It's under-discussed as with any discussion of an empirical study of ML systems, ie., treating them as targets of analysis. As soon as you do this, they're rev

It's under-discussed as with any discussion of an empirical study of ML systems, ie., treating them as targets of analysis. As soon as you do this, they're revealed to exploit only statistical coincidences and highly fragile heuristics embedded within the data provided. And likewise, pretty universally discriminatory when human data is inovlved