Slashdot Mirror
Game AI Conference Explored
Academia Blog Grand Text Auto has up a long set of notes from last week's first AI and Interactive Entertainment conference, which includes keynote talks from Doug Church, Will Wright, Chris Crawford and Damian Isla of Halo 2. From the Doug Church talk: "none of the AI detail gets attention in a 30 second ad or magazine blurb...also, if a character in battle only lives a minute, there's not much fidelity players can even perceive...industry has been promising good characters for a long time, not delivered...
players are cynical, don't want to hear it anymore...hard to back out of the fakery"
I wonder if the experiments in simulating the human brain will come in handy with making an AI. I mean if we can understand how the human mind makes decisions, we can better code an AI that can make decisions the way we can based on environment, situation... how we feel about the world... etc.
If you like what I've said here, and want to read more, go to http://www.krillrblog.com
I'm covering the overall architecture of the brain right now in my intro. psych. class.
It looks remarkably like a computer architecture when broken into components. You have your I/O neurons, interfacing with a component that discards noise. From here it is put onto a bus, into what the textbook labels as the "executive" (CPU). The executive can store and load from short-term memory (registers)---there are different kinds for various senses and parts of cognition---and do other brain-type things. It can also store and retrieve from long-term memory.
It is a computer---just a very, very analog kind.
It looks a whole lot like a computer because that's the dominant metaphor that we're using and we're filtering our explanations through that idea.
At the time of Descartes, they thought that everything looked kind of like it was clockwork and so they made explanation using that metaphor. Freud did a pretty good job of explaining the psyche in terms of conflicting forces and Jung did it by populating our heads with stereotypes.
Metaphors are cool that way - they are flexible and once you get a halfway good metaphor, you start noticing all kinds of similarities (for example have you noticed that EVERYONE either looks like a pig or a rat? try it!). But it's good to remember that the map is not the territory.
I have a lot of opinions about Cyborgs and Architects
if the NPCs gave chase to you DEPENDING on where, when, how and who/which NPC saw you.
Functionality most likely implemented through a huge set of IF-THEN statements.
That's to some degree all that AI is, whether it's an expert system or an Eliza-style conversational program. You really don't get away from if-then until you start talking about computational intelligence, like neural networks, Dijkstra-style pathfinding, and the like.
I wonder if the experiments in simulating the human brain will come in handy with making an AI.
AI for a computer game? Hardly.
One of the most effective, fun game AIs I've played against recently is that in Halo - it's probably no more advanced than that in some other games, but it has some great application of smoke-and-mirrors and does a good job of presenting obvious cause-and-effect behaviours to the player.
Kill a Covenant Elite, and all the lowly grunts nearby will panic and try to run and hide. But to actually shoot that Elite, it's probably taking cover behind a rock, waiting for you to attack - it doesn't just meander into combat, shooting blindly.
All sorts of things like this - simple 'IF foo THEN bar' behaviours which the player can learn, understand and anticipate can be great, so long as they're fun to play against. Some hyper-intelligent enemy that can figure out precisely where the player is and attack unseen might be programmatically more advanced, but isn't necessarily more fun to play against.
In-the-field tactics are probably best left to the game AI, but higher-level, map-specific scripted strategies can give the illusion of some overall plan behind the enemies' actions (plus they can be designed to be fun to fight against, rather than being whatever the AI might extrude - fun, crap or otherwise).
Neural networks or whatever might be more 'realistic', but they won't necessarily be better to play against...
Tedious Bloggy Stuff - hooray?