Please do write the article--if you continue to believe your claims--and please include sufficient technical detail so that people who work with machine learning can verify that the measures are at least plausible.

The reason this is important is that one of the biggest challenges in ML has been to be able to tune them to get the style of output that we think is socially desirable. This includes desiring patriarchy.

It is hard to minimize error against optimal patriarchy. It's much easier to minimize click-prediction error, clap-count prediction error, duration of engagement prediction error, and stuff like that. The numbers are all just sitting right there. Easy mode.

It's not impossible that Medium would have decided to make their jobs a lot harder in order to make themselves somewhat less money and be somewhat less popular, because it is possible to curate training sets until you get the kind of bias you want in the outcome, but it's a lot of work.