Deep Dive·

Decoders of Entropy: How the Oracle Actually Works

Are the I Ching, Tarot, Runes and Geomancy just different 'neural nets' that decode the same randomness? Surprisingly close. We trace the real mechanics — from latent-noise generative models and the free-energy brain to stochastic resonance — and show exactly where numerology and gematria fit (and why they don't).

The question

If you cast the I Ching, draw Tarot, throw Runes, or run a radionics-style scan, you're using the same random source each time and getting a different kind of answer. That invites a sharp intuition:

Aren't the oracle systems basically different neural nets that decode the same entropy — each with its own way of turning randomness into meaning?

That intuition is more right than it sounds. Let's make it precise — and find the one place it needs correcting, which is also the most interesting part.

Every oracle is a codebook

Strip away the mysticism and an oracle is a codebook: a fixed map from raw bits to a structured symbol.

SystemInputCodebookOutput
I Ching6 bitsKing Wen mapping1 of 64 hexagrams
Geomancy4×4 bitsthe 16 figuresJudge + witnesses
Tarota draw78-card deckcards + positions
Runesa drawElder Futhark1 of 24 (+blank)

Information theory says something important here: a channel's bits carry no meaning on their own. As one analysis of Shannon information puts it, mutual information "tells us how many bits can be transmitted, but not which bits, nor what they mean — that is left entirely to the codebook of the observer." Each divination tradition is a 3,000-year-old, human-tuned codebook over the space of human situations. Same entropy in, different symbol out — exactly your intuition.

The machine-learning parallel is real

Here's where it stops being a metaphor. Modern generative neural networks — VAEs, GANs, diffusion models — do precisely this: they take a random latent vector z (literally noise) and their decoder turns it into structured output. "Noise sampled from the normal distribution can be decoded into a high-quality image"; the latent space is shaped so that random points map to valid data (generative models overview).

So the oracle pipeline and a generative model are the same shape of computation:

random latent z  →  [ decoder + codebook ]  →  structured, meaningful output
   quantum bits   →   I Ching / Tarot / …    →   a hexagram / a spread

Different oracle systems are different decoder architectures over the same noise — which is exactly how a quantum-oracle app is built: one entropy engine, many decoders.

The correction: there are two decoders in series

Here's the part the intuition misses. The oracle system is only a shallow decoder — it maps bits to a symbol. The symbol is not yet meaning. The deep decoder, where meaning is actually generated, is you (and now the LLM that interprets the draw):

z → [Decoder 1: the oracle codebook] → symbol → [Decoder 2: your brain + the AI] → meaning

Three well-established sciences describe Decoder 2:

  • Projection. An ambiguous symbol is a Rorschach blot. In projective testing, "the more unstructured the stimulus, the more the test-taker reveals about their own personality" — you supply the meaning by pattern-completing the symbol against your own life (projective tests). Add apophenia, the brain's drive to find patterns in randomness (apophenia), and the Barnum–Forer effect, where general statements feel uncannily personal (Barnum effect), and you've explained most of the felt accuracy.
  • The predictive brain. Karl Friston's free-energy principle says the brain is a generative model that constantly decodes sensory entropy into a structured world-model and minimizes prediction error (free-energy principle). The oracle just hands that model a fresh, intention-framed chunk of structured noise to chew on. The deepest "neural net decoding entropy" in the loop is literally your cortex.
  • Stochastic resonance — the strongest honest "it works." In a nonlinear system, adding the right amount of noise lets a weak, sub-threshold signal cross the detection threshold; this is measured in real neurons (stochastic resonance). Your half-formed hunch about a decision is a sub-threshold signal. A genuinely random draw is calibrated noise that can nudge it into awareness — no psychic content required.

Three families — and where numerology and gematria fit

This decoder lens also sorts the entire zoo of divination cleanly. Methods split into three families by what they take as input (Methods of divination; the classic source Magic and Divination in Islam groups them as sortilege, letter-number interpretation, and astrology):

1. Aleatory — entropy decoders (input = randomness). I Ching, Tarot, Runes, Geomancy, the radionics Scan, dice, bibliomancy, casting lots. This is the family the neural-net-on-noise model fits. (A lovely historical note: lot-casting "was not always divination in the sense of predicting the future, but rather a means of determining a course of action or deciding between courses of action" — a decision aid, which is the honest modern use.)

2. Deterministic — correspondence transforms (input = your own data). This is where numerology and gematria live — and crucially, they are a different computational class. There is no entropy at all:

  • Numerology maps your name and birthdate through a fixed cipher (Pythagorean uses 1–9 by alphabet position; Chaldean uses 1–8 by sound) to a Life Path, Expression, or Soul-Urge number (Pythagorean vs Chaldean).
  • Gematria (Hebrew) and its Greek twin isopsephy assign each letter a fixed value (Aleph = 1 … Tav = 400); a word's value is just the sum, and words of equal value are held to be connected — "love" (ahavah, 13) and "one" (echad, 13) → "love is oneness" (gematria). As Lon Milo DuQuette notes in Llewellyn's Complete Book of Ceremonial Magick, there are dozens of gematria techniques; the Kabbalah literature even maintains a "Sepher Sephiroth," a numerical dictionary.

So numerology and gematria fit the platform — they're still codebooks that map an input to a symbol a human then interprets — but they don't fit the quantum cast, because their input is you, not noise. In engineering terms they're a compute step (deterministic transform), not a draw. Same decoder abstraction, different input.

3. Interpretive — omen reading (input = observed signs). Astrology (birth time/place → chart), palmistry, tasseography (tea leaves), augury. Here you read patterns in given phenomena rather than decode noise or compute from data.

The unifying spine: **every system is a codebook mapping some input to a symbol space that a human

  • AI then decode. They differ only in the input — noise, your data, or observed signs.**

The honest fork

What is the entropy decoding from? Two readings, and your app should never blur them:

  • Mainstream (well-supported). The randomness carries zero information about you. It's a neutral key that unlocks your own latent model — a structured prompt that forces a fresh angle on what you already, partly, know. Every mechanism above lives here.
  • Fringe (contested). Intention subtly biases the entropy, so the symbol is non-randomly meaningful — Jung's "acausal connecting principle," the PEAR lab, the Global Consciousness Project. Unproven, widely doubted, and the only reading in which the noise itself carries signal.

A well-built quantum oracle keeps this seam visible — for instance by labeling whether a given draw came from a true quantum source or a fallback cryptographic one — so you always know which claim is on the table.

So: was the intuition right?

Yes — and it was a good one. Refined, it reads:

Each oracle system is a different shallow decoder/codebook over one shared quantum latent. The deep decoder that generates meaning is your brain plus the LLM. Numerology and gematria are decoders too, but of your own data, not of noise. And whether the latent carries real signal (synchronicity) or is a projective key (cognitive science) is the open question — with every defensible mechanism locating the meaning in the interpreter, not the noise.

Which is also the most useful way to use one. A genuinely unpredictable draw breaks your habitual framing and hands you a symbol you didn't choose. Whether or not a single particle ever felt your intention, that is a real engine for reflection. The randomness is real. The meaning is yours to make.


Sources & further reading

The decoder / ML parallel: Generative models (VAE/GAN/diffusion) · Shannon information & meaning

Why it works (cognitive science): Projective tests · Apophenia · Barnum–Forer effect · Free-energy principle · Stochastic resonance

The families & deterministic systems: Methods of divination · Gematria · Pythagorean vs Chaldean numerology · Lon Milo DuQuette, Llewellyn's Complete Book of Ceremonial Magick · Magic and Divination in Islam (on sortilege) · C. G. Jung, Synchronicity: An Acausal Connecting Principle