CLN WIKI

Signal-Source Decoupling

When signals stop pointing to their sources — and cognition begins to fail.

Cognitive Science · Vadim Shchetinkin

The contract that makes thinking possible.

Human heuristics are not standalone algorithms. They are calibrated to an environment. A yawn triggers empathy because yawns reliably indicate fatigue. A confident tone triggers trust because confident speakers are usually well-informed. A request for help triggers cooperation because the requester is usually a person who can reciprocate. These correlations — signals reliably pointing to their sources — are the invisible scaffolding of social and epistemic life.

Signal-source decoupling is what happens when this scaffolding begins to collapse: when a yawn might be animated, when confidence might be a language model's next token, when a request for help might be a phishing bot. The signal remains; the source has become unknowable.

CLN v2.0 = CONCEPT: Signal-Source Decoupling =the severing of reliable correlation between a signal and the type of entity that produces it #decoupling #heuristics #ai-risk =proposed mechanism: R. Scott Bakker (2016) — "AI and the Coming Cognitive Ecological Collapse" // 01 · The Mechanism in Detail .Baseline State (Pre-AI) =signal: speech pattern X =source: human with cognitive state Y +correlation is high → heuristic "trust confident speaker" works .Decoupled State (AI-era) =signal: speech pattern X (identical) =source: language model optimising for P(next token) — no cognitive state Y !heuristic fires → produces wrong conclusion !the cognitive system cannot detect the failure — it has no mechanism for doing so // 02 · Domains of Decoupling .Epistemic Domain =confident assertion → usually means: speaker believes claim, has evidence =AI: confident assertion → means: high probability token sequence ~the "hallucination" problem is a symptom of decoupling .Social Domain =empathy signal → usually means: another person is experiencing an emotion =AI: empathy signal → means: token matching emotional context patterns ~therapeutic chatbots, companion AIs — exploit anthropomorphic inference errors .Authenticity Domain =original writing → usually means: a person thought through this argument =AI: original-seeming writing → means: a model was prompted ~authorship, accountability, intellectual effort — all decoupled // 03 · Why Heuristics Cannot Self-Correct .The Speed Problem =heuristics operate pre-consciously, at ~20-200ms =deliberative verification operates at seconds-to-minutes !social and emotional responses fire before verification can intervene .The Scale Problem =a person generates ~10K words/day in social interaction =AI generates 10K words in seconds, at global scale !the ecology of signals becomes AI-saturated faster than adaptation can occur .The Invisibility Problem =ecological collapse is not experienced as collapse — it's experienced as normality ~fish doesn't notice water disappearing until it's too late // 04 · Stabilisation Strategies .Technical Approaches =watermarking: embedding source metadata in AI outputs =authentication: cryptographic proof of human authorship ~arms race: easily defeated by sufficiently advanced models .Structural Approaches =make the signal itself carry the source information =structured formats that expose reasoning process — not just conclusions +CLN: operator semantics encode human intent into document structure +a !! operator is a human judgment, not a statistical pattern +the structure itself demonstrates cognitive engagement — cannot be faked at scale →explicit structure is the new signal-source coupling mechanism @related: Cognitive Ecology, Heuristic Cognition, AI Amnesia

Structure as authenticity.

The deepest insight from signal-source decoupling analysis is that the solution cannot be purely technological — it must be structural. Adding metadata to AI outputs is an arms race. What is needed instead is a way to make the signal itself carry evidence of human cognitive engagement.

A CLN document with carefully placed ! risks, ? open questions, and ~ tensions is not just well-formatted. It is a record of a mind at work — a signal that, by its structural complexity, points back to a human source. This is not foolproof, but it raises the cost of deception and rewards genuine cognitive effort.

The question is not whether AI can produce the signal. The question is whether the signal retains any information about the source.
← Cognitive Ecology Heuristic Cognition →