The gender biased #AppleCard credit decision is a fascinating example of how algorithmic credit decisions can enter choppy waters
Even more fascinating is Goldman’s response. They can’t identify the reason there was such a difference in credit limits but they are sure that decisions are not based on “factors like gender”
In fact, they have no way of guaranteeing that. The fact they exclude protected characteristics or “do not know your gender or marital status” does not mean these factors were not used. These are #AIFalseDreams
Want to know more about #AIFalseDreams? Check out my job talk paper- scholar.harvard.edu/gillis/job-tal…
I even have a section in which I demonstrate the ability to predict “marital status” from fairly innocuous variables. I’m sure #AppleCard can do a much better job with their data
Last thought- I would love to see proof that #AppleCard prices on creditworthiness alone. My guess- this is about being able to charge women more for credit, unrelated to creditworthiness. #AI #Algorithm #MachineLearning #Bias
I also recently discussed this on #DeepBackgroundPod
