What would you do if you don't find solution to friendliness problem while it would be clear that strong AI is within one year? What is the second best option after trying to develop AI friendliness theory?