March 29, 2009

Awards, Articles, Conclusions

by Nick Montfort · , 4:23 pm

3 Responses to “Awards, Articles, Conclusions”


  1. Malcolm Ryan Says:

    Yes, GDC is over. There was some interesting discussion about making games with more meaningful stories. Some seemed to think we were already doing it, some that we had a long way to go, and some that the whole idea was impossible and a distraction from what games should be about. (I’m never really sure what that ‘should’ means.) If nothing else, it shows that the idea is still very firmly on the agenda, although I’m not sure that many new answers are being found.

  2. Andrew Stern Says:

    Of course, Malcolm’s comment could have been written for any of the past 15 GDCs. ;-)

    I missed the conference, which also included an AI summit in the first 2 days, and in parallel to that, a AAAI symposium on Intelligent Narrative Technologies — http://www.cc.gatech.edu/conferences/aaai-int2/submission.html — a sequel to the one Michael, Nick and I attended in DC in 2007 — http://grandtextauto.org/2007/11/14/intelligent-narrative-technologies-2007/ .

    Let’s search for some blog writeups of these two events…

    I did happen to arrive in SF just as GDC was ending, and had some drinks with some folks who were at GDC itself (including Michael), who thought GDC was a bit tame this year.

  3. Mark J. Nelson Says:

    This isn’t really a proper blog writeup, but I went to many of the AI Summit events, and it was interesting on the whole. About half was a tutorial of standard methods for people new to the field, plus some tutorial-style overviews of new technologies, like how to multithread your AI algorithms. The other half was more brainstorming and panels of how to take the next step in AI. For that part, I’d say the same summary Malcolm gave kind of applies, “these are hard problems, and we’re not quite sure how to solve them, but they’re worth discussing”. It is interesting, though, that I think industry may have caught up to the cutting edge of academia in the space of what the problems with expressively-controllable AI are. Most techniques you can think of, from planning to machine learning to classic knowledge-rep literature, are being mined and adapted by one fellow or another, and some (Richard Evans, Damian Isla, etc.) are very up on the literature.

    Speaking of expressiveness, in some informal discussions, many game-AI people were actually sort of surprised that the “expressive AI” thing is a Mateas innovation and not what the entire field of AI is doing. The reaction was sort of, “but of course the main issue in AI is how to have it be expressively controllable by a designer; that’s the whole point” (with the exception of a few narrow areas like path planning, which are more traditional AI “solve the problem optimally” problems). There were also some discussions about where the line blurs between authoring tools and AI— if your AI is “just” an FSM, but you have some excellent tools that let a designer easily generate and edit lots and lots of those FSMs, is that worse than some fancier runtime system? When does it matter?

    Probably the most interesting to me was a panel following up Chris Hecker’s suggestion a while ago that we should look for whether there’s a nice structure/style decomposition for AI, akin to the triangle mesh for 3d graphics, or Photoshop for raster graphics. That branched off into a more general discussion about declarative versus procedural representations of AI, with widely expressed skepticism that there was some way to take the procedurality out of AI except in narrowly defined areas. An example where it does work was that “structure” for expressively moving soccer players is basically joint angles, and “style” is the functions that compute those joint angles. Stuart Reynolds argued for a “behaviour capture” analogous to motion capture that uses ML to automatically fill in such parameters from example gameplay traces. Most of the other panel members were of the contrary option that figuring out what the parameters even are in a particular space is the whole problem (i.e. what the AI-relevant structure of the space is), barring the magical appearance of a clean structural decomposition of All Intelligence.

    The latter panel invoked Michael in absentia a few times to argue the pro-procedural viewpoint, that AI isn’t just setting a bunch of sliders on a to-be-discovered set of sliders, but inherently procedural; and therefore that we need procedural designers, not to figure out how to factor out the procedures so designers can set the style declaratively. Facade was also mentioned as “in many ways, still a high-water mark for game AI”, so at least this subset of people definitely seem interested in going in that direction, but are mainly undecided on how to do so in practice.

Leave a Reply

Powered by WordPress