The claim
The owner, at U10, on Friston et al.’s description of action as “sampling selectively what one expects to experience” (LIT-1181, NOTE-960): “I think this selectivity maps to the authority space in my coupled model.”
In CLAIM-297’s circuit, frame → retrieval → interpretation → endorsement → future retrieval policy, influence w decides which operators act, the operators that act decide what is retrieved, and what is retrieved is what the next judgement observes. The current allocation of influence selects the evidence. In CASE-099’s two-hypothesis linearization:
- Selective sampling is the β arrow. The specialist with authority retrieves evidence for its own hypothesis, and so moves the meaning toward it. In Friston et al., action samples the data the agent expects, and unexpected perturbations are “explained away by action and not perceived”.
- Policy selection on current beliefs is the γ arrow. Active inference scores policies by expected free energy, computed under the agent’s current beliefs, so the prevailing interpretation decides which policy, and so which sampling, gets authority.
The model’s γ is the coupling from interpretation to endorsement. It is not active inference’s γ, the precision over policies, which in this mapping sits on the authority side.
With both arrows present, CASE-099 says the symmetric state is captured once βγ > (1 − α)(1 − δ). Friston et al. see the self-confirmation and accept it: “this weakness becomes the strength”. Their anchor is outside the loop, in selection on whole agents: “agents who do not have the right priors cannot exist”. In the model’s terms, the free-energy agent is a coupled loop run without independent scoring (no Regime B, CLAIM-330), whose errors are corrected only by the removal of organizations that get them wrong.
What it adds to the thesis
- The selecting work Friston gives action is done inside the model. It is done by allocating authority among operators, with no action on the world. This is CLAIM-341’s thesis at the place where active inference is most explicit about why action matters.
- The anchor question moves. If selective sampling is authority, then what anchors a mind is whatever scores its operators independently of the interpretation they serve (CLAIM-342, QUESTION-121). Friston’s answer is selection across agents, which is not available within one.
- Active inference is the commitment regime of CLAIM-340. Predictions are made before the sensations they are scored against. CLAIM-340 says this protects against capture only when the interpretation bends the reading of the evidence no more than it bends the predictions. Friston’s selective sampling bends what arrives.
What it does not say
- It does not show that action and retrieval are the same. Action changes hidden states of the world, which can then deliver what the agent did not expect. Retrieval draws from the agent’s own store. The mapping needs the store to play the world’s part: to hold distinctions that a query can reach and that the current interpretation did not put there (CLAIM-298). Whether it can is the claim’s open point, and its defeater.
- It is CLAIM-337’s move run in the other direction. That counter says the judgement loop is action and perception one blanket inward. This claim says active inference’s sampling is authority one blanket outward. Both say the two are formally alike. They disagree about which description is primary.
- Friston et al. do not use the word authority, and the mapping is not theirs.
Note of 2026-10-21: which facts become legible as evidence
A22 of the exchange of 2026-10-20: “Authority can affect not merely how evidence is interpreted, but which facts become legible enough to count as evidence at all”. That is this claim’s authority arrow (influence decides what the next judgement observes), stated for institutions, with the categories and reporting requirements an institution keeps in place of retrieval. Whether that is a new channel or CLAIM-340’s κ′ is QUESTION-253. The term is filed as TERM-064.