Much of machine learning is training on category variables and then using them to make predictions about individuals.

There aren't many ML algorithms trained just on single individuals. https://twitter.com/pmddomingos/status/1353096987399213056

In some Machine Learning algorithms categories like "race" and "sex" are a way to smooth over societal abstractions and wrap them up in a single variable that within context can be (sadly) highly predictive.

ML entrenches demographic effects, not erases them.

Given some responses I get, I'm not only a fan of machine learning, but a practitioner and advocate. ML can be excellent and very useful, but we should always keep our eyes wide open about what it does and doesn't do objectively.