Future of coding should also means more equity. I came accross a paper earlier mentionning that latest NLP algo intensify gender bias in Information Retrival system (source: https://arxiv.org/pdf/2005.00372.pdf). Basically, the more performant is the algo the more it is able to infuse the bias of the training dataset. When it comes to NLP to fight gender bias, there are some technics like gender swapping,biais fine tuning or learning gender-neutral embedding exist to mitigate the bias (futher reading - https://arxiv.org/pdf/1906.08976) ⬇️ I attached a screen shot of my browser (Ecosia) when looking for 'doctor' in images

Hi Maeliza it's pretty disheartening that such biased(/racist) search results are still evidently the total norm. quick check of google which while better on a gender basis, is still very racially biased... Repost from last month but so relevant to this topic, and again sad how little has changed since it's publishing... https://en.wikipedia.org/wiki/Algorithms_of_Oppression

Yes that's true. We should go for a chane one dataset at the time. I know that Google implemented the gender swapping technic to mitigate gender bias at scale in translation. For the race bias they tag the identified 'race' to provide more balanced output of images request

🙂 https://futureofcoding.slack.com/archives/C5T9GPWFL/p1591498636036100

Thanks !