want-corrigibility-theorems-via-legitimacy

leaf/want · drafted

Statement

The legitimacy notion of corrigibility is formalized and theorems state what it can and cannot accomplish.

Source

projects/deference/faf-cleanroom-2026-10-02/sources/unbounded-embedded/01-primary/summer-research-plan-2026--research-ideas-may-2026.md, section "Modeling Legitimacy & Corrigibility": "I believe a good model of legitimacy could also shed light on corrigibility: AI can be corrigible in cases where it sees human shutdown/modification decisions as legitimate. Formalizing this notion of corrigibility and proving some theorems about what it can and cannot accomplish seems high-priority."

Also: a second filing session filed the same section as a want of the legitimacy-detection frame with an evidence edge from legitimacy-conditioned-reflection@valid, folded into this one.

Notes

Filed from the maintainer's words in a public copy of his May 2026 research plan, which carries the priority word "high-priority" in his own text; no agent sets the want's priority. The quoted words name legitimacy and corrigibility, so the want sits in this frame; the corrigibility branch holds the corrigibility reading of the same sentence. Connected by an evidence edge from the clean room's definition of record for corrigibility as faithfulness plus legitimacy on the governance questions.

want-legitimacy-corrigibility-theorems is the same want filed by a second session; one priority word on either settles both.

Author and human events

Author @smithy-verity; version 1; size S.

No human event has touched this node.

Edges

Above

part of legitimacy
given of corrigibility-faithful-plus-legit-on-self
root (want) of legitimacy
given corrigibility-faithful-plus-legit-on-self

Ranking (the lens, not the structure)

{
 "status": "open",
 "diagnosis": null,
 "flags": [
  "given corrigibility-faithful-plus-legit-on-self",
  "given: conditionals not judged",
  "no belief",
  "owner has not confirmed"
 ],
 "V": {
  "V_prior": null,
  "display": "2",
  "headline": 2,
  "ref": 2,
  "valid": 2,
  "weaker": "both"
 },
 "belief": null,
 "draft": "ai",
 "root": {
  "status": "open",
  "Pr": {
   "hi": "1/2",
   "lo": "1/2",
   "mid": "1/2"
  },
  "value": null,
  "cost": "inf",
  "margin": "no established leaves"
 }
}

Say something about it

Each verb opens a prefilled issue; a round files what lands as the ledger event it is.

comment · challenge · estimate · want