Timothy C. May wrote:
At 5:07 PM -0600 12/11/96, Igor Chudov @ home wrote:
Correlation is not an evidence of discrimination, at least to me.
Nor is it to me. So neither of us will likely object to the neural net-based lending programs which feed in a bunch of applicant data points, train the net by providing feedback on who repaid their loans and with what complications, etc. Even if such nets end up rejecting "otherwise-qualified" (a la your other post) applicants in such a way that the accept/reject ratios appear strongly correlated with certain ethnicities?
(Another member of the list sent me private e-mail about his experiences writing a "scoring program" for a bank making just the kinds of loans we're talking about here. He recounted his bank's very real experiences with loan paybacks by various ethnic and national groups. Nothing very surprising, to me.)
My readings on neural nets made an impression that they are not necessarily good at all. - Igor.