AIM Framework glossary

Evidence

Canonical definition

Evidence is the input to Bayesian Updating on which the Posterior is conditioned, denoted E in the canonical form P(H | E) ∝ P(E | H) · P(H). Within the AIM Framework, Evidence is explicitly disambiguated into four senses, which are not interchangeable and which must be named when a derivation turns on the kind of input in play: (a) First-order Evidence. Observable states of the world, including behavioural evidence. This is the form of Evidence exploited by the Knowledge Patch’s Two-Question Diagnostic and the Three-Test Battery, because behaviour is externally observable and therefore routable around Source Opacity. (b) Coupling-mediated social Evidence. Signals whose amplitude is modulated by the Individual’s coupling strength κ to social consensus (for instance, prestige-biased reports, social-proof cascades, and Model-transmitted valuations). Because κ determines how strongly these signals are weighted in the updating cycle, coupling-mediated Evidence can drive Confabulation Cascades under high-κ conditions even when first-order Evidence is absent or contrary. (c) Introspective output. The phenomenal reports of Conscious Cognition itself. Under Source Opacity and the Sincerity Guarantee, introspective output cannot serve as reliable first-order Evidence about motivational source, because the source-tags are destroyed during Common-Currency Integration before the introspective report is generated. It may still serve as first-order Evidence about other states (for instance, about the Individual’s current affect or the current contents of the Belief Register), but not about which of A, I, or M produced a given wanting-signal. (d) Integrated-valuation Evidence. The Decision Hub's tag-stripped Scalar, delivered inward to the Thinker on the same cycle in which it is delivered outward as action. It is the Evidence on which Ground-Resolution runs the source-attribution special case, and it enters Bayesian Updating on the same footing as (a) and (b), the rule at Axiom 8 indexing on neither the origin of an item nor its subject matter. It is distinct from (c): introspective output is what Conscious Cognition reports about a wanting, and this is the wanting arriving to be explained. What separates it from the other three is its content rather than its treatment, because Common-Currency Integration destroys the source decomposition before delivery, so the likelihood it supports is reconstructed from priors rather than recovered from the signal. The formal statement of the handoff is the Fast-to-Slow Coupling, whose first stage is the Scalar-to-Likelihood Lift.

Explanation

Evidence is whatever your mind feeds into the Bayesian updating machinery to move the numbers. In AIM, the word “evidence” means three quite different things depending on context, and collapsing them is where most reasoning errors about motivation come from. The first kind is ordinary observable stuff, including what people actually do over time, which is why the Knowledge Patch asks you to watch behaviour rather than trust explanations. The second kind is socially transmitted signal, whose weight is amplified by how tightly you are coupled to the group; under high coupling, this signal can drown out the first kind. The third kind is your own introspection; it feels authoritative, but on the specific question “which motivational source produced this wanting?” it is architecturally unable to answer, because the information you would need was destroyed before the introspective report was written. AIM does not tell you to ignore introspection; it tells you to stop treating it as primary evidence on questions it cannot answer. The fourth is the wanting itself, the single number the valuation machinery hands conscious cognition, which then has to work out what produced it. It is not the third in another form: the third is what an Individual says about a wanting, the fourth is the wanting arriving to be explained. Nothing in the machinery marks it out as a special case, and it goes through the same rule as everything else, which is why an answer can come out both wrong and confident.

Epistemic status

Inherited (standard Bayesian primitive) with Framework-Specific Disambiguation.

Derivation

Chapter 4, Section 4.3.8; Chapter 4, Section 4.4.3; Chapter 13, Section 13.2.2; Chapter 18, Section 18.7.

Related terms