nucleation

CLAIM-535: An organization that acts on what it observes has a policy acceptable in every circumstance exactly when, for each observation, the circumstances it cannot tell apart share an acceptable action; so every circumstance having an acceptable action does not give a policy, a finer observation never removes one, and where acceptable actions overlap the coarsest observations that suffice are several and incomparable, so which distinctions must be made legible depends on which acceptable action the organization commits to

2 min read

Properties11
StatusActive
Rolethesis
Warrantdemonstrated
Tagsagency, information-theory
Linewill-organization, formal-apparatus
Bears onQUESTION-285 — What organization turns an interaction into a measurement, a measurement into a record that something uses, and observation into inquiry, without a conscious investigator; and what keeps an apparatus that revises its own discriminations answerable to what it did not produce?, QUESTION-133 — Can a population of operators whose demands are jointly infeasible under one functional organization develop a differentiated or compositional repertoire that makes them jointly feasible, without suppressing, weakening or deleting any of those demands?
Rests onCLAIM-350 — A coarse-graining preserves a deterministic judgment exactly when the judgment is constant on its fibres: the judgment then factors through the coarse description, uniquely on its image, and otherwise two states the judgment tells apart have been merged
GroundsTHEORY-1096 — Blackwell's comparison of experiments, its easy direction: an experiment is a family of distributions indexed by a common parameter, and if one experiment is a garbling of another by a Markov kernel, every risk attainable with the garbling in any decision problem with bounded losses is attainable with the original, on any outcome spaces
ComplementsCLAIM-523 — For a normatively relevant discrepancy to change what an organization does, three distinct achievements are needed and any can fail while the others hold: legibility, by which the discrepancy becomes recognizable; volitional efficacy, by which a corresponding demand can move action; and correctability, by which a fitting revision can be carried out within deadline and resources; none establishes that the demand or the revision is justified, and whether the norm-satisfying histories the agent can still realize are empty depends partly on what its earlier actions destroyed, CLAIM-354 — Freedom of will and autonomy come in degrees along several dimensions, among them viable reachability, control cost, robustness to disturbance, governance reconfigurability, revisability and motivational integration, and agents compare by inclusion of what they can viably reach under matched conditions: a partial order with incomparable pairs, not a scalar, and a gradient only of a declared measure, CLAIM-152 — Any individuation built on sufficiency is relative to the target it is sufficient for: fix the outcomes differently and a different partition wins, so no boundary drawn this way is universal, CLAIM-537 — In tolerance-bounded compatibility on a graph with disturbed offsets, |Dx − a₀ − w| ≤ δ for |w_e| ≤ η_e, a compatible configuration exists for every disturbance exactly when |cᵀa₀| + Σ η_e|c_e| ≤ Σ δ_e|c_e| on every cycle, and one configuration serves every disturbance exactly when η ≤ δ and |cᵀa₀| ≤ Σ(δ_e − η_e)|c_e|; an agent that chooses its configuration from what it observes needs the second condition within each class of disturbances it cannot tell apart, so whether a compatible state can be selected lies between the two and depends on what the agent can read
UsesTERM-064 — legibility, of a distinction to an observer for a task: whether the observer, with the rules, categories and resources it actually has, can recover the distinction from what it is given and bring it to bear on the task; a relation among a representation, an observer's rule class, its resources and a task, bounded above by what the source makes possible for an ideal receiver, and directed, since what is legible to a governing institution and what of its governing is legible to others are different relations
Cited as “rests on” byCLAIM-532 — Self-formation, in the sense the record holds (the transformation dimension of a will-organization), is the production by an agent's own activity on a discrepancy it cannot yet resolve of a maintained capacity, a new distinction it can make legible or a new channel by which it can act, that outlasts the discrepancy and serves a class of later ones; it is told apart from learning an action by transfer to discrepancies of the same class after the inducing conditions are gone, and from a capacity installed or upheld from outside by interventions on the processes that produce and maintain it

Graph View

The claim

Circumstances w ∈ W, an observation map h: W → O, and for each w a set A(w) of acceptable actions. A policy π: O → actions is acceptable everywhere when π(h(w)) ∈ A(w) for every w.

  • Criterion. Such a π exists exactly when ∩_{w ∈ h⁻¹(o)} A(w) ≠ ∅ for every o. If each such intersection is nonempty, pick π(o) in it. If one is empty, any π(o) fails some w in that class.
  • Possibility is not enough. A(w) ≠ ∅ for every w does not give a policy, as in CLAIM-537’s triangle.
  • Refinement. If h = g ∘ h′, any policy π for h gives π ∘ g for h′, so a finer observation never removes an acceptable policy (Blackwell’s easy direction, THEORY-1096).
  • Policy-relativity. With W = {1, 2, 3}, A(1) = {a}, A(2) = {a, b}, A(3) = {b}, the partitions {1, 2} | {3} and {1} | {2, 3} both suffice, neither refines the other, and the trivial partition does not. So there is no unique coarsest sufficient observation. Which distinctions the organization must make legible depends on which acceptable action it commits to where it has a choice. CLAIM-350’s unique coarsest partition is the case with one acceptable action per circumstance.

So legibility, in TERM-064’s sense, has a positive job here: to make legible the distinctions that separate circumstances with no common acceptable action, under the policy the organization adopts.

What it does not say

  • Anything about memory, the cost of observing, or limits on action. One step only. A32’s “bounded actuation” half is unproved and not filed.
  • That the acceptable sets are the right ones.
  • That choosing what to observe under cost is covered. That is active sensing and dual control (Feldbaum), not registered because its citation could not be checked.

Backlinks