Direct link to the page on the theory.

That's really interesting! (ha!) I recommend reading the full page for good examples, but here's a summary:

Apart from external reward, how much fun can a subjective observer extract from some sequence of actions and observations? His intrinsic fun is the difference between how many resources (bits & time) he needs to encode the data before and after learning. A separate reinforcement learner maximizes expected fun by finding or creating data that is better compressible in some yet unknown but learnable way, such as jokes, songs, paintings, or scientific observations obeying novel, unpublished laws.