General-Game Playing

http://games.stanford.edu/ is about general game-playing AI (deterministic games): if you change the game, the player can still adapt. They seem to be striving for more intelligence in game AI. Their competition seems to sell itself as a test of general intelligence.

This paper seems interesting. It defines a language for encoding all games.

But it feels to me like they're making up problems in an artificial world.

The space of possible games is huge, and the distribution of games that you draw from is going to create a serious bias in what counts as "intelligent".