Module Map
One loop: generate models → derive consequences → risk them against reality → update beliefs → detect paradigm/program capture → rebuild when pressure demands.
Inference Loop
Generate candidate models (compression, breadth, predictive risk).
Derive testable consequences inside a rule-set.
Prefer claims that forbid outcomes; attempt to break them.
Bayesian coherence; MaxEnt to encode ignorance honestly.
Paradigms & programs determine what counts as “data” and “method.”
Incompleteness + underdetermination require rebuild paths.
Thinker Lenses
Primary lenses that structure the lecture; auxiliary anchors that enforce limits and improve model selection.
- Aristotle: validity as form; syllogistic discipline.
- Popper: falsifiability; severe tests; conjectures.
- Kuhn: paradigms; crisis; incommensurability.
- Lakatos: research programmes; progressive vs degenerating.
- Quine: holism; web of belief; revisability.
- Jaynes: probability as extended logic; MaxEnt.
Hume (induction wound) · Peirce (abduction + inquiry) · Gödel (formal limits) · Kolmogorov (probability axioms) · Solomonoff (universal induction).
0 · Substrate: Logic, Truth, Signal
Logic is chosen; truth is stable signal under pressure; epistemology is the protocol for revising what counts as true.
Core upgrades
From “truth as label” → “truth as persistent alignment under stress.”
toggle
- Logic is chosen, not given. Classical logic (bivalence, monotonicity, explosion) is a tool; alternative logics can dominate in incomplete, adversarial, or self-referential environments.
- Truth = stable signal under pressure. Not only correspondence, but persistence against distortion, incentives, and time.
- Mapping: logic = syntax of transformation · truth = stable signal · valid inference = no injected noise (relative to chosen logic) · epistemology = update protocol.
1 · Validity & Soundness
Validity is structural (relative to a logic). Soundness is validity plus true premises (signal-bearing inputs).
Classical core + logic-relativity
Classical core for hygiene; overlays for revision/contradiction/necessity.
toggle
- Validity: impossible (in the chosen logic) for premises true and conclusion false.
- Soundness: valid + premises true.
- Logic-relativity: the same form can behave differently across classical / paraconsistent / non-monotonic / modal logics.
- Signal view: soundness = signal-preserving transformation applied to signal-bearing inputs.
2 · Fallacies as Exploits and Heuristics
Fallacies are structural exploits (formal) and evidence/meaning exploits (informal). Some “fallacies” are conditional heuristics about testimony and incentives.
Two classes + adversarial reinterpretation
Pattern-recognition layer: name the move; then audit its function.
toggle
- Formal fallacies: invalid structure (e.g., affirming the consequent).
- Informal fallacies: relevance/evidence/meaning failures (ad hominem, straw man, equivocation, base rate neglect).
- Adversarial clause: “fallacy” can be weaponized as a rhetorical shutdown; also, some are rational shortcuts when the target is testimony and incentives.
3 · Deduction, Induction, Abduction
Deduction preserves truth; induction amplifies with uncertainty; abduction generates hypotheses. Hume exposes induction’s wound; Peirce repairs inquiry with abduction.
Three-mode engine
Abduct → Deduce → Risk → Update (repeat).
toggle
Modes
- Deduction: rules + premises ⇒ conclusion (must follow if premises true).
- Induction: observations ⇒ expectations (probabilistic; never certain).
- Abduction (Peirce): effects ⇒ best explanatory causes (hypothesis generation).
Compression (short code) · breadth (many phenomena) · depth (fits secure anchors) · predictive risk (testable consequences) · priors/hypothesis-space audit (attack surface).
Hume & Peirce
- Hume: no deductive guarantee the future resembles the past (problem of induction).
- Peirce: inquiry survives by self-critique; abduction supplies hypotheses the other modes can test and update.
4 · Popper: Falsifiability and Critical Rationalism
Science as conjectures exposed to refutation. The demarcation line: claims must forbid outcomes.
Risk ethic
Prefer hypotheses that can be hurt; test by attempted refutation.
toggle
- Conjectures: theories are creative guesses, not induced from data.
- Falsifiability: scientific claims exclude possible observations; “explains anything” fails demarcation.
- Sophisticated falsification: anomalies strike bundles (theory + auxiliaries + instruments).
5 · Quine: Web of Belief and Revisability
Tests hit whole systems; many revision strategies preserve global coherence. Underdetermination means multiple webs can fit the same data.
Holism
No single experiment forces a single belief to fall; revision is reweaving.
toggle
- Analytic/synthetic line dissolves: even logic/maths are (in principle) revisable under enough pressure.
- Underdetermination: different ontologies can match the same observation set.
- Operational consequence: falsification becomes comparative and program-level, not single-claim deletion.
6 · Kuhn: Paradigms and Revolutions
Science runs inside paradigms (methods, standards, exemplars). Anomalies accumulate; crises open the gate for paradigm shift; standards change with the shift.
Paradigm diagnostics
Track what questions are thinkable, which anomalies are buried, and how legitimacy is allocated.
toggle
- Normal science: puzzle-solving; paradigm presupposed.
- Crisis: persistent anomalies destabilize standards.
- Shift: rival framework wins; concepts and criteria mutate; incommensurability can appear.
7 · Lakatos: Research Programmes
Hard core + protective belt; evaluate trajectories (progressive vs degenerating) by novel, risky predictions vs ad hoc patching.
Programme-level evaluation
The unit of analysis becomes a moving research complex, not a single claim.
toggle
- Hard core: protected assumptions.
- Protective belt: auxiliaries that absorb anomalies.
- Progressive: predicts novel facts that succeed.
- Degenerating: post hoc fixes; epicycles; appearance-saving.
8 · Jaynes: Probability as Extended Logic
Coherence constraints force degrees of belief to obey probability theory; Bayes updates; MaxEnt encodes ignorance without smuggling assumptions.
Coherence + MaxEnt
Deduction is the {0,1} edge-case of plausible reasoning.
toggle
- Probability = logic of uncertainty (under rationality desiderata).
- Bayes: posterior ∝ prior × likelihood.
- MaxEnt: choose the least-assuming distribution consistent with constraints.
- Attack surface: priors and hypothesis space; evidence channels can be adversarial.
9 · Falsifiability Under Holism
Popper’s demarcation survives as a minimum norm, but real theory-choice is comparative and embedded: webs, paradigms, programmes, and Bayesian updates.
Integration
Popper posture + Quine holism + Kuhn history + Lakatos criteria + Jaynes updating.
toggle
Prefer hypotheses/programmes that (1) forbid outcomes (falsifiable), (2) make risky novel predictions (Lakatos-progressive), (3) compress broad data with minimal code (AIT intuition), (4) update coherently under new evidence (Jaynes), while (5) tracking paradigm constraints that decide what counts as evidence (Kuhn) and (6) recognizing tests strike bundles (Quine).
10 · Scientific Method as Adversarial Protocol
Beyond the classroom skeleton: formal hygiene → probabilistic coherence → falsificatory risk → programme/paradigm diagnostics → adversarial constraints → meta-audit.
Six levels
Treat “data” as a move in a game; build costs-to-fake and pre-commitment.
toggle
1) Formal hygiene
Choose the logical substrate; enforce validity; flag formal fallacies.
2) Probabilistic consistency
Represent uncertainty numerically; update by Bayes; use MaxEnt for ignorance.
3) Falsificatory risk
Prefer claims that can break; design severe tests and pre-commit failure criteria.
4) Programme / paradigm diagnostics
Classify programmes as progressive/degenerating; track anomalies & burying behavior.
5) Adversarial constraints
Replication across independent incentives; pre-registration; open data; commitment mechanisms.
6) Meta-epistemic audit
Periodic interrogation of priors, missing hypotheses, taboo zones, and incentive structure.
11 · Incompleteness, Compression, Limits
Gödel sets formal limits; Kolmogorov/Solomonoff link simplicity to description length and prediction; compression punishes overfitting in abduction.
Two constraints
No final system; no internal total safety proof. Build reset paths.
toggle
- Gödel: sufficiently expressive consistent systems contain true statements unprovable within; cannot prove own consistency from inside.
- Algorithmic Information: explanation ↔ compression (minimum description length); flexible models that fit anything pay with long code.
12 · Integrated Sovereign Logic–Epistemology Stack
Full-stack summary: chosen logic + validity/soundness + fallacy firewall + three-mode inference + Popper risk + Quine web + Kuhn paradigms + Lakatos programmes + Jaynes coherence + incompleteness + reset.
Operating law (compressed)
A recursive protocol that stays coherent under contradiction, uncertainty, institutional power, and adversarial data.
toggle
- Choose logic (classical core; non-monotonic/paraconsistent/modal overlays as needed).
- Enforce validity (structural hygiene) + soundness (premise truth as stable signal).
- Detect fallacies (formal exploits; informal evidence/meaning exploits; track “fallacy” as rhetoric).
- Abduct candidate models under constraints (compression, breadth, depth, predictive risk).
- Deduce risky consequences; pre-commit to break conditions.
- Attempt refutation (Popper) while recognizing bundle impacts (Quine).
- Update coherently (Jaynes: Bayes + MaxEnt); continuously audit priors/hypothesis space.
- Diagnose paradigm & programme dynamics (Kuhn/Lakatos): anomalies, burying, epicycles, power.
- Respect limits (Gödel/underdetermination) and maintain reset protocols for rebuild when pressure clusters.
A claim earns epistemic weight only by surviving explicit structural hygiene, adversarial testing risk, coherent updating, and programme-level scrutiny—under the constraint that no formal system can certify its own completeness from within.
Resource Index (All Links)
Grouped for fast retrieval: Baselines → Induction/Abduction → Popper/Demarcation → Quine/Kuhn/Lakatos → Jaynes/Bayes → Limits/AIT → Media.