| I heard or listened to an anecdote about how credit reporting would actually be a perfect application for ML, but ML companies are fearful of stepping into a minefield of "you can't do this" regulations. If an algorithm found some correlation between your "credit worthiness" (however their proprietary IP chooses to define it) and race or color of skin (or some proxy of), then using that information would be (legally) discrimination, say most legal experts speculating on a hypothetical (but bound to happen eventually) case. The fact is, yes, ML could probably discover a bunch of novel correlations that increase effectiveness by 7.8%, but since we can't really describe what "model" or way of thinking about our world the deep neural net is using, it's too easy to build a racist and discriminatory credit report. Kinda like that microsoft twitter chatbot that had to be taken down because it was learning from (and repeating) the racist garbage that a bunch of trolls were feeding it. Point is, it's a minefield. But it's going to happen. Someone's going to try it. And that's what the dark side of these companies like CoreLogic is: when real people's life events are influenced by the outcome of these databases, what's a societally-acceptable error rate? And what mechanisms are in place to check it? Not saying CoreLogic is using racist ML credit bots. Not even saying the problem here is racism. I'm saying we've decided to put rules in front of credit reporting companies because, even if it's unintentional, it can lead to outcomes we all decided "you know, we have these ideals we're trying to strive towards and this isn't it." You also don't need a conscious conspiracy to output a racistbot -- microsoft didn't set out to make one, they just didn't think to build in a troll filter. So yeah, CoreLogic isn't scamming renters out of such small amounts, but they are benefitting from an environment that allows such scams to exist. Put another way, if there was some pending legislation that would make it harder for these scams to work (eg CoreLogic being legally obligated to also publicly report how many inquiries an inquirer has made with CoreLogic for the purpose of this specific property), do you think CoreLogic would spend lobbying money protecting their interests and preventing this at the expense of the consumer? Of course they would, reporting regulations are a huge operating expense! (wait... aren't you a company specializing in reporting...?) That's why the author made a few swipes at the less-regulated corner of the credit report industry. |