Just found an interesting google blog on equal opportunity for machine learning.

Some thoughts to head off discussions. So it would appear to me that an oracle model can never be a planetist. (Now that we are going to Mars and all, let’s reframe the problem in terms of planetist and planetism–discrimination based on planet of association)

The fact is that if a model is predicting loan defaults with 100% accuracy, then it cannot be a planetist. The challenge to that will surely be that we do not believe it is an oracle. How does the oracle know if a person will or will not default on a loan? It might be a planetist all along? We will never know know because 0 is not large enough sample to be representative when you don’t allow any Marsians to get loans.🖖🏻