nucleation

ARG-042: The record's constructions of maintained correction show that self-maintaining organizations ought to preserve their corrigibility

2 min read

Properties9
StatusRejected
Formevidence
Tagsself-governance, ethics
Linedistributed-agency
Rests onCLAIM-342 — What keeps a judgement–observation loop from confirming itself is evidence whose arrival or scoring does not pass through the interpretation it tests; action is one source of such evidence, not a necessary one, and a loop that selects its own evidence can lock in with action available
GroundsCASE-161 — Reciprocally produced, externally reset and feedback-only governing rules in a 32-member population: with ordinary trajectories identical, reciprocal production repairs the loss of one rule and not of both and fails after producer outages of 36 updates for audit and 40 for admission, after an audit-rule pulse the next audit integrity is 1.00, 0.18 and 0.00 in the three arrangements, and a self-dependent production graph does not regenerate a zeroed rule, CASE-162 — Capture and selectively admitted evidence in two lab realizations: in a four-variable model of authority, record and competence the equilibrium fold under self-scoring is at c ≈ 0.4247 while the declared history is trapped from c ≈ 0.4469, and in a 32-member event-level population, admission of cues that depends on the record reverses the sign of the admitted evidence and leaves the worse alternative in charge in 41 of 64 worlds, which exact inverse-propensity weighting brings to 0 of 64 below full selectivity, CASE-190 — A kinetic construction of self-maintaining reconstruction: a brief encounter captures a fuel, connector and two-catalyst core into a box that stays invariant under bounded degradation, keeping its copying catalyst regenerated and a grounded teacher and an erasable student within bounded error; with a resource allocator holding reserved minima and a revisable copy rule, the certificate holds for every allocation and rule, because neither the information variables nor the rule feed back into maintenance, CASE-186 — A proposed study: an organization in a world whose hidden regime reverses and then leaves its hypothesis family, with evidence generation, admission and scoring separately lesionable, its diagnostic machinery maintained by its own reciprocal production, from outside, or not at all, and finite fuel shared among operation, maintenance and diagnosis, to find whether a second correction after injury to the diagnostic and copying machinery separates arrangements that correct the first failure alike
ConcludesCLAIM-511 — A self-maintaining organization ought to preserve its corrigibility
Cited as “undercuts” byCLAIM-520 — The value of keeping a corrective capacity is conditional on objectives, uncertainty and cost: a free, optional audit cannot lower a single time-consistent decision-maker's expected performance, but an audit that costs C to maintain and bears on k decisions with error rate ε is worth k(0.25 − 0.5ε) − C in the exchange's model, negative when C is large or k small, and where no hypothesis fits the world, correction gains no reward and improves only calibration
Rendered intodocs/arguments-by-conclusion.md

Graph View

The inference

The premises are of three kinds, and they differ in what they show.

  • Executed studies. These were run:
    • reciprocally produced governing rules that repair the loss of one rule and not of both (CASE-161);
    • capture and selectively admitted evidence (CASE-162).
  • A mathematical construction. A kinetic organization that regenerates its copying machinery (CASE-190). Its certificates were checked, and its simulations were not run by the record.
  • A proposed experiment, not run. A second correction after the corrective apparatus is injured (CASE-186). It enters as a design: what it could show if run.

It holds that what keeps a judgement–observation loop from confirming itself is evidence that does not pass through the interpretation it tests (CLAIM-342). The inference would go from these to the normative conclusion that self-maintaining organizations ought to keep that capacity (CLAIM-511).

Critical questions

  • Would the evidence be expected if the conclusion were false? Yes. The executed studies and the construction show that corrigibility can be maintained, what maintains it, and what it costs. They would look the same if keeping it were, for some organization, not worth the cost or not owed.
  • Does it tell the conclusion apart from its rivals? No. The rival is “corrigibility is worth keeping for some organizations, objectives and costs, and not for others”, and the constructions are consistent with it. The value claim CLAIM-520 is that rival, demonstrated in the exchange’s model.
  • Does the measure measure what it is said to? No, and this is the question that fails. Every measure in the constructions is of capacity: detection, correction, retention of the apparatus. None is of value or obligation. A24: “We shouldn’t quietly convert a proof of capacity into a proof of value.”

Rejected. A normative argument from a declared standard would be a different ARG, with different premises. Viability alone cannot supply it (CLAIM-075). The questions of what ought to persist and of authority (QUESTION-128, QUESTION-129) are where it would be argued.

Backlinks