RE: Re[2]: early AGI apps

From: Reason (
Date: Sat Nov 09 2002 - 18:57:19 MST

--> Stabbert

> Saturday, November 9, 2002, 12:54:31 AM, Reason wrote:
> R> The military and the gaming industry (and the military gaming
> industry) are
> R> ripe for something that commoditizes AI in the game sense. By
> this I mean
> R> customizable, highly abstracted decision making and knowledge
> R> representation/processing tools. Everything is roll-your-own,
> which is an
> R> insane state of affairs if you stop to look at it. Let's take
> the gaming
> R> industry: as much effort goes into AI (human simulation,
> essentially) in
> R> your average game as goes into the 3D engine. 3D engines are
> commoditized,
> going for $50k ->> $300k per game license. The space is
> competitive. Why are
> R> people still building their own AI (human simulation) code
> from scratch each
> R> time? It's certainly not anything to do with complexity or level of
> R> customization required...
> As far as gaming goes this is not really accurate. The AI is very
> limited, mainly because of processing time constraints, it is very
> gametype- and engine-specific (different for RTSs, FPSs, etc.) and it
> is not being built from scratch each time: with Unreal's UT2K3 engine, for
> instance, licensees get the AI code right along with the 3D engine
> code. A general "human simulation" package would at this point be far
> too costly to run, resource-wise.

My specific interest in game AI is not FPS bots -- that's a solved problem
so far as I'm concerned. That is indeed a pseudo-commoditized space in human
simulation; it's basically a packages for navigating 3D space within certain
parameters. Tightly tied to 3D engines, yes. As soon as you move away from
run-and-gun behavior, however, you see that there's really nothing out

I'm more interested in things that can mimic people, societies and plots in
MMOGs and CRPGs. Speech, moods, goals, emotions. (So is the US military,
funnily enough). I think it's fair enough to say that everything except FPS
bots are pretty much being built from scratch each time. Just look at any
recent interview with Peter Molyneux -- he's probably doing the best work in
this sort of societal human simulation thing, but he's nowhere near nor does
he appear to be headed in the direction of a general engine.

And yes, you wouldn't be running a general human simulation; the goal is
really to approach human complexity as the architecture allows. But it seems
that much of what is needed to perform the MMOG/CRPG type of
social/plot/goal/mood human simulation can be parallel or have some
conceptual overlap with early GAI groundwork.


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:41 MDT