) so we're not in the
privileged position of traditional authors, screenwriters, playwrights
etc (single player games are fundamentally different in this regard) and
are instead rather in the three-steps-removed position of trying to
manipulate levers which move levers which prompt players to create
situations and communications which cause those emotional
rollercoasters.

It goes back to expressive fertility (cf laws): a medium which is very
fertile in the sense that it aggressively, even actively, supports and
encourages expression.

And there's the hook back to whimsy.

> As allies, they become the social context in which the players
> operate. They decide to build a town here, dig a well there, clear
> this land, war on that guild, put out that forest fire, etc. And they
> need help.  The players have a purpose, a meaning, for their
> gameplay.

This rankles me with the question:

  Why are they necessarily NPCs?  Why aren't they other players?

Is there some special quality to NPCs that make them more attractive or
in fact more functional for such operations than other humans (players,
GMs, or human-driven NPCs)?  This isn't to say that every possible role
that could be occupied by a human must be occupied by a human, but
rather in observance of the fact that if NPCs doing XYZ are interesting,
then humans doing XYZ are inherently more interesting.

> It fits into the social context of the game.

Especially given the relationship between "other humans" and "social
context".

> As enemies, they become more difficult to predict.  They make plans of
> conquest.  Intelligence suggests planning.  That planning can be
> spotted and interpreted by the players, providing them with more
> complex scenarios for combat.  Instead of 'pulling' a steady stream of
> monsters, players can manage and react to what the NPC monsters are
> doing.  It gives purpose and meaning to gameplay.

Absolutely, but again, is this an improvement over other humans/players,
and if not, are in fact NPCs the most efficient and effective way of
accomplishing this end or are they simply the more familiar?

> The value of NPCs is orthogonal to the fundamental pursuit of a goal.
> NPCs provide the goals, and they make them more interesting.

I'd insert a bunch of "can" and "may" qualifiers in there.

Not so long ago the big cry was to simulate a "real world" with "real
behaviour" and "real physics" with everything simulated "just the way it
would happen if it were REALLY REAL!".  It was going to be great.  Then
we found out we were producing games not simulations.  My sense is that
we are heading into a similar dichotomy with NPCs.  The great cry is
that "NPCs need to be more real!", "NPCs need to be believable!" , "NPCs
should have Real People Personalities(tm) and Simulated Real
Emotions(tm)!"  We have similar sounding supporting logic statements
being made for why such RealPeopleNPCs(tm) are such a Good Idea, heavily
salted with personal belief and (excuse me) arm waving claims of, "It
will be great!"  I don't buy it.  The arguments seem to equate to:

  It would be great if we did it with people, so if we can do it with
  software that is "Just Like People" (for some non-Turing definition of
  "Just like People") it will be just as good if not better, and cheaper
  too!

Uhh, yeah.  Given the roles that NPCs fill I don't see that the first
assertion of, "It would be great if we did it with people" has been
demonstrated and instead see considerable suggestion that it is in fact
an article of unsupported faith that doesn't sound right.  Then again,
perhaps I am just being a curmudgeon as Mike Sellers so politely
suggests.

--
J C Lawrence
---------(*)                Satan, oscillate my metallic sonatas.
claw at kanga.nu               He lived as a devil, eh?
http://www.kanga.nu/~claw/  Evil is a name of a foeman, as I live.

_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev