I think the scientific method is to learning the rules to chess through observation as religion is to arguing over what happens when you land on Free Parking.
Doug Hofstadter once used "Bongard Problems" to make the same point. Each problem is made of two sets of 6 cards with simple patterns drawn on the cards. The solver must figure out the very simple idea which separates the sets.
For instance, set I might only have shapes made from of three line segments, while set II has shapes made from four line segments.
Overgeneralizing, which Feyman mentioned, is just one of the challenges the cards make explicit. Another is the need to simultaneously see the patterns at multiple levels of abstraction. It's hard to explain, but I hope anyone who's interested in this sort of thing would check them out.