Say you took a small disk shaped object like a hockey puck with a window on it and you filled it with sand. 50% white sand and 50% black sand. Inside the puck would be blades that are attached to a motor and rotated slowly to constantly change the pattern. The pattern formed in the window would be truly random wouldn't it? You could mount this to a PCIE card with a camera...
And they are less cool than avalanche noise, when you're asking nerds like us.
IMHO, one of the best devices to get low amounts of randomness is still the dice (cube). Casino-grade, if you're anal about it. I use that to throw Diceware passphrases.
I'd certainly dispute the "memorable" part for most people on Earth.
You could use this on $LANGUAGE of your choice, or $CORPUS of your choice.
(Also, it seems much more memorable to me than an equivalent random stream of bits are.)
Still, neither "wrowd", nor "aut", nor "settea" looks or sounds English-like to me. I wouldn't even begin to know how to pronounce it.
It's nice in a sense, sure, but I suppose I could even remember (real) French words better, despite all those accents and me not speaking French.
"aut": a logarithmic unit of measurement indicating the level of automation of a process. "I think we can get another couple of auts out of the sentiment-analysis pipeline."
"settea": a genus of shrubs in the family Settaceae.
I frown on casually lying to people over the internet. The term 'OK' originally stood for 'all correct'.
http://www.etymonline.com/index.php?term=OK
http://www.straightdope.com/columns/read/503/what-does-ok-st...
You're trying to allude to NOKD, which is (1) completely unrelated, and (2) not lost.
The uncertainty principle isn't the aspect of quantum mechanics that is used in these RNGs. They only need the fact that they can perform an action (measurement of some observable) whose outcome is not predictable. They do not attempt to measure different observables, since this would not help produce randomness. It is in essence just a (theoretically ideal) dice roll.
This is one interpretation of QM. There are many theories explaining eigenbasis collapse (the "random-looking" phenomenon), and some of them are deterministic.
The important thing here is that the data stream is hard or impossible to predict. It doesn't have to be truly random, because we don't even know if "truly random" is a sensible thing to ask for. What we're actually hoping is that you can't predict the result of a measurement, whether or not it's because of randomness or something else.
Most RNGs also don't rely on the uncertainty principle (or, more generally, non-commutative measurements), but instead on e.g. radioactive decay. This is an eigenbasis collapse mechanism (with the two eigenstates being "this thing hasn't decayed" or "this thing has decayed"), but it's not based on uncertainty. This is easier to build physically.
Basically, it removes the possible attack vector of replacing any command-and-control circuitry with your own, suborned, components.
[0] - https://www.llnl.gov/news/lawrence-livermore-scientist-devel...
Also, it integrates very well. The smaller the diode, the better randomness you'll get.
But, yes, with enough precision a resistor would do.
For example, perhaps the black grains have a slightly different density, so they might tend to congregate more on the upper or lower side of the window, ever so slightly.
In general, natural processes seldom produce perfectly unbiased streams of independent coin flips. There's usually some post processing.
If two such devices were identical in every way and started from the same conditions (i.e: the sands were placed in the devices the exact same way) then over a sufficient amount of time the sand mixture in the two devices would reach an identical equilibrium state of mixing. At least that's how two gases would act. No idea about sand.
Your argument basically boils down to "chaos theory is wrong, because I /can/ know everything".
Over time the sand could (perhaps) grind itself into flour and not exude the same mechanical properties anymore.
Finally, even if that worked, it might be good for what, 2,3, maybe 5 samples per second. If you take too many pictures before the sand moves, it's no longer random. What if I need millions of random numbers ?