[MUD-Dev] AI not worth doing in our games?

[Robert Zubek]

From: Freeman, Jeff

> I want to say: In most cases we don't need smarter AI.  We can
> achieve the same or better results with more diverse AI and/or
> some slight-of-hand.

> e.g. A "real" AI for pigeons in the park would be big, complex,
> and result in flocks of pigeons wandering around the park.  We can
> make pigeons do that without a big, complex AI, though.  So if
> that's the desired result, we should just do that.

Maybe we can put it this way: intelligence is in the eye of the
beholder, and it's contingent on the context and the task at hand. A
mobile robot may not be able to create new math theorems, and a
theorem proving engine can't move around a simple office building,
but we can't easily say either is less intelligent - their domains
of competence are completely unlike each other.  As behavior
designers, we need to consider first what kinds of competence we
want the agent to have in the world, and only then what kinds of
mechanisms are needed to achieve it.

Another problem is that there's a wide range of levels of
abstraction for intelligent behavior. We can model just the pigeons'
feeding and avoidance behaviors as force fields, or we could model
cognitive processes that lead to those behaviors, or the neural
processes that lead to those cognitive processes, and so on - but as
we go up in generality, we also lose control over the performance of
the task at hand. FOL inference can be coerced to produce the same
avoidance motions as a simpler behavior-based model, but the
generality will cost us increased computational complexity, which
may not be a reasonable cost if the generality isn't really
necessary.

> The areas in which faking it just won't do are areas of
> competition with players.  But even in those areas, the players
> don't want the AI to be competitive anyway.

But the distinction between "real AI" and "faking it" is really
blurry.  Maybe it's just that players want the AI to have broader
competence that what's been done thus far?

Assuming human performance as the ideal of intelligence and
competitive performance (or is this a bad assumption? ;), our NPC
AIs lack almost everything that characterizes human players. The
only aspect we've been able to model thus far is small-scale
tactical performance - the running around and shooting and hiding
bit, which has been modeled deeply and narrowly. But larger-scale
tactics, such as working with other members of your group, strategic
thinking, spatial reasoning, conversing with other players, emotive
displays, boasting and trash-talking, engagement in social rituals -
these we still have no idea how to model properly. It seems these
kinds of competence would make the game more fun, though, exactly
because they would make the agent competitive (and cooperative) in
more interesting ways...

Rob

--
Robert Zubek
http://www.cs.northwestern.edu/~rob


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