Can you fool a computer at generating random numbers?

I'd like to see people talking about the DieHard tests, and potential problems with it. Or about the needs for different types of randomness for different applications - modelling requires large quantities of random numbers but needs to be able to recreate the same sequence later. That would be very bad for cryptography. Games need unpredictability, but true randomness isn't great. Music shuffle systems are often made less random to prevent jarring changes.

Spooky huh?

--

Am I wrong in thinking that this will only work with at best a random coin flip precision? Not to mention that the few flips people will be willing to make means you lack any significant amount of data to make such a statement.