Summary of Random to Whom? Dice, Tetris, and a lunch bet over what random means

Overview of Random to Whom? Dice, Tetris, and a lunch bet over what random means

Adam Gordon-Bell and Don spend this episode unpacking a deceptively simple question: can computers generate random numbers? What starts as a playful lunch bet turns into a wide-ranging tour through game design, probability, cryptography, historical random-number generators, and the philosophy of randomness itself. The core conclusion is that randomness is not an absolute property of a number, but a relationship between a generator and the observer trying to predict it.

The Big Idea: Randomness Depends on the Observer

The episode keeps returning to one central claim:

• A sequence is “random” if it can’t be predicted better than chance by the party evaluating it.
• That means randomness is not really a property of the number itself.
• Instead, it is a property of the process and the knowledge of the observer.

In other words:

• To a player, a loot box, dice roll, or card shuffle may be random.
• To someone who knows the algorithm or state behind it, it may become predictable.
• So the real question is not “Is this random?” but “Random to whom?”

Experiments and Examples Used to Explore Randomness

Human-generated randomness

Adam starts with a simple test: asking Don to alternate keypresses randomly between two keys.

• A predictor program quickly starts guessing the pattern.
• Don improves a bit over time, but the sequence still isn’t truly random.
• This shows that humans are generally bad at generating unpredictable sequences.

Tetris and player perception

The episode uses classic Tetris as an example of how people often misjudge randomness:

• Random piece generation can create unlucky streaks.
• Players remember the times they desperately needed a long “I” piece and never got one.
• This led to “fairer” systems like the 7-bag randomizer, which is actually less random but feels more fair and playable.

Wolfenstein and Doom

The hosts discuss early game randomness in id Software titles:

• Wolfenstein 3D used a hand-built table of pseudo-random numbers.
• Doom reused the table but changed how it was seeded so replays would be deterministic.
• This demonstrates that many “random” systems in games are really pseudo-random, designed for consistency and reproducibility.

Compression as a randomness test

They also use compression as a practical test:

• Truly random data should be hard to compress.
• Doom’s repeated patterns compress heavily.
• More random-looking data, like digits of pi or computer-generated random output, resists compression.

Historical and Scientific Context

Early attempts at randomness

The episode traces the history of random-number generation:

• Stanislaw Ulam: inspired by solitaire, he helped develop simulation-based thinking.
• John von Neumann: famously argued that there is “no such thing as a random number,” only methods for producing them.
• Middle-square method: a historical algorithm where you square a number and take the middle digits as the next seed. It can look random at first, but often falls into loops.

RAND, books of random numbers, and statistical testing

Before modern computers, randomness was often sourced from physical processes:

• RAND published A Million Random Digits with 100,000 Normal Deviates.
• These were generated from physical processes like roulette-style methods.
• Such tables were used for experiments, sampling, and classification.

RANDU: a famous failure

The episode highlights RANDU, an IBM random-number generator that looked fine until people discovered its outputs lay on a small number of planes in 3D space.

• It appeared random until viewed from the right angle.
• This became a famous example of how something can pass casual inspection yet fail mathematically.

Modern Randomness: Good Enough, or Truly Random?

Python’s PRNG and predictability

The episode demonstrates that even modern pseudo-random generators can be cracked:

• Python’s random module uses MT19937 / Mersenne Twister.
• Once enough outputs are observed, the internal state can be reconstructed.
• So it is random enough for ordinary use, but not secure against prediction.

Physical sources of entropy

They then pivot to sources that seem closer to “real” randomness:

• Dice
• Coin flips
• Card shuffles
• The operating system’s /dev/urandom
• Cloudflare’s lava-lamp entropy wall
• Random.org’s atmospheric noise source

The key insight is that even these are only random relative to what observers can measure and predict. If you had perfect information about every atom and force, many of these would be deterministic in principle.

Main Takeaways

• Randomness is contextual. It depends on the predictor and the amount of information available.
• Pseudo-random is often enough. For games, simulations, and many practical systems, “random enough” is all that matters.
• People are bad at judging randomness. We expect streaks to “look random,” but true randomness often looks biased or unfair to humans.
• Game designers often reduce randomness to improve player experience. Tetris’s 7-bag and Spotify/iPod shuffle changes are examples.
• Security requires stronger randomness than games. A sequence may be random for gameplay but not safe for cryptography or gambling.
• There may be no truly random number in the absolute philosophical sense—only systems that are unpredictable to us.

Conclusion

The episode ends with a qualified win for Adam’s side of the bet: he successfully demonstrates that a computer can produce numbers that are random enough for the intended observer and purpose. But the deeper answer is more nuanced:

• If “random” means impossible for anyone, anywhere, in any possible model to predict, then maybe true randomness is unattainable.
• If “random” means unpredictable to the relevant observer, then computers absolutely can generate random numbers.

The final takeaway is the episode’s title idea: randomness is always “random to whom?”