The paper "How to Lose at Tetris" (http://www.geom.uiuc.edu/java/tetris/tetris.ps) says that you don't even need to know what the player is doing. And that even almost all [1] random tetris games are bound to be lost, even with perfect play.
[1] "Almost all" is a technical term in probability theory and means "with probability 1" (which is not the same as "all games".)
I wonder whether that is equivalent the same technical term from set theory which mean "all except for a finite number of exceptions", and thus with a finite underlying set, it's correct to say that "almost all" elements have a property when in fact none of them have it...
It's similar, but with a finite set it wouldn't show up unless you had some outcomes have probability 0 (in which case, why not just delete them from the outcome space?).
Where you need the idea is in continuous spaces - like a uniform [0,1] random variable. With probability 1 it's not a rational number, etc.
[1] "Almost all" is a technical term in probability theory and means "with probability 1" (which is not the same as "all games".)