There will be bias in the data, but that can be looked at in one place, and corrected once. At which point everyone gets the benefit. This means errors are foun

There will be bias in the data, but that can be looked at in one place, and corrected once. At which point everyone gets the benefit. This means errors are found fixed and "released" broadly and quickly. This is much better than trying to disseminate training and information to judges who may or may not actually absorb it. For example new information on bias could be encoded in an algo but is possibly not teachable...because people are biased.