She had been doing that job for awhile. IMO the reason this particular paper got the treatment it did is subtle but simple, based on the following observations:
* Many jurisdictions impose strict controls on the use of racially biased methods (eg: redline laws like the fair housing act in the US).
* Different advertising companies have access to different targeting methods (eg facebook has your social graph, apple have your app use, google have your search history).
* Language models are a key technology for google to maintain advertising relevance & thus keep adwords competitive
"Models trained on modern language reflect societal biases" may seem like an obvious fact, but once google says it publicly they no longer have a basis to deny it in court.
* Many jurisdictions impose strict controls on the use of racially biased methods (eg: redline laws like the fair housing act in the US).
* Different advertising companies have access to different targeting methods (eg facebook has your social graph, apple have your app use, google have your search history).
* Language models are a key technology for google to maintain advertising relevance & thus keep adwords competitive
"Models trained on modern language reflect societal biases" may seem like an obvious fact, but once google says it publicly they no longer have a basis to deny it in court.