There is an entire statistical discipline devoted to evaluating algorithms without understanding their internals, or at least with minimal knowledge thereof. It

There is an entire statistical discipline devoted to evaluating algorithms without understanding their internals, or at least with minimal knowledge thereof. It's called "machine learning" (a term you criticized me for using). According to the PP article, race is not an input to the algorithm so it can't give -0.5 to black people solely for being black.