Which of these are reasons that it's often unrealistic to expect an ML system to be $100\%$ accurate?

You might not have enough data

Data can be mislabeled

Data can be ambiguous

All of the above.