Further, proxies for protected classes might be hiding undetected within other factors used in machine learning models Despite the complexity, these factors might unintentionally be too coarse to accurately capture the qualifications of members of protected classes