Thanks to Damian for inviting me to his blog! I'm looking forward to seeing its development as well. :)
First, a quick introduction. I've been working on AI research for the last six years or so. In my previous life I was a roboticist, getting autonomous robots to behave intelligently in natural human environments full of unpredictable people. But then I realized games are much more interesting - and metamorphosed into a game AI researcher. My dissertation, finally defended in April, was on a new technique for social interaction through natural language. But I think I've had enough of academia, and I'll be joining a small simulation studio this summer. ;)
Since AIIDE was the last time I spoke with you folks, I just wanted to write a few words about some of the interesting bits I noticed at the conference. There were some common themes to the talks and conversations, some points revisited independently by multiple speakers, that I thought might be good predictors of future trends. I'd be curious to hear what you folks thought of them as well.
First, there was the sense of impending complexity explosion. The increased computational power of next gen platforms, and players' appetite for higher-fidelity simulation (realistic physics, open worlds, complex world manipulation, etc) will force us to adopt much more granular world representations. But a fine-grained world model will only complicate AI. Game worlds used to be full of simplifying assumptions, and much of game AI development concentrated on their efficient exploitation. But as these simplifications are being removed - as we move from static grid maps to dynamic and malleable worlds, from pre-rendered animations to runtime movement generation, from single-state behaviors to complex and naturalistic performance, from single units to large groups - the old techniques will not necessarily fare so well. Doing A* on a static grid map is vastly different from having hundreds of units simultaneously navigating and reacting to their environment in a dynamically-changing 3D world.
But to (mis-)quote Doug Church, even though it looks like we're standing still, we're still moving forward; though it looks like our AI isn't doing new kinds of things, it's doing existing things in ways that are more general, more scaleable, more comprehensive. Personally, it seems like a very positive trend - knowledge representation is crucial to good AI.
Second, the production process. It seems that low-level content production could really benefit from some AI as well. Current console titles are already expensive and risky to develop (to say nothing of the 2006 and 2012 console generations). So maybe we could develop more intelligent tools, that would do some of the mundane work for us, and let us do authoring using high-level representations?
Abstraction and composition are standard ways of dealing with complexity, and they're frightfully effective. We can try to proceduralize some of the low-level authoring processes into abstract elements, and move them into the engine, into game elements, or even the tool chain. Easier said than done, of course, since that introduces all new kinds of risks. :) But there has to be some valuable low-hanging fruit in this space.
Third, there were some mentions of complex AI perhaps leading to novel kinds of gameplay. Action games, sports games, fighting games, tycoon games - these have been made for a while, and we've gotten pretty good at doing good AI for them. But could better AI actually drive gameplay towards some new styles of games?
This relates to the complexity management theme as well. Simplifying assumptions both enable and limit what game AI can do. So as we remove simplifications, we will hopefully end up removing the limitations as well.
So I suppose what I found most interesting about the conference was the persistent theme of increasing complexity of game AI, and how to manage it. Games are becoming much more complex, and we'll have to continue finding novel ways of dealing with it - through better knowledge and behavior representation, through the proceduralization of the authoring process, through the expansion of the role of AI to include more than simple low-level smarts. It may look like we're standing still. But I'd like to think of it as stopping for a moment to put on running shoes. :)