the problem is 'bias' is often based on a model that is statistically and operationally accurate, but unfair from a higher order social moral framework. a suff

the problem is 'bias' is often based on a model that is statistically and operationally accurate, but unfair from a higher order social moral framework. a sufficiently layered net will pick up second and third order consequential effects that are not formally protected by anti-bias laws, but end up effectively mapping onto the sorts of people in categories the laws were intended to protect. that's the fundamental problem as I see it, and it's necessarily complicated and nuanced.