As I mentioned before, I was at an AI conference last week. The conference was a mix of invited talks by industry people and academic paper presentations. It didn’t quite gel into one unified experience, however.
The audience was split maybe 75% academics (professors and PhD students), 25% game industry folk. The two groups have very different agendas about what they want out of such a conference. For me, there are two kinds of talks that I hope to hear. The first is a GDC-style talk. These are talks where the draw is either a talk about a game that is exciting or the speaker is renowned for past/current projects and we go just to hear what they have to say, even if it’s not strictly relevant. An example of the former would be me going to see any and all talks about Guitar Hero at GDC. An example of the latter would be going to see anything Will Wright is talking about, whether it’s Spore or Astrobiology.
The second kind of talk I’m looking for is a clear explanation of applicable algorithms or techniques which I can actually walk away and implement into my game. I don’t have really any good examples of this one. They don’t really happen. And this is what I keep hoping AIIDE will be, but it’s not quite there yet.
This is the 3rd year of the conference, so it’s still fledging. It’s quite small in comparison to a show like GDC, with only around 100 attendees. I missed last year, but made it the first year. This year was better than the first, in my opinion, because they had a ton more people speaking from industry than the first year. Admittedly, they were almost all invited talks rather than paper submissions, but that’s ok in my mind. Well, the first year they had Will Wright, but I don’t think one Will Wright can compare to all the speakers we had this year. I’ll hopefully get around to talking about the actual speakers and what they had to say in specific later this week.
As to the academic community, my impression was that a lot of them view this as yet another opportunity to get published, which is of paramount importance to them. AIIDE is funded in part by AAAI and papers are published through them, so it’s a pretty respectable AI publication, I guess. Hey, I’m published now, as a result of getting a paper accepted to last year’s conference. But that doesn’t matter too much to my career, unlike those in academia. Well, maybe I’ll find a way to post my article at some point, not sure where my soft-copy of that thing went anyways. But I digress.
A colleague of mine received his PhD from MIT and recently entered the game industry to work at my former employer, Mad Doc Software. We had a long talk about the conference and he commented on how frustrating he found the industry talks. He didn’t enjoy being shown ideas without the information on how to implement them (those would be the GDC-style talks). These weren’t useful to him. To me, these talks are about getting excited and energized to possibilities. I love hearing someone who thinks in a way different than my own go off in different directions to get me thinking (I’m thinking of Will Wright and Chris Hecker who spoke (partially about Spore) at this year’s AIIDE). But it’s true that there’s nothing that you can take home and implement from one of these talks. All you take with you are new thoughts that might lead to new innovations.
The other big problem is the disconnect between what we consider AI in the games industry and what academic AI is. They have time to do nothing but work on AI problems, with all the resources around them devoted to that problem. AI engineers are very lucky if they get 20% of the CPU on the newest consoles (and less on PC titles with low min specs) to do their work. We can’t even work on the same problems. Our goals are to help provide an interesting and challenging interaction with the player. And to mostly not do anything too stupid, which is the kiss of death for game AI. Their goals are to do something new and advance the state of knowledge. Of course there’s some amount of overlap with some areas of academic research and some areas of game AI but it isn’t clear where that lies.
What I’d like to see is tutorials about writing game AI from the perspective of people with knowledge of both sides, not just one or the other. There are techniques we’re using in the industry that might be sub-optimal but we use them because they’re all we know. We’re not reading academic AI journals when we do research, we’re reading the AI Game Programming Wisdom series - a book filled with articles written by people in the industry on what they’ve done and learned, frequently without reference to what the academic community has done. But it’s readable and understandable by folks like myself who don’t enjoy reading papers filled with Greek letters.
Don’t get me wrong, I quite enjoyed the conference. It got me thinking about different problems and what it is we do as AI engineers. Some of the academic presentations were enlightening as well, though few contained anything that I’m going to run out and implement any time soon. But they also contributed to the thoughts a-whirling in my mind, and that’s a good thing. I’m just trying to think of how we can make this better and make it work for everyone involved. Because I think this conference is important and needed, so I want it to succeed.