Open questions
When does manual engineering fail?
How do you know learning was useful?
What comprises evidence of generality?
What if you can't understand resulting hypothesis?