Alignment Workspace

pic-goal-in-state

leaf/picture · drafted

Statement

The operator's correction carries no information about value but writes a goal register the agent's utility is indexed by, so the agent is corrigible to modifications through that register while its objective over end states is fixed.

Source

Hudson 2024 (a lack of resistance to having goals modified is what cannot be captured in a ranking of end states) and Hudson 2025's transformation; the clean-room package projects/deference/faf-cleanroom-2026-10-02/packages/lit-mdp-corrigible, findings F-1 (Theorem 3.1's corrigibility half is definitional under one reading) and F-17 (the transformation is inert on a physical channel).

Author and human events

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

No human event has touched this node.

Edges

Above

Ranking (the lens, not the structure)
{
 "status": "open",
 "diagnosis": null,
 "flags": [
  "AI number"
 ],
 "V": {
  "V_prior": null,
  "display": "2",
  "headline": 2,
  "ref": 2,
  "valid": 2,
  "weaker": "both"
 },
 "belief": null,
 "draft": "ai"
}

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