This is actually a big problem with deep machine learning based algorithms: they're guaranteed to be discriminatory but there is no obvious mechanism for making

This is actually a big problem with deep machine learning based algorithms: they're guaranteed to be discriminatory but there is no obvious mechanism for making them obey laws that prohibit discrimination (which often means they get away with it based on a technicality). In a perfect world this wouldn't be a problem because you'd just fix the root issues (i.e. eliminate poverty and crime in ethnic groups, fill leading roles in tech with women, etc etc). In the real world, things are not that straightforward (especially in the US where a lot more data can be collected and used to feed the algorithms than elsewhere).