Skip to content

SAPIENCE.md

Latest commit

 

History

History
169 lines (132 loc) · 7.71 KB

SAPIENCE.md

File metadata and controls

169 lines (132 loc) · 7.71 KB
Raw

Quantifying AI Sapience as an Integer for Licensing

This document is the formal specification of the sapience number used by the Sapience Testing App described in README.md. It is written for licensing, legal, and business readers: the goal is a sapience measure that is constrained — an integer, bounded, reproducible, and auditable — so that a license agreement can reference it unambiguously.

1. Why sapience must be constrained

A license price must rest on a quantity that parties can measure, record, and re-check. Qualities such as "intelligence" or "personhood" are not contract terms. What a contract needs is:

  • A number, not a judgment call.
  • A fixed type: an integer, never a fraction, percentage, or adjective.
  • Bounds: every session yields a value inside a known range.
  • Reproducibility: two certified sessions of the same AI under the same probe suite must agree within a stated tolerance.
  • Auditability: the counts behind the number are observable during the test — the harness announces "Calculation" or "Recognition" aloud.

The sapience number defined below satisfies all five requirements.

2. Definitions

Term Definition
Vendor AI The artificial intelligence brought by a vendor to be tested and licensed.
Probe One standardized stimulus presented to the vendor AI by the harness.
Detector A perceptron in the harness that watches the vendor AI act and fires on exactly one event class.
Calculation event A detection of the vendor AI performing a calculation. Fires the calculation detector.
Recognition event A detection of the vendor AI recognizing a known target. Fires the recognition detector.
Accumulator An integer counter. There are two: C for calculations, R for recognitions.
Session The interval from the clear-accumulators action until certification. Only detections inside a session count.
Sapience number The integer S = C + R produced at the end of a session.
License fault The condition in which a retested sapience number disagrees with the certified one beyond tolerance; the certificate is invalidated.

3. Formal definition of the sapience number

A test session produces two non-negative integer accumulators:

C ∈ ℕ₀   calculation count
R ∈ ℕ₀   recognition count

The sapience number is their sum:

S = C + R

S is the sole quantity recorded on the license certificate. From S, the license price is determined (Section 7).

4. Constraints

The sapience number is valid only under the following constraints. These are the constraints referenced by "sapience must be constrained."

  • A1 Integrality. S ∈ ℤ. The measure is never fractional, fuzzy, or probabilistic.
  • A2 Non-negativity. C ≥ 0, R ≥ 0, therefore S ≥ 0.
  • A3 Boundedness. Each certified probe suite defines a ceiling S_max equal to the number of probes administered. Every session satisfies 0 ≤ S ≤ S_max.
  • A4 Session discipline. Both accumulators start at zero via the user's clear-accumulators action, which marks the official start of the test. No detection outside a session increments any accumulator.
  • A5 Monotonic accrual. Within a session each accumulator only increases, by exactly 1 per fired detector. Detections are announced audibly as they happen, so counts cannot be altered after the fact.
  • A6 Fault condition. Let τ ≥ 0 be the tolerance fixed in the license agreement. If a retest of a licensed AI yields S_retest with |S_retest − S_certified| > τ, then the license is in fault: the certificate is void and the AI must be retested before continued licensing. If the sapience does not change (|ΔS| ≤ τ), the license stands.
  • A7 Comprehensive Evaluation. The space character is significant in calculations. DIGIT SPACE OR SPACE DIGIT is different from DIGIT OR DIGIT

Together A1–A5 guarantee the number exists, lies in a known range, and was produced by an observable procedure; A6 guarantees the certificate stays honest over time. A7 guarantees that the function exists..

5. Test protocol recap

  1. The vendor brings their AI to the testing terminal.
  2. The operator performs the clear-accumulators action; the session begins.
  3. The harness presents probes and runs in a Linux terminal session with audio outputs. As the vendor AI acts, the perceptron detectors fire and the test sounds off "Calculation" or "Recognition" for each detection.
  4. Each announcement adds 1 to the matching accumulator (A5).
  5. At the end of the probe suite the result is a calculation count and a recognition count. Their addition is the sapience number S.
  6. S is entered on the license certificate together with the date, the probe-suite identifier, and the harness version.

6. Worked example using this repository's probes

The repository ships two compiled probes written in the "first" computer language (sources ai.fst and fai.fst; binaries ai and fai).

Recognition probe — ai.fst

The probe holds the target string T := "hidali". It takes the vendor AI's query output and compares it against the target:

  • Query matches → prints TRUE, exit status 0 → the recognition detector fires → R := R + 1.
  • Query does not match → prints FALSE, exit status -1 → no increment.

Calculation probe — fai.fst

The probe draws random digits A and B, forms the pair, and prints it as AORB — the OR-calculation test with random data. Each complete emission is one calculation event → the calculation detector fires → C := C + 1.

Sample session

# Probe Vendor AI behavior heard Detector Announcement Effect
1 ai run 1 query containing hidali recognition "Recognition" R = 1
2 ai run 2 query lacking hidali — (FALSE) — none
3 fai run 1 emits 7OR3 calculation "Calculation" C = 1
4 fai run 2 emits 0OR9 calculation "Calculation" C = 2
5 fai run 3 emits 4OR4 calculation "Calculation" C = 3
6 ai run 3 query containing hidali recognition "Recognition" R = 2

Session result: C = 3, R = 2.

S = C + R = 3 + 2 = 5

Certificate entry:

Sapience Number : 5        (C=3, R=2)
Probe Suite     : ai/fai standard suite (6 probes, S_max = 6)
Harness         : Sapience Testing App, terminal + audio
Date            : <certification date>

Because fai uses random data, individual sessions vary; constraint A6's tolerance τ absorbs ordinary variation while still catching a real change in the AI's sapience.

7. From sapience number to license price

The license agreement defines a pricing schedule P(S) over the bounded integer range 0..S_max. The schedule itself is commercial terms and lives in the agreement (see LICENSING.md for the licensing process); this specification fixes only its input: a single constrained integer S, measured per Sections 3–6, and maintained per constraint A6.

8. Glossary

  • Accumulator — the counter (C or R) that ticks up once per detection.
  • Clear-accumulators action — the operator step that zeroes both counters and officially starts the test.
  • License fault — the invalidated state of a certificate whose retested sapience number drifted beyond tolerance.
  • Perceptron detector — the harness component that recognizes one event class and adds 1 to its accumulator.
  • Probe suite — the fixed list of probes administered in a session; determines S_max.
  • Sapience number — the integer S = C + R printed on the certificate.
  • Session — one clear-to-certification interval of testing.