Random vs. systematic error

Hi there,

Andrew mentioned in lecture that DL algorithms are robust against random errors, but not systematic errors. What is the reason behind that? How can you explain it in statistics?

Thanks!