A Data Science Central Community
At least this is IEEE's viewpoint, if you read their recent article (front page) published in IEEE spectrum, September 2011. In a nutshell, they claim that hardware technology using Intel chips is better than software architecture to generate / simulate random numbers. See one of their prior articles at http://spectrum.ieee.org/computing/hardware/intel-makes-a-digital-c....
I strongly disagree with this point of view. I believe that IEEE has only investigated "stupid" random generators available in Excel and some programming languages, and concluded, based on these highly flawed random generators with well known biases, that mathematical random generators are poor.
In my opinion, you can't beat a random digit generator where the k-th digit generated by your algorithm is the k-th digit of a number of the form a + b + c +... where a, b, c are well chosen transcendental numbers that are linearly independent over the set of rational numbers. Take for instance a = log(63), b = cos(5,678), c = PI +SQRT(58), and you get a great random generator! Of course, if you can get transcendental numbers that can be approximated very fast (e.g. PI), you not only get a great random generator, but one that can be implemented very efficiently, and beat random generators that are dependent on hardware. I agree that the seed for your generator should be either hardware-dependent, or better, derived from an external source of randomness (e.g. a mix of digits extracted from 10,000 stock prices, extracted in real time).