A good learning algorithm should be able to handle noise and uncertainty in the data effectively.
"You don't have to understand everything to love it."