Sign in
Ask AI

AIM Framework glossary

Bayesian Lock-In

Canonical definition

Bayesian Lock-In is the stable equilibrium reached when Bayesian updating on a restricted Hypothesis Space or on endogenously generated Evidence drives the Posterior on an incumbent Hypothesis to near-certainty, so that competing Hypotheses carry negligible remaining weight and subsequent Evidence is interpreted through the incumbent Belief rather than revising it. In the motivational special case, Bayesian Lock-In is the stable false equilibrium reached when the Confabulation Cascade has sufficiently narrowed an Individual's Hypothesis Space that Mimetic Recognition approaches zero and all new evidence is interpreted through a self-model that attributes desires to authentic sources regardless of actual origin. The mathematical terminus of the individual-level Confabulation Cascade where P(H_authentic \| all evidence) → 1, following necessarily from repeated application of Bayesian Belief Dynamics [Axiom 8] to Source-opaque inputs.

Explanation

Bayesian Lock-In is the terminus of the Confabulation Cascade at the individual level. At this point, certainty in the Confabulated narrative is so high that even contradictory evidence cannot shift it, the Individual does not consider 'maybe this is mimetic' because the conceptual category is unavailable. Instead, any disconfirming evidence is reinterpreted through the existing self-model, generating elaborated Confabulations rather than doubt. This is not irrationality; each update was locally valid. The pathology arises because the evidence itself, the Individual's own narrative, was structurally corrupted at source. Bayesian Lock-In explains why self-correction through reflection alone cannot escape mimetic entrapment: the reflection process uses the same corrupted self-model that generated the problem. Escape requires external intervention, the Knowledge Patch, the Two-Question Diagnostic, or environmental restructuring that changes the evidence stream, to inject Mimetic Recognition and/or change the Individual’s social context before the cascade completes. Source-channel specification. Bayesian Lock-In operates differentially across A-priors, I-priors, and M-priors. A-priors are partially corrigible because A-experience (deficit, satiation) provides exogenous evidence that can update the prior; I-priors are partially corrigible because I-engagement provides feedback (audience-removal persistence, competence development) that can update the prior; M-priors are architecturally non-corrigible without the Knowledge Patch because the M-channel does not provide self-disclosing evidence (preconscious transmission, Source Opacity). The Lock-In dynamic is therefore most severe and most architecturally guaranteed for M-priors, which is why the Knowledge Patch's intervention is M-specifically targeted.

Epistemic status

Framework-Specific Principle.

Derivation

Chapter 13, Section 13.3.3; Chapter 4, Section 4.4.9; Appendix A, Section A.5.8.1; Appendix A, Section A.5.6.

Also in the entry

Distinction from Bayesian Fixed Point. Bayesian Lock-In is the stable false equilibrium at which the Confabulation Cascade terminates in an Individual, named for the narrowed Hypothesis Space that holds the Individual there; the Bayesian Fixed Point is the mathematical property of that equilibrium. Lock-In emphasises the Hypothesis-space narrowing that produces the equilibrium, the source-channel asymmetry, and the architectural reason the Individual cannot self-correct (the M-Hypothesis becomes unavailable to Conscious Cognition); Fixed Point emphasises the formal property of P(H*\|all evidence) → 1 under endogenous evidence dynamics. The two terms name one equilibrium, architecturally and formally respectively.

Related terms

Ask the AI to explain "Bayesian Lock-In" in context