3 ms·
TLDR: There are two different intuitive definitions of predictive fairness. Given populations with different base crime rates, it is impossible to meet both fai
by danielvf 10y ago
TLDR: There are two different intuitive definitions of predictive fairness. Given populations with different base crime rates, it is impossible to meet both fairness criteria.
If you can survive the WP payroll, the clearest article I've found on this is here:
https://www.washingtonpost.com/news/monkey-cage/wp/2016/10/17/can-an-algorithm-be-racist-our-analysis-is-more-cautious-than-propublicas https://www.washingtonpost.com/news/monkey-cage/wp/2016/10/1...
The full paper is here: https://arxiv.org/pdf/1610.07524v1.pdf https://arxiv.org/pdf/1610.07524v1.pdf
- rahimnathwani 10y agoThe clearest I've seen is this interactive visualisation: https://research.google.com/bigpicture/attacking-discrimination-in-ml/ https://research.google.com/bigpicture/attacking-discriminat...
- danielvf 10y agoThat is great to play with, although it seems to gloss over the fact that you can't have both kinds of fairness.
- rahimnathwani 10y agoIt defines 4 separate objectives (profit maximisation and 3 types of fairness), and shows that they each result in different thresholds. Just click any of the 4 red buttons. It then goes on to advocate a particular kind of fairness.