You don't spend two years doing a Game only to realise later that is not fun

That does happen on occasion, the commonly-cited example being Half-Life. How awesome would it have been if the Valve team hadn't had to waste so much time, money, and personal energy on their initial failed prototype?

Unfortunately most studios ship their failures, either because they don't realize they built something crappy or because the alternative is bankruptcy. A cynic would say that if AI can reduce the cost of experimentation, it will only result in more bad games, while an optimist would argue that it will result in more good games. I think we'll find that they're both right.