Andrew Ng believes that a key way to improve data consistency is to direct a good labeler to use a strategy employed by another good labeler in cases where there are multiple correct ways to categorize data. Then it is important to continue to adjust the rules for the algorithms until consistent results are produced.