A good learning algorithm should be able to handle noise and uncertainty in the data effectively.
"The best way to learn is to do it, to make mistakes, and to learn from them."