Having to know the structure of a hash/map/dict is strictly superior to having to know the methods and fields of an object. For any given task the two are equivalent, except that one is easily modified/reused/passed to other callers without creating an extremely brittle class hierarchy you're only going to have to rearchitect when your requirements change. As an IBM researcher looking at Watson waaaay before Watson actually existed: "OOP requires you to treat all extrinsic information as if it were intrinsic to the object itself." This is tantamount to saying you must have a complete and consistent programme design before you write a single line. Which is crazy (and probably the reason UML exists.) But don't take my word for it, think it through, it's the only conclusion you can reach, I think.