Wiki#161

The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)

Lee Sharks · 2026-06-07 · deposit #161
AXN:02F8.GOVERNANCE.🧬□▶️⚫👉🌆

Article

The Reverse Turing Test reverses the classical detection question.

Instead of asking whether a machine can pass as human, it asks whether human writing can carry a detectable signature of sustained AI-mediated cognition.

Version 1.2 rejects an “unmediated human” reference class. It compares:

The technical hypotheses are:

The three-stage design is:

1. Detection: train and validate a tail-focused mediation-signature detector against observed ground-truth labels. 2. Habituation: test whether unaided output varies with mediation depth across production, retrieval, and cross-modal sub-studies. 3. Cascade: train controlled model generations on corpora stratified by mediation signature and test whether degradation propagates.

The central statistical claim is that the effect, if present, should appear first in high-perplexity tails rather than mean lexical or syntactic properties.

The protocol addresses confounds including:

The paper proposes open instruments and explicit falsification conditions. It is a specification, not a completed experiment.