Yeah, in a sociological level I totally understand how being discriminated for being black is more serious than being discriminated for something unlikely like

Yeah, in a sociological level I totally understand how being discriminated for being black is more serious than being discriminated for something unlikely like being a redhead, but a major thing at play here is machine learning, should the engineers add code to the system in order for it to skip looking skins colors? What if skipping skin color is not enough and the machine "bypasses" such restrictions by detecting the shape of the nose (e.g. black people have slightly different noses than white people); at the end of the day what the machine is doing is just statistics and if you poke too many holes on it -for moral reasons or else- you may as well don't use statistics at all.