[extropy-chat] AI: credit reporting

Kevin Freels kevinfreels at hotmail.com
Wed Apr 7 18:46:45 UTC 2004


As a mortgage broker by day, I read hundreds of credit reports every month. What has always impressed me was how accurate the scoring is when you are using it to calculate risk. There are people out there that you would never suspect as being a credit risk, but the algorithms used to calculate their scores say otherwise. Then, sure enough, two years later they are being foreclosed on!

So many things are considered in credit reports: Proprotions of balances to limits, length of history of tradelines, number of tradeline, tradeline types, length of credit history, and I think even employment and address changes are calculated into the mix. The end result is amazingly accurate.

Because of this accuracy, insurance companies and employers are starting to use these scores as a basis of determining a person's value to their companies. Insurance rates are getting adjusted based on credit scores because lower score people are more likely to file a claim. It seems to me that it is no longer a "credit" score, but a "decency score". 


What is even more amazing though is how occasionally "it" can be "fooled". "It" can also make "mistakes" on occasion by giving a person a low score (indicating higher risk) even though a review of the data shows no signs of why this score was reached.Each of the three main credit reporting agencies usually render different scores. Each one employs their own proprietary algorithm for calculating thses scores. Each company has a huge investment in these algorithms and as such, will probably never release them to the public for scrutiny. 

What I am wondering is whether or not these algorithms could represent the first baby steps towards AI. Are we not already using it as such? How involved do you think these algorithms are? 

Kevin Freels
PS What is the Atheist version of Easter?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20040407/f4dced0f/attachment.html>


More information about the extropy-chat mailing list