Could you point us to that paper?
Somehow I can't believe your theory that you can generate trully random set in deterministic system.
Sure, you can do the opposite. Just pass output from a noisy (but with limited max noise level) line through a schmitt trigger that can handle this noise and you beautifully remove all the randomness from the data. But if you know all inputs and all processes within a system exactly, you can predict the output.

The point I disagreed with is:
"Thus if the user has access to a true/pseudo random number, why use the RNG at all !!"
My answer: Because of efficiency.