Implementation of Machine Learning into Artificial Intelligence?

[MFL]

Hey everybody, I took a turn for the weird today and was reading through DARPA Press releases and stumbled on one that was talking about probabilistic Programming languages and building tools to take advantage of them,

http://www.darpa.mil/NewsEvents/Releases/2013/03/19a.aspx ,

and it got me thinking about a topic I like to regularly bring up here every so often, how Half Life 2 didn't take advantage of its artificial intelligence.

Now I honestly don't know too much about programming, machine learning, or artificial intelligence at all, so I was wondering, if we had any members here who were knowledgeable about the subject, if I could ask you a few questions. Mainly, how close are we to seeing this kind of programming being implemented into the games we play? Even more specifically, first person shooters and real time strategy games, if only because it's easier for me to imagine the possibilities in those genres. Or is talking about it in that context really not making any sense in the first place?

 

[Maestro]

Creating learning machines is a prohibitively expensive endeavor, and it would yield little in the way of gameplay improvements compared to the cost. Not to mention, even in real life, the responses of enemies are never based on what you have already done. Their response is determined by what they know. Unless they are the Borg, then there is no way they could 'adapt' to your tactics over the course of a game.

 

[Vegeta897]

Creating learning machines is a prohibitively expensive endeavor, and it would yield little in the way of gameplay improvements compared to the cost. Not to mention, even in real life, the responses of enemies are never based on what you have already done. Their response is determined by what they know. Unless they are the Borg, then there is no way they could 'adapt' to your tactics over the course of a game.

Not in an overall sense, but AI in games can certainly remember things you've done if they're programmed to look for it and act accordingly.

But actual AI that learns on its own is nowhere near feasible yet for games. Our CPUs aren't even fast enough to handle a bunch of AI that have a lot of routines. In the current method of programming AI it's all about creating routines and determining how the AI chooses which ones to do. Arma for example has one of the most complicated set of AI routines in games, and it takes a serious toll on modern CPUs when you have dozens of actors all doing this. Not to assume Arma's AI is perfectly optimized, but it's safe to say our current CPUs aren't able to handle the most of what we can get out of AI routine programming, so looking into real AI that actually learns on its own is beyond the scope of current computers.

 

[MFL]

Well damn my jimmies for rustlin.