5 ms·

> Ant uses advanced machine-learning algorithms and custom programmable chips to crunch huge quantities of user data in a few seconds, to determine whether to g

by vorg 9y ago

> Ant uses advanced machine-learning algorithms and custom programmable chips to crunch huge quantities of user data in a few seconds, to determine whether to grant a customer a loan Is this legal in the U.S. and E.U.? These AI systems can increase profits by saying who should get loans, but not why they should. If a U.S. business can't give clear criteria on how it decides on whether or not to grant loans, and instead allows an AI to decide based on any possible unknown criteria which could include ethnic profiling, then wouldn't they get sued soon enough?

dis-sys 9y ago
Your question is based on the assumption that they do reject loans on regular basis. The truth is they don't. They are so willing to blindly grant loans to the extent that the total amount lent out by them was found to be several times higher than the legal ceiling set by Chinese laws. The authority had to step in and force it to suspend such irresponsible lending. It was widely reported by various Chinese news outlets back in Jan 2018.
jon-wood 9y ago
My assumption here would be that it depends what data you allow the AI to consume. It’s definitely a fuzzy area though, even where you’ve got control over the data being rated on, for example there was recently some noise in the media about car insurance being more expensive if you have a hotmail.com email address.
wjnc 9y ago
A few thoughts (EU/NL): -It's hard to find out why you were denied a loan and full transparency is not required. You could ask for how the company has processed your personal details and try to find out but hard to know and hence hard to complain. Even GDPR doesn't change that a black box model for the modeller is incomprehensible to a customer. -Companies can to a certain degree operate as they want to. (I think that's fine too. Business should benefit both parties.) Discrimination is not always very clearcut and often on proper grounds (think risk-based). -In the Netherlands companies cannot automate denial of service / business. There has to be a human element to that. -Actuaries have long learned not to use for example names in modeling. That would be a good example of adding better risk insights through illegal discrimination (names give a good delineation of ethnical background, at least a for few generations).
Amygaz 9y ago
Totally legal. There is even blog/advertising for it: https://www.techemergence.com/artificial-intelligence-applications-lending-loan-management/ https://www.techemergence.com/artificial-intelligence-applic...
- jsemrau 9y ago
  The information they collect is not available for credit scoring anywhere else but China. While many services exist that offer "social scoring" or a combination thereof (Fico XD score), none of them rank well and offer loan approvals outside a few hundred dollars in working capital loans.
petre 9y ago
There's a book about this called Weapons of Math Destruction by Cathy O'Neil. Yes, it's legal. The more opaque the model and the algorithms are, the more they fuel discrimination.

by vorg 9y ago

dis-sys 9y ago

Your question is based on the assumption that they do reject loans on regular basis. The truth is they don't. They are so willing to blindly grant loans to the extent that the total amount lent out by them was found to be several times higher than the legal ceiling set by Chinese laws. The authority had to step in and force it to suspend such irresponsible lending. It was widely reported by various Chinese news outlets back in Jan 2018.

jon-wood 9y ago

My assumption here would be that it depends what data you allow the AI to consume. It’s definitely a fuzzy area though, even where you’ve got control over the data being rated on, for example there was recently some noise in the media about car insurance being more expensive if you have a hotmail.com email address.

wjnc 9y ago

A few thoughts (EU/NL): -It's hard to find out why you were denied a loan and full transparency is not required. You could ask for how the company has processed your personal details and try to find out but hard to know and hence hard to complain. Even GDPR doesn't change that a black box model for the modeller is incomprehensible to a customer. -Companies can to a certain degree operate as they want to. (I think that's fine too. Business should benefit both parties.) Discrimination is not always very clearcut and often on proper grounds (think risk-based). -In the Netherlands companies cannot automate denial of service / business. There has to be a human element to that. -Actuaries have long learned not to use for example names in modeling. That would be a good example of adding better risk insights through illegal discrimination (names give a good delineation of ethnical background, at least a for few generations).

Amygaz 9y ago

Totally legal. There is even blog/advertising for it: https://www.techemergence.com/artificial-intelligence-applications-lending-loan-management/ https://www.techemergence.com/artificial-intelligence-applic...

jsemrau 9y ago
The information they collect is not available for credit scoring anywhere else but China. While many services exist that offer "social scoring" or a combination thereof (Fico XD score), none of them rank well and offer loan approvals outside a few hundred dollars in working capital loans.

petre 9y ago

There's a book about this called Weapons of Math Destruction by Cathy O'Neil. Yes, it's legal. The more opaque the model and the algorithms are, the more they fuel discrimination.