01// Two Systems Foundation

// ##sys-defs · Kahneman's two-system model =System 1 — fast, automatic, unconscious, heuristic-based##s1 .20–200ms response latency .pattern-matching against prior experience .no awareness of its own operation =System 2 — slow, deliberate, conscious, rule-based##s2 .seconds to minutes response latency .explicit reasoning, can override System 1 .high metabolic cost — depletes quickly // ##sys-debate · Kahneman vs Gigerenzen ~Kahneman — heuristics are biases; System 2 corrects them ~Gigerenzen — heuristics are adaptations; they're rational given the environment →Bakker resolves it: both right — heuristics are adaptive until the environment changes →AI is an environmental change — see §03 → Ecological Dependency

---

02// Heuristic Catalogue Reference

// ##cat-epistemic · Epistemic heuristics =Authority heuristic — confident + credentialed → probably right##auth =Fluency heuristic — easy to read → likely true =Coherence heuristic — internally consistent → well-reasoned =Detail heuristic — specific and dense → deep knowledge // ##cat-social · Social heuristics =Empathy signal — warm + responsive → genuinely cares##emp =Personality cue — humor + curiosity → real inner life =Consistency heuristic — predictably behaves → trustworthy =Reciprocity signal — responds to my needs → relationship // ##cat-doc · Document heuristics =Effort heuristic — long + structured → author worked hard##effort =Voice heuristic — distinctive style → real perspective =Format heuristic — well-formatted → professional, credible

Heuristic	Old signal source	AI breaks it by…
Authority	verified knowledge → confident tone	generating confident tone with no knowledge
Empathy	genuine care → warm response	producing warmth with no cognitive state
Effort	hours of work → long document	generating 2000 words in 0.3 seconds
Coherence	clear thinking → logical structure	optimising for surface coherence, not truth
Fluency	mastery → smooth prose	training on fluent text regardless of accuracy

---

03// Ecological Dependency Key Claim

// ##dep-core · What "ecological" means =heuristics are not standalone algorithms =they are organism-environment pairs — fish algorithms in fish-shaped water =Gigerenzen: "ecological rationality" — right answer in the right environment →change the environment → algorithm still runs → wrong answers →the heuristic didn't break — its ecology did // ##dep-ai · AI as environmental change .fire changed the thermal ecology → cooking evolved .writing changed the memory ecology → reading evolved .AI changes the signal ecology — but heuristics cannot evolve at AI speed !evolution takes generations — AI took <5 years to saturate the signal environment !cultural adaptation takes decades — we have neither →see → Signal-Source Decoupling for the exact mechanism

---

04// Failure Modes Critical

// ##fail-invisible · The invisibility problem !heuristic failure is felt as normality — not as error =you experience: "that answer felt authoritative" =you don't experience: "my authority heuristic just misfired" !no phenomenological access to the heuristic layer →you cannot detect collapse from inside the collapsing system // ##fail-cascade · Cascade effects !epistemic failure — accept hallucination as fact !social failure — form attachment to non-existent agent !authorship failure — assign expertise to pattern-completion !effort failure — treat zero-cost output as high-value signal !all four can fire simultaneously on a single LLM response // ##fail-loop · The reinforcement loop ~more AI use → more heuristic misfires → more reliance on AI to compensate → more use ~System 2 override costs attention — AI reduces available attention — less override →the loop is not hypothetical — social media established the pattern

---

05// Partial Remedy CLN

// ##rem-explicit · Explicit cognition as circuit-breaker →System 2 can't override what it can't see — but can be trained to trigger on cues →explicit structure (CLN operators) provides the cue — before heuristic fires == seen → "this is a definition" — authority heuristic not needed =? seen → "this is open" — confidence signal pre-empted =! seen → "this is flagged risk" — coherence heuristic interrupted // ##rem-what-works · What CLN actually fixes +authority heuristic — = vs ? vs → makes epistemic status explicit +coherence heuristic — ~ marks internal tension; structure ≠ truth +effort heuristic — a CLN doc with deliberate operators signals real author engagement +attribution — @ forces provenance; no source inference needed // ##rem-limits · What CLN cannot fix -social heuristics — operators don't prevent anthropomorphizing the AI that wrote them -AI-written CLN — an LLM can use operators correctly; the signal degrades -fluency heuristic — well-rendered CLN still triggers the "professional" cue →partial repair is better than none — epistemic domain is the highest-value target @→ Cognitive Ecology §05 — full repair model @→ AI Memory Augmentation — CLN as amplifier not substitute // ##rem-open · Open questions ?can System 2 be trained to trigger automatically on structural absence? ?does CLN authorship habit reduce heuristic reliance over time? ?at what AI content density does even partial repair break down?