Sign in
Ask AI

AIM Framework glossary

Confabulation Rationality Lemma

Canonical definition

The Confabulation Rationality Lemma is the formal result that a rational Bayesian reasoner converges on a sincerely held false attribution whenever the true generative Hypothesis is unable to accumulate probability mass, and that the convergence is experienced from the inside as truth-discovery rather than as the failure mode it is. The result is general: it governs any epistemic system updating on any subject matter, and the true Hypothesis it excludes may be of any kind. Inability to accumulate mass has two architecturally distinct sufficient conditions, and the Confabulation Rationality Lemma holds under either. Under Type I Exclusion the true Hypothesis is absent from the representable Hypothesis Space, its prior undefined rather than small, and posteriors converge on the least-wrong available Hypothesis. Under Type II Exclusion the true Hypothesis is representable, and probability mass has already concentrated on a competitor by two mechanisms operating jointly. The first is order of encounter: because the evidential force an observation carries is weighed against the Belief-state it meets rather than being intrinsic to the observation, a Hypothesis that occupied the explanatory slot early is met as background by everything arriving after it. The second is reaffirmation: the incumbent is restated on every cycle, from the Belief Register's own prior state recalled as Evidence under the Self-Coupling Permanence Principle and from the correlated attributions of other Individuals arriving through interpersonal coupling, so that the mass the incumbent holds tracks the count of its restatements rather than the weight of what supports them. The two mechanisms ordinarily compound: a Hypothesis that arrived earlier and is restated often will dominate one that arrived later and is not. Under both mechanisms each increment of concentration lowers the Interference Term, and a lower Interference Term admits less movement from the next observation, so the concentration extinguishes the capacity that would have reversed it. Neither condition is prior to the other and neither is the more common. The updating is rational under both, in the same technical sense: the reasoner assigns posterior probability in proportion to prior weighted by likelihood, and does so correctly on the information available. What differs is which part of the input is defective. Under Type I Exclusion the Hypothesis Space is incomplete; under Type II Exclusion the Evidence stream is contaminated by restatement of the reasoner's own output and of the outputs of those coupled to the reasoner. The Confabulation Rationality Lemma is content-agnostic under both, operating identically regardless of which true Hypothesis is excluded and regardless of subject matter. Sincerity is a deductive consequence under each condition, by a different route: under Type I Exclusion because the true Hypothesis is unavailable to the updating process, and under Type II Exclusion because Coupling-Induced Opacity denies the reasoner any marker distinguishing endogenous restatement from exogenous Evidence.

Explanation

A mind arrives at a confident wrong answer by two routes, and on both of them the mind is working properly. On the first route the right answer is not on the list of answers the mind can give, so it gives the best of the ones that are, and nothing in the result tells it that an answer was missing. On the second route the right answer is on the list and is passed over, because the mind keeps meeting a different answer instead: it meets its own earlier answer again whenever it returns to the question, and it meets the same answer from everyone around it. Repetition does not arrive labelled as repetition. It arrives looking like corroboration, and an answer restated five hundred times feels better supported than one restated twice, whether or not anything was learned on any of those occasions. The second route therefore does not require that the truth be unavailable. It requires only that something else was already in place and was said more often. In both cases the confidence is the correct response to what the mind had to work with, which is why sincere confidence in a wrong explanation tells you nothing about whether the right one was available. Extension: The Confabulation Rationality Lemma's domain is general and its instantiations are not ranked. Any true Hypothesis absent from a Hypothesis Space, or present in one and dominated within that Hypothesis Space, produces the same output, and the framework's own case is one instance among them rather than the result's subject. That case is the M-Hypothesis: Source Opacity (Axiom 3) destroys the source-tag before the valuation reaches Conscious Cognition, and the three-source taxonomy is absent from default Hypothesis Spaces, so the Mimetic source is excluded under Type I Exclusion and, once AIM Vocabulary Acquisition has supplied it, remains exposed to Type II Exclusion. What distinguishes the case is that the exclusion is architectural rather than incidental. It is not that the Confabulation Rationality Lemma is about it. Further instantiations of the Confabulation Rationality Lemma under Type I Exclusion include: (a) the Information Gap Principle, which applies the lemma to causal explanation under conditions where distributed or systemic causes exceed the reasoner's computational tractability, producing sincerely held agent-based scapegoat narratives; (b) Constitutive Narration, which applies the lemma to identity content where no mind-independent Referent exists, producing sincerely held self-narratives that construct rather than describe the self; (c) Confabulation Compounding, which is the lemma iterated across time, with each posterior becoming the prior for the next update, producing progressive convergence toward the Individual Fixed Point; (d) the Non-Malice Corollary, which applies the lemma to attributions of intent in systemic harms, predicting that observers will sincerely believe in malicious agency where structural causation is unavailable to them; (e) the Intractability Confabulation, which applies the lemma to disciplinary closure, producing the sincerely held narrative that a phenomenon is too complex to be modelled when the framework required to model it is unavailable to the discipline. The worked instantiation under Type II Exclusion is the Confabulation Race, in which an incumbent attribution already loaded into the Belief Register occupies the explanatory slot before deliberation begins and is unseated only by repeated effortful I-Override. In every instantiation the same Bayesian mechanism produces sincerely held false Beliefs; what varies is which true Hypothesis is prevented from accumulating mass, and by which of the two conditions. Intervention specification. Confabulation is not an irrationality to be corrected by demanding more careful introspection; under both conditions it is the rational output of a Bayesian system whose input is defective. The two conditions call for two interventions and neither substitutes for the other. Type I Exclusion is answered by supplying the missing Hypothesis, whatever that Hypothesis is; within the framework's own case, by Hypothesis-space expansion through AIM Vocabulary Acquisition, the Knowledge Patch's first operation. Type II Exclusion is answered by the operations acting on the concentration itself: Doubt Generation, Epistemic Suspension, and the Two-Question Diagnostic, which sustain a non-zero Interference Term against closure-demand and supply likelihood ratios high enough to outweigh accumulated reaffirmation. Supplying a Hypothesis without them adds a candidate to the list that never accumulates mass. The Confabulation Rationality Lemma is the formal architectural reason that the Knowledge Patch is necessary, and its two-condition structure is the reason the Knowledge Patch requires more than one operation.

Epistemic status

Framework-Specific Named Result.

Derivation

Chapter 11, Section 11.2; Appendix A, Section A.5.6.

Also in the entry

Source: Chapter 11, Section 11.2; Appendix A, Section A.5.6.

Related terms

Ask the AI to explain "Confabulation Rationality Lemma" in context