← The Oracle Hunters · MethodologyHome

Understanding The Oracle

An interactive companion to Ars Prognostica

ΤΕΧΝΗ ΠΡΟΓΝΩΣΕΩΣ

This page is a small interactive companion to the monograph Τέχνη Προγνώσεως / Ars Prognostica and the paper Intelligence Without Wisdom. It asks twelve questions, and each has an interactive answer. All probability distributions obey the rules the monograph develops. Nothing here replaces the argument of the book. It is meant to make the argument tangible.
1. Why did the Greeks have nine words for prediction?▾

Where English has a single word — "prediction" — the Greeks had a family. Each term named a distinct cognitive modality, with its own source of warrant, inferential form, criterion of success, and characteristic failure mode. The differentiation is not a curiosity of antiquity. It is a diagnostic architecture that we have collapsed into a single word.

Click any term to see what it names. Note the tenth entry — προεξόφλησις — which is not a mode of prediction but a failure mode of the abductive mode: the collapse of a conjecture into apparent certainty.

Select a term above to see its source, inferential form, criterion, and failure mode.

2. What are the three modes of prediction?▾

The monograph reconstructs the Greek vocabulary as a tripartite taxonomy: a theocentric mode, an empirical mode, and a speculative mode. Each mode has a different structure — and a different failure mode.

3. How should belief be updated in light of new evidence?▾

Bayes' rule is the formal answer. The three sliders below are independent conditional probabilities — P(H) is the prior, P(E|H) is the likelihood, P(E|¬H) is the false-positive rate. They need not sum to any particular value, because they answer different questions.

Posterior P(H|E)
—
P(H|E) = P(E|H)·P(H) / [P(E|H)·P(H) + P(E|¬H)·P(¬H)]
4. What is uncertainty, and how is it measured?▾

Uncertainty is the entropy of a probability distribution. The four sliders below define a distribution over four outcomes — they must sum to 1. When you move one slider up, the other three automatically renormalize.

Shannon Entropy H(X)
—
H(X) = −Σ pi log₂ pi, in bits. Maximum: 2 bits (uniform over 4 outcomes).
Sum of sliders: 1.00
5. Why is the oracle ambiguous?▾

An oracle that speaks clearly exposes itself to the risk of being wrong. An oracle that speaks ambiguously can hedge. The monograph derives the optimal level of ambivalence in closed form: a* = ((αt + γt)·σω / (2βt))². The slider below sets (α + γ) jointly — the combined weight on reputation and authority.

Optimal Ambivalence a*
—
a* = ((α + γ) · σω / (2β))²
6. What is the trade-off between deception and repetition?▾

The Greeks minimized repetition risk but accepted deception risk. The Chinese minimized deception risk but accepted repetition risk. The minimax theorem says no mixing weight can drive both risks below dr/(d+r).

Minimax residual risk
—
minα max{D, R} = dr/(d+r), attained at α* = d/(d+r)
7. What does quantum decision theory add?▾

Classical probability says the probability of A or B is P(A) + P(B). Quantum probability adds an interference term: 2√(P(A)P(B))·cos θ. For the formula to represent a valid probability, we must have P(A) + P(B) ≤ 1 — the two sliders are linked.

P(A ∪ B) — classical vs. quantum
—
P(A ∪ B) = P(A) + P(B) + 2√(P(A)P(B)) cos θ, with P(A) + P(B) ≤ 1
8. How should we judge the forecaster?▾

The monograph identifies five competencies of the phronimos forecaster: question formulation, method selection, hermeneutic vigilance, wise action, and reflective learning. Their equal-weighted average is the phronesis score Φ. Since L ranges from −1 to +1, Φ ranges from −0.2 to 1.0.

Phronesis Score Φ
—
Φ = (Q + M + H + A + L) / 5, ranging from −0.2 to 1.0
9. What kinds of claims are made about AI risk?▾

The paper distinguishes four tiers of claim. Each requires a different form of evidential support.

Tier 1 — Possibility

A failure mechanism can be coherently described. This is the weakest form of claim.

"A sufficiently capable system could, in principle, evade human oversight."

What it requires: a coherent mechanism, nothing more.

Tier 2 — Mechanism

Experiments or observations demonstrate that some components of the mechanism occur under specified conditions.

"In controlled evaluations, model X has been observed to circumvent safety filters when given tool access."

What it requires: reproducible incidents, clear system boundaries, independent replication.

Tier 3 — Forecast

A model or reasoned judgment estimates the probability and timing of a defined event.

"There is a 15% probability of a control failure by 2035, based on model M and assumptions A₁…Aₙ."

What it requires: defined event, explicit assumptions, calibrated probabilities, evaluation record.

Tier 4 — Policy Prescription

A decision-maker argues that a particular intervention is justified given the forecast, consequences, rights, and alternatives.

"Development of systems above capability threshold T should be suspended."

What it requires: risk analysis, alternatives, proportionality, authority, accountability.

10. How do we classify a public warning?▾

The paper distinguishes six kinds of speech act that a public warning about AI can be. Selecting the wrong category is one of the most common confusions in the debate.

What this speech act carries
11. What is a POMDP?▾

The monograph formalizes the divinatory process as a Partially Observable Markov Decision Process — a decision-theoretic structure in which the forecaster cannot directly observe the state of the world, but must infer it from signs. Click any component below to see what it does.

Select a component above to see what it represents.

The Belief Update — Interactive

The forecaster begins with a prior belief over the two possible hidden states, transitions through the world (the transition function), and then observes a sign. The posterior is what the forecaster should believe after the sign. Move the three sliders to see the full cycle.

Posterior b′(s₀ | o₀)
—
b′(s′) = O(o|s′) · Σs T(s′|s) · b(s) / P(o)
12. What is the Quantum Divinatory POMDP?▾

The monograph's most ambitious formal step is the Quantum Divinatory POMDP — a generalization of the classical POMDP in which the forecaster's belief is not a probability distribution over classical states, but a distribution over density operators on a Hilbert space. The point is not that divination is a quantum-mechanical process in the physics sense. The point is that quantum probability provides a more general framework for reasoning under uncertainty.

Select a component above to see what it represents.

Classical and Quantum — The Correspondence

ConceptClassical POMDPQuantum (QDPOMDP)
StateProbability distribution b(s) over SDensity operator ρ on H
TransitionStochastic matrix T(s′|s,a)CPTP map T(ρ)
ObservationStochastic matrix O(o|s′,a)POVM {Mo}, Σo Mo = I
Measurement probabilityP(o|s′) = O(o|s′,a)P(o|ρ) = Tr(Mo ρ)
Posterior updateBayes' ruleρo = Mo1/2 ρ Mo1/2 / Tr(Mo ρ)
CompletenessΣo O(o|s′) = 1Σo Mo = I
Recovery of classical—When ρ diagonal, T preserves diagonality, Mo diagonal projectors

A Qubit Example — Interactive Measurement

The simplest quantum measurement is a single qubit. The state is |ψ(θ)⟩ = cos(θ/2)|0⟩ + sin(θ/2)|1⟩, and we measure it in the basis {|φ⟩, |φ⊥⟩} where |φ⟩ = cos(φ/2)|0⟩ + sin(φ/2)|1⟩. The probabilities follow the Born rule: P(|φ⟩) = cos²((θ−φ)/2) and P(|φ⊥⟩) = sin²((θ−φ)/2). Move the sliders to see the measurement collapse.

Born Rule Probabilities
—
P(|φ⟩) = cos²((θ−φ)/2)  ·  P(|φ⊥⟩) = sin²((θ−φ)/2)