Exploring a meta-rule space isn't hard.

Exploring a meta-rule space and finding rule sets that create results that humans find interesting is much harder.

If you skip that you tend to create systems that seem too random, or too repetitive and obvious.

This is why most algorithmic art is so uninteresting. It falls on one side or the other of the randomness/repetition valley, without hitting that sweet spot that the best human creativity finds.

That's not even thinking about emotional associations, intellectual abstractions, social awareness, and models-of-mind. You might need all of them to make a genius-level creative AI, but I'd guess that the more you have, the more likely you are to get something worth keeping.