To everyone commenting on how this is proof games today are too easy or things like that, this study has nothing to do with the difficulty of the games. NP-hard is a term in computer science refering to the time required for a computer to solve a given problem. Think of it as how much time it would take an AI to solve any given level for the game (time in this case not being a matter of seconds or minutes, but more of the relative difference between cases, for example will a level twice the size take twice as long to solve, four times as long, or more?). NP-hard means that as the size of the case grows (i.e. bigger levels), the time taken for a computer to solve it generally increases exponentially. Any modern game is likely to be NP-hard as well, but it would be more difficult to prove.
This is a pretty big simplification, but if you want to read more, Wikipedia has several articles about computational theory, though they are not very beginner-friendly:
http://en.wikipedia.org/wiki/NP-hard.