Although I'm pretty sure there isn't a (is male)? give 10x function in the algorithm, I wonder if this was a case of putting certain metrics into the algorithm that highly correlate to gender.
I work in financial services as a statistical model developer and we usually go through extensive legal review before we can use a model for a specific application. We always avoid using less controversial demographics (e.g. customer tenure as a surrogate for age) before we build models for less legally risky endeavors like marketing because of Fair Lending laws. It concerns me that gender made it into a credit risk model which have an even higher standard for legal oversight where I'm at. Fingers crossed that this is only a rare oversight on Goldman Sachs part.
3 comments
[ 12.9 ms ] story [ 284 ms ] thread