Truly, señor Lucas is a one-man comedy gold mine.

What makes me smile when I read comments like this one from Lucas Arts or from EA "calling" for better game AI, I wonder "who are they calling"? Am I to understand that these $Billion companies expect us to fund this work by ourselves (sorry facetious rhetorical comment there, hehe)?

George, we need your money, not your call to action.

Operators are standing by!

did you guys make it to the siggraph keynote he gave? it was interesting to say the least...

Aren't we going round in circles here? This thread started with a link to a summary of Lucas' SIGGRAPH keynote.

Here's the trick though. As a game developer, we need for the publishers to *realize* that they need good AI. I'm working on a title where our publisher has come to us and asked us to do some of the things we've wanted to do with AI for a long time. That's exactly what we want! The more press AI gets, the better it is for us developers who don't have to explain why they want to hire several full-time AI programmers for something that can be faked with scripting. Then we get to work on incorporating some of the dreams we've had for ages, because the mass market is starting to recognize AI and ask for it.

Great point. What you're describing is the way it really *should* work -- the interest needs to come straight from the publisher, and the publisher needs to be listening to the market, who really do want better AI.

We need more visibility for AI in the press, and a better understanding of AI in the media and the general public so that more publishers will start to think this way.

I know it's kind of a broad question, but does anyone have good ideas for ways to improve our visibility and/or better educate the media about game AI?

1) Stop promising reactive AI and then drop me into a world where the AI is dumb as rocks in all but the obvious situations. The best AI I've played in a long time came from an augmented FSM, not from a neural network or genetic algorithm. AI needs to learn what graphics did, and that's that a simpler model can look and act more complex than something that attempts to be like the real thing. The best example I can think of is in Killzone, where the article I read on suppression fire and tactics was great and a nice piece of tech, except that in reality in the game the AI is still maddenly dumb and easy to defeat even on Hard mode.

2) Give them something tangible, something they can notice. Spend more time on making the characters look and act real all the time instead of making them hunt me more realistically. If the character is hunting me, make sure he looks like he's hunting me. Feeding internal AI mental state into outward appearance goes a long way to making me believe in your AI. If you can't sell it in 10 seconds, then you have nothing to sell.

3) No matter how great your conversation AI is for the guy I buy my arrows from, if your monsters all blind rush me all the time, I see no AI and therefore think it sucks.

SL, I really can't disagree with the general gist of your comments, but I just have to ask regarding point #1 ... did anyone on this blog actually suggest using neural networks and genetic algorithms for AI?

If you'd like to criticize the academic AI researchers who are doing machine learning, I completely understand, but I hope everyone reading this understands that we're a completely different group of people.

Paul T:

I think you may read a little too much aggression into my comments. Please chalk it up to the limitations of text based communication.

As for this group of people, I read this blog because I respect the work people here have done and I hope to learn something as I continue in our chosen field.

The original question was, how do we make publishers and the press care. The suggestions I made above are general suggestions to AI people across the field, not to anyone in particular. For instance, I have read a whole lot about Facade and how great it is, and I agree, it is compelling, but I see very little there that is useful in making a -game- more compelling, or anything there that would make a publisher care about AI. Not for the cost and amount of work it entails, at least.

Every year I read articles on "what's next beyond the FSM" and "how can I create great AI conversations" and yet none of those articles address the very real problem of people in game AI overusing technology because it sounds cool when it adds little to nothing to the gameplay. The way to make publishers care is to just provide better AI and to do it in ways that they see as being cheap and effective. My only point was, and not to anyone in particular but to everyone that's ever worked in AI, KISS. Keep it simple stupid. It's the oldest and best rule I know.

Of course, I am very open to hearing different opinions. These are just my own.

I think SL up above nailed the core of it. To sell AI, what makes it different and cool has to be instantly apparent. Technology doesn't sell; demos do.

One thing we're finding is that better AI doesn't just make better gameplay -- it makes new kinds of gameplay. There are things you can do in a game with robust AI that just don't come up if you're dealing with bots-on-rails. Selling that isn't easy, but I think once it takes hold publishers will see what can be done with better AI. I hope so anyway!