Tell me more ×
Game Development Stack Exchange is a question and answer site for professional and independent game developers. It's 100% free, no registration required.
up vote 2 down vote favorite
2

In my work I analyze and design Evolutionary Algorithms (EAs), but for fun I'm learning game design, and, of course application of EAs in games. So I'm wondering if there have been any games where AI actually works as AI, i.e. learns from past solutions to develop new ones. I suspect it should be something like tic-tac-toe or chess.

share|improve this question
2  
Killer Instinct on SNES did this. It learned what combos it "saw" you doing and would start breaking them later. – ashes999 May 28 '12 at 22:31
1  
Black and White had a learning AI, but it wasn't learning in the same way you're asking about. It was trying to learn from the player. – Laurent Couvidou May 28 '12 at 22:54
1  
While this is an interesting question, I don't think it's a fit for the site. It's not constructive and there's no correct answer. Additionally, something very similar has been asked before: gamedev.stackexchange.com/questions/17447/… – Byte56 May 28 '12 at 22:54
1  
gamedev.net has an AI forum where this discussion might get some good traction, if you can drag them away from their creepy love affair with neural nets =P – Patrick Hughes May 28 '12 at 23:10
@Byte56: thanks, I didn't know that. MSE and SO accept quite open questions, so I thought it's the case here as well. Will keep it in mind though. – sigma.z.1980 May 29 '12 at 10:06

4 Answers

up vote 3 down vote accepted

Since you mentioned genetic algorithms and not specifically unsupervised learning, I thought you may find this interesting: http://boxcar2d.com/

It uses genetic algorithms to mutate better cars for the terrain based on each generations fitness. This is one of the most direct uses of EAs that I've seen that seem to be what you're looking for.

If you're looking for unsupervised learning as opposed to EAs, I can't help much since my experience is in the latter. Hope this was helpful.

I'de note this could also be done for tic tac toe or chess as you mentioned, the difficulty is just in modeling the genomes correctly to come to a solution (not necessarily a correct one).

share|improve this answer
awesome toolbox thanks) – sigma.z.1980 May 30 '12 at 10:32
up vote 3 down vote

While I cannot think of any commercially available games which did this, you may be interested in some of the AI challenges that resulted in such AI, perhaps most notably StarCraft's Berkeley Overmind. http://overmind.cs.berkeley.edu/

share|improve this answer
2  
+1 great link, thanks for sharing – ashes999 May 29 '12 at 1:57
up vote 0 down vote

I would imagine that it's just too easy to fudge the AI and make them act intelligent instead of work intelligently. Most commercial game engines come with some basic AI capacity that eliminates most of the incentive to really build something new and cool anyway.

For the cases you mentioned, tic-tac-toe has been completely solved for quite a while now (http://xkcd.com/832/) and while chess has not (way too many permutations to really be viable) the basic algorithms, ie those that would be useful in the real time setting of a game, aren't really learning algorithms, they simply rank available moves.

share|improve this answer
up vote 0 down vote

I can't cite specific examples, but the guest on this show http://www.se-radio.net/2008/04/episode-92-introduction-to-game-development/ says that's a common practice to use Genetic Programming (or other heuristics) to make an AI

The advantage of using this natural computing approaches to AI is that you can create artifical dumbness without too much trouble, wich is generally desirable. Because as we know, for some games a perfect AI is trivial to do, but a 'human-like' one is not.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.