Optimal learning does not mean getting 100% correct all the time