Hacker News Reader

IIRC Amazon tried something like this when feeding a neural net with resumes; but the neural net still managed to pick up the sexist hiring process by learning what kinds of activities were typical "girl" or "boy" activities. Humans with experience could essentially do the same thing (even subconsciously).

I'm sure that kind of redacting names &c would help, but it wouldn't be perfect.

https://www.reuters.com/article/us-amazon-com-jobs-automatio...

View on HN

> (...) but the neural net still managed to pick up the sexist hiring process by learning what kinds of activities were typical "girl" or "boy" activities.

I'm not sure that's a valid take.

It's my understanding that Amazon calibrated their machine learning model to answer the question of "help me find more people like the one we already have", and they just so happened to have within their ranks way more people with a specific academic and professional background, which was happened to refer to be men who did activities dominated by men.

Reply on news.ycombinator.com