The Probability-Cipher Unity — When Cardano's Dice Became Shannon's Entropy

A gambler counting dice in the 1560s and a Bell Labs engineer proving secrecy in 1949 were solving the same problem: how much ignorance remains after you see a signal. Gerolamo Cardano wrote Liber de Ludo Aleae before probability had a stable language; Claude Shannon gave that language a unit, the bit. The distance between them is not 389 years of trivia. It is the path from gambling odds to unbreakable encryption.

The case

Cardano's move was blunt: count favorable outcomes, count total outcomes, divide. A fair die gives a 1 in 6 chance of any face. Two dice give 36 ordered outcomes. Once you can count uncertainty, you can reason about games without superstition.

Shannon's 1949 paper, "Communication Theory of Secrecy Systems," makes the cipher version exact. A cryptosystem has messages, keys, and ciphertexts. Perfect secrecy means seeing the ciphertext does not change the odds of any plaintext:

P(M = m | C = c) = P(M = m)

That is a probability statement, not a vibe about difficulty. The one-time pad works because each message-length random key maps a plaintext to a ciphertext with no statistical preference left over. If the key is truly random, used once, and at least as long as the message, the attacker learns zero bits about the message from the ciphertext alone.

The same counting instinct appears in physics. Boltzmann's entropy counts microscopic arrangements compatible with the same visible state. Shannon's entropy counts possible messages compatible with the same received signal. Cryptographic entropy counts possible keys compatible with the same ciphertext.

Where it shows up

Object	Date	Randomness question	What leaks
Fair die	1560s Cardano draft	Which face appears?	Biased dice
Pascal-Fermat letters	1654	How should an unfinished game be split?	Bad odds
One-time pad	1917 patent lineage; Shannon proof in 1949	Which key was used?	Reused key
Enigma traffic	1930s-1945	Which machine setting produced this text?	Repeated formats
Voynich tests	1912 rediscovery; modern statistics	Language, cipher, or generated text?	Word patterns

Enigma is the clean warning. Its key space was large, but German traffic often carried repeated structure: weather reports, predictable headers, operator habits. Shannon's term for this is redundancy. Natural language is not random; that is why compression works and why cryptanalysis has a grip.

The Voynich Manuscript sits near the same fault line. It has about 240 vellum pages and roughly 38,000 word-like tokens, but its script still has no accepted reading. Michael A. Greshko's Naibbe cipher, published in Cryptologia in 2025, uses dice and playing cards to produce Voynich-like text from Latin or Italian. It does not solve the manuscript. It shows that a hand-doable cipher can imitate some of the manuscript's statistics.

What's contested

Shannon's theorem is settled under its assumptions. The open fight is over the source of randomness. Physical dice, card shuffles, radio noise, thermal noise, and deterministic pseudorandom generators do not fail in the same way. A cipher can be perfect on paper and broken in practice by a weak random source.

The Voynich side is less settled. Naibbe is a model, not a decipherment. Gordon Rugg's 2004 Cardan-grille work and later statistical studies show that generated pseudo-text can mimic parts of Voynichese, while other analyses argue the manuscript has structure that simple generators miss. The live question is not "is it weird?" The live question is which statistical signatures survive every proposed generator.

Why this has to do with other realms

This page belongs in cryptography, history, and physics because entropy is a portable object. In concept thermodynamic entropy, a gas hides microstates behind pressure and temperature. In concept information theory, a message hides alternatives behind a signal. In concept one time pad, a key hides plaintext behind ciphertext.

That bridge matters for concept voynich manuscript because undeciphered writing is not only a linguistic puzzle. It is an entropy audit. The analyst asks whether the marks behave like language, like encrypted language, or like a generator built to fool the eye.

An open question

If a 15th-century scribe had dice, cards, vellum, and patience, what test would separate meaningful ciphertext from generated nonsense without assuming the answer first?

Key sources

Gerolamo Cardano, Liber de Ludo Aleae (written c. 1560s, published 1663) - early systematic treatment of dice probability.
Blaise Pascal and Pierre de Fermat, correspondence on the problem of points (1654) - gambling as the seed of formal probability.
Claude E. Shannon, "A Mathematical Theory of Communication" (1948), Bell System Technical Journal - entropy enters communication theory.
Claude E. Shannon, "Communication Theory of Secrecy Systems" (1949), Bell System Technical Journal - perfect secrecy as a probability condition.
Michael A. Greshko, "The Naibbe cipher: a substitution cipher that encrypts Latin and Italian as Voynich Manuscript-like ciphertext" (2025), Cryptologia - to verify: final issue and page details.
Gordon Rugg, "An elegant hoax? A possible solution to the Voynich manuscript" (2004), Cryptologia - generated-text benchmark for Voynich claims.

Abhishek's take

What grabs me here is that probability did not begin as pure math. It began as a way to stop being fooled by dice, then became a way to measure secrecy, heat, compression, and language. I trust this kind of idea more when it survives that many costume changes. The page I want next is simple: when does randomness become a tool, and when does it become an alibi?

Tags: #probability #cryptography #entropy #shannon #cardano #randomness #voynich

Abhishek