Data mining without prejudice

Interesting article from MIT about a paywalled article in Science about a new technique for data mining developed at MIT, the upshot being apparently that it's doing curve fitting with no preconceived notions of the variables being fit. Or something. I need to read it after some sleep.