Newsgroups: rec.arts.int-fiction
Path: news.duke.edu!newsgate.duke.edu!nntp-out.monmouth.com!newspeer.monmouth.com!feed2.onemain.com!feed1.onemain.com!feed1.newsreader.com!feed2.newsreader.com!uunet!ash.uu.net!world!buzzard
From: buzzard@world.std.com (Sean T Barrett)
Subject: Re: What's this thing we call AI?
Message-ID: <GJx9L1.BE5@world.std.com>
Date: Wed, 19 Sep 2001 18:19:48 GMT
References: <a69830de.0109152302.7596d47e@posting.google.com> <a69830de.0109170920.5bce523d@posting.google.com> <GJtnsv.3Ir@world.std.com> <3ba86df2.395305184@news.verizon.net>
Organization: The World Public Access UNIX, Brookline, MA
Lines: 56
Xref: news.duke.edu rec.arts.int-fiction:92765

Jason Melancon <jasonmel@afn.org> wrote:
>buzzard@world.std.com (Sean T Barrett) wrote:
>> The moment the NPC AI starts coloring its actions based on the
>> potential player experience *as a game*--or, alternately,
>> the moment the NPC begins coloring its actions based on our
>> goals for the player instead of the NPCs goals--you are to
>> my mind writing game-mastering AI, not NPC AI.
>
>If you were an NPC in my game, wouldn't you care whether you
>accidentally locked me in the attic?  I would hope you would let me
>out when I started making noise.

It depends on the NPC, doesn't it? On the NPC's goals. Like I said.

>My point, of course, is that being
>humane and not locking people in attics is a perfectly desirable,
>albeit sophisticated, goal.

The attic is merely a particular example, and not the best one
for my side of this issue. IF tends to involve a very fragile
simulation; in the real world, we rarely get so stuck that a
goal is totally unachievable, but it's easy in an IF to, e.g.,
consume a non-renewable resource that the player needs to win
the game. I'm all for flagging such resources and having NPCs
not consume them in deference to the author's goals for the
player, even if the NPCs goals are in conflict with the
protagonist's; but to me, that's not NPC AI, that's game "AI".

This sort of thing happens automagically in pen&paper RPG's;
the gamemaster takes on the persona of an NPC and provides the
"AI" for that NPC, but at the same time the GM leavens the NPC
actions with the GM's goals for the entire game/session.

[re: Emily's post]
>I doubt this is your main point, but couldn't we make a distinction
>between intelligence and free will, two orthogonal properties of an
>NPC?

"Free-will" is nearly as loaded a term as "intelligence". I think
people generally use it to mean "having a choice over what action
to take", but you seem to be using it here to mean "having a choice
over what goal to achieve", since an AI which is pre-programmed
with only one choice per scenario can hardly be an AI. (And of course,
save random numbers, any AI will actually only exhibit one behavior
in any exactly identical situation, so it's hard to argue that any
AI of any size really exhibits free-will, hence the problem with
the term. Some people argue that free-will is as much an illusion
as recent posters have suggested consciousness is.)

Even limiting the goals doesn't really address the problem of
an AI which makes plans and achieves them; to achieve a plan one
often creates side-effects on the world, and producing the list
of "allowable" side-effects is complicated, especially if the AI
doesn't know what the consequences of all actions are in advance.

SeanB
