7 Sources of Randomness


The modern computer age runs on random numbers. Patternless strings of digits are essential for the data encryption that promises privacy and security online. And since computers—fundamentally deterministic machines designed to follow set procedures—cannot deliver true randomness, we must source our entropy from the physical world.

A random number generator (RNG) that taps into the inherent randomness of a physical process typically consists of (1) a transducer to convert some aspect of the physical phenomenon into an electrical signal; (2) an amplifier to boost the amplitude of random fluctuations to a measurable level; and (3) an analog-to-digital converter. Here’s a sampling of the real-world sources of randomness we’ve exploited over the years.


First a nod to a low-tech RNG: dice! Small throwable objects with multiple resting positions have been generating random numbers at least since 2500 BCE, when Mesopotamians playing the Royal Game of Ur tossed tetrahedrons. The ancient Egyptians and Indians also enjoyed dicing, as did the Romans. Impressive as this Roman 2nd century icosahedral (20-sided) die is, however, we can do six times better now. Need a random number between 1 and 120? Anyone? 

As long as they're not loaded and nothing in the environment (or means of tossing) favors certain outcomes over others, dice are a reliable way to produce mostly random digits. The going is slow, though.


To fuel its postwar appetite for random numbers, the RAND Corporation needed more than dice—120-sided or otherwise. In 1947, engineers devised an electronic simulation of a roulette wheel, which they connected to an early computer. The setup churned out numbers at a rate of about one per second, eventually producing enough to fill—after filtering, processing, and testing—RAND’s 1955 publication A Million Random Digits with 100,000 Normal Deviates. Though the book’s contents were primarily useful in statistics and experimental design, its title seems to have flummoxed the New York Public Library, which reportedly indexed the random number table under the “Psychology” heading. The tongue-in-cheek Amazon reviews of the 2001 reissue are also good for a laugh.


A caesium- or cesium-137 nucleus can, via a process called beta decay, become a barium-137 nucleus, liberating an electron as it does so. And the laws of quantum mechanics decree that there’s no way to tell when a given nucleus of caesium-137 will decay; no way to tell, given a collection of caesium-137 nuclei, when the next individual atom in the group will decay; and thus no way to tell how the intervals between consecutive decays will compare. Autodesk co-founder John Walker harnessed this quantum randomness to create HotBits, an online resource that furnishes users with “genuine random numbers” by measuring a pair of intervals between caesium-137 decays and emitting a zero or one bit based on the relative length of the two intervals. 


In 1996, Landon Noll, Robert Mende, and Sanjeev Sisodiya of Silicon Graphics, Inc. filed a patent (US 5732138) for a “method for seeding a pseudo-random number generator with a cryptographic hash of a digitization of a chaotic system.” The chaotic system in question? A LAVA LITE, its blobs of colored wax set in unpredictable motion by the heat of the incandescent bulb in its conical base. Dubbed lavarand, the patented system used a digital photograph of a lava lamp to generate a 140-byte seed for a pseudo-random number generator. The lavarand website has been inactive since 2001, its archived version sadly devoid of trippy imagery.


In 1997, Mads Haahr and some friends walked into a Radio Shack and told the sales guy they needed the cheapest radio he had. They wanted their computer to listen to static, they explained. Haahr et al. had decided to source entropy for random number generation from a radio picking up atmospheric noise. Atmospheric noise is radio noise caused by natural atmospheric processes, primarily lightning discharges in thunderstorms. They needed the cheapest radio available because many units have noise filters that only permit users to tune to frequencies that stations are using to broadcast.

Almost 20 years later, Haahr’s Random.org still relies on atmospheric noise to advance its “mission ... to produce the highest quality true random numbers and make them available to the world in useful forms.” Website visitors use Random.org’s numbers to hold drawings, to drive online games, and for lotteries, sweepstakes, and scientific applications.

Some argue, it’s worth noting, that only quantum phenomena—that beta decay in #3 above, for instance—are truly nondeterministic. Proponents of RNGs that rely on physical phenomena without quantum-random properties (atmospheric noise, say, or lava lamps) counter that these phenomena are complex and chaotic enough to make it infeasible for humans to forecast their behavior. Randomness tests can also be performed to certify the output of these RNGs.


The lavarand operation (see #4 above) went dark in the early aughts because Landon Noll and a new collaborator, Simon Cooper, had invented an improved RNG: LavaRnd. Instead of lava lamps, LavaRnd uses a webcam with the lens cap on as the source of entropy. The thermal noise emitted by the webcam is digitized and stripped of any unwanted predictability. Unlike lavarand, LavaRnd is patent-free, open-source, and in the public domain. As Noll told WIRED in 2003, “We’re trying to give people the ability to generate random numbers themselves.”


In 2015, You-Qi Nie and colleagues at China’s Hefei National Laboratory for Physical Sciences announced that they’d devised a quantum RNG capable of yielding 68 billion random bits per second. 

Let that big number sink in.

This when the fastest commercially available quantum RNGs could only produce a million bits per second. These generators work by sending a stream of photons through a beam splitter with a 50-50 chance of transmission and reflection. The series of transmissions and reflections is then translated into a string of 0’s and 1’s. Single photon detectors can only detect so fast, however, and the equipment’s limitations cap the speed of bit production.

To achieve their record-shattering rate, the Chinese physicists operate their laser at its threshold level. This enables them to measure photons generated by spontaneous emission, an entirely random quantum process. An interferometer converts fluctuations in the phase of these photons to intensity changes, which are then measured by photodetectors. And since photodetectors work much faster than those slowpoke single photon detectors, voila! As the MIT Technology Review put it, “Organizations that need a practical system that offers secrecy guaranteed by the laws of quantum physics may not have much longer to wait.”